Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

When the term "earnings premium" comes up, many people tend to think of the Trump Administration's One Big Beautiful Bill (OBBB) and its Do No Harm institutional accountability. A new regulatory regime that tens of thousands of academic programs now face to keep access to federal financial aid.

That view would be wrong, and also parochial.

Omar Khan of the Transforming Access and Student Outcomes in Higher Education organization (TASO) published a piece at Wonkhe this week arguing that UK policy on graduate earnings contradicts itself. The government wants the graduate earnings premium to rise (i.e., postsecondary graduates to increase their earnings over non-graduates). It also wants parity of esteem for young people who don't go to university. Khan calls out the problem: "no amount of rhetorical flourish can make both graduates and non-graduates earn more than the other in their monthly payslips." A premium is a comparison. Somebody has to be on the wrong end of it.

Khan is describing a different regulator than the US Department of Education (ED), a different dataset, and a different country than the OBBB Do No Harm provisions. He is also describing the US, however, which has its own version of the contradiction.

The Trump Administration has been emphatic that not everyone needs a degree and that alternative pathways deserve real standing. Meanwhile the accountability framework it finalized in July runs an earnings premium test with the same metrics for all undergraduate credentials, ignoring the differences. Undergraduate certificates—the shortest, cheapest, most obvious alternative credential on the board were left out by Congress in the OBBB statute, but ED swept these certificates back in under gainful employment authority and folded both into a single earnings premium metric.

Based on the latest ED data, 29.0% of undergraduate certificate programs and 6.6% of associate degree programs would fail the earnings premium rules, while just 1.2% of bachelor degree programs would fail. The actual policy is about to put a massive constraint on non-degree student programs.

So which is it? Are we building short-form pathways or auditing them out of existence? The UK can't decide whether it wants the premium up or the gap closed. The US can't decide whether it is promoting the alternatives or testing them to death.

The earnings premium is not solely an American idea being argued in American terms. Not only is the earnings premium idea bipartisan in the US, it is international and may be growing in scope.

What Khan Actually Argues

The Khan article's setup is that two claims circulate in UK policy debate simultaneously, often from the same people. The first: too many graduates earn too little, meaning that the earnings premium should be higher, and courses that don't deliver should shrink. The second: non-degree pathways are undervalued and deserve parity of esteem.

Khan's move is to take the first claim seriously and follow it. If the test for whether higher education is worth it is that graduates must out-earn non-graduates, then a policy success means non-graduates fall further behind. You cannot campaign for a wider gap and equal standing at the same time.

Then he adds in the demographic effects. Drawing on TASO's own linked-administrative-data work, he notes that students from free school meal (FSM, how England classifies low-income students) backgrounds earn less at every qualification level—but they still gain from every additional qualification they obtain.

Khan's conclusion is not that the earnings premium metric is unusable. It's that the public debate is incoherent, and that "learners are ill-served by a contradictory public debate and policy context." To make the point that there needs to be a balanced set of policies that are not unidimensional, he lays out the theoretical earnings of graduates and non-graduates, asking which is the appropriate level of overlap. Having two policies where one argues for more overlap and another that argues for less overlap is not coherent.

It is important to point out that Khan and others (like myself) that are arguing for more coherent policies are not in fact arguing against any accountability or standards for alternative pathways. Trying to force fit all of education into one set of metrics and not recognizing the tradeoffs involved is the problem—using the earnings premium as currently constructed might make sense for bachelor degree or graduate degree programs, but using it for short-duration undergraduate certificates or associate degree programs is problematic.

What England Actually Has, and What It's About to Get

One big transatlantic difference is that England has never attached a funding consequence to a graduate earnings threshold. To date there has been no institutional accountability attached to earnings premiums.

But that may be about to change, at least in England. The Office for Students (OfS) consultation that closed in December proposes adding "a benchmarked salary measure derived from the Longitudinal Education Outcomes dataset," looking at salaries three years after graduation, as one of three student-outcome indicators. A second consultation on implementation is scheduled to take place in the fall.

A few things worth noting, which I'll develop further over time.

England put a Longitudinal Education Outcomes (LEO)-derived earnings metric into a regulatory instrument in October 2017—the Teaching Excellence Framework (TEF) specification included the share of graduates earning above the median salary for 25-to-29-year-olds. That was five years before the first draft elements of the US's updated gainful employment rule (through negotiated rulemaking) that added an earnings premium test. But England then dropped it while the US kept theirs and expanded it to essentially all of higher education.

A lot of the issues around earnings premium arise from whether data sources are used appropriately, and last year the UK's statistics watchdog told the Department for Education (DfE) that its "graduate premium" framing "could potentially be misleading for users, especially when presented in isolation from prior academic attainment." DfE retired the term and rebuilt the series to control for prior GCSE attainment.

Intermingling

The US and UK moves towards earnings premiums have not happened in separate vacuums even if there is no apparent straight line of influence—the documented citation trail between the two systems is thin, and I've spent enough time in it now to be careful. But people move, conferences overlap, terminology and concepts migrate, and the former head of OfS and an architect of the English student-outcomes regime spent the last two years at Harvard writing about American accountability regulation. There's more here than coincidence and less than a conspiracy, and it's worth exploring in the future.

OEB Session in December

Along with Glenda Morgan and Ed Cramp from Duane Morris, I'll be taking up this thread—two decades of US metric-driven accountability, and what the rest of the world might reasonably learn from it—on a panel at OEB in Berlin, December 2–4. More on that session closer to the date.

In the meantime: read Khan's piece. He does an excellent job pointing out the incoherence of the earnings premium über alles debate (my term, not his), and it helps show how much ROIs and earnings premiums are becoming international subjects.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

What did I read this week? Three things in my inbox caught my attention. They cover very different subjects—graduate migration, college athletics, and quality regulation—but all ask how higher education demonstrates the value it claims to create. The Strada report asks which institutions build a state’s college-educated workforce. The GAO report asks what institutions receive in return for their growing investment in athletics. Nicola Dandridge’s paper asks what educational quality means and who has the authority to define it.

Homegrown talent

A new Strada report provides a fascinating analysis of interstate migration among working-age adults with bachelor’s or graduate degrees. This is an important issue as states increasingly focus on workforce development. The findings were counterintuitive to me: I had pictured college graduates moving to cities in search of work and expected to find more evidence of brain drain.

Instead, the Strada report shows that states build their college-educated workforces primarily by educating and retaining their own residents—not by competing to attract mobile graduates from elsewhere.

Most college-educated workers remain surprisingly close to home, although the proportions vary considerably by state. Two-thirds still live in the state where they attended high school: 55% attended high school and college and now work in the same state, while another 12% left the state for college but later returned.

The type of institution students attend is also associated with whether they remain in their home state. Broad-access institutions are particularly important producers of homegrown talent.

Approximately two-thirds of graduates from broad-access public institutions attended college and now work in their home state. Only around half of graduates from the most selective public institutions followed that path. Broad-access private nonprofit institutions contributed to homegrown talent at rates similar to those of public institutions.

As a reminder, Glenda Morgan has set up a newsletter in alliance with On EdTech. It is called On Student Success and focuses on just that—research and news and commentary on efforts in higher education to improve student learning and career outcomes. I asked Morgan if I could cross-post today’s article, which is excellent in my opinion. If you haven’t done so already, I highly recommend that you sign up for that newsletter today.

With that, enjoy today’s post by Morgan. [- PH]

Higher education has spent decades treating enrollment growth as evidence of institutional success. Presidents are praised for adding students, programs, buildings, and geographic reach. “If we’re not growing, we’re dying,” one university president told researchers from the National Center for Higher Education Management Systems (NCHEMS).

But what happens when the number of prospective students stops growing? In a stagnant or shrinking market, institutions cannot all maintain their enrollment by capturing a larger share. Growth becomes increasingly redistributive: one institution’s gain is more likely to produce losses elsewhere.

Two recent reports—one on American public higher education from NCHEMS and another on English universities from the Higher Education Policy Institute (HEPI)—show what this looks like in practice. Higher education systems are, like it or not, stratified by prestige, selectivity, wealth, and recruiting power. That is a description of institutional position, not a judgment about educational quality or public value.

Better-resourced institutions expand into markets traditionally served by other colleges and universities. Institutions with less power in the hierarchy respond by adding programs, recruiting new student populations, or moving into the territory of still more vulnerable institutions.

Each decision may be rational for the institution that makes it. Collectively, however, these decisions can leave higher education systems more hierarchical, less stable, and less capable of serving students who depend on affordable and geographically accessible options.

The US: Enrollment moves up the hierarchy

NCHEMS released a policy brief with the Lumina Foundation describing how enrollment changes in the United States have shifted the roles played by different kinds of public institutions.

NCHEMS argues that the long period of enrollment growth allowed states to rely on competition and student choice to shape public higher education. When there were enough students to go around, expansion at one institution did not necessarily threaten others.

That assumption no longer holds. Public research universities generally possess greater visibility, stronger finances, and a wider recruiting reach than regional universities and community colleges. As demand softens, these advantages allow them to capture a growing share of students and public resources.

We see this most clearly in undergraduate enrollment. Enrollment at public research universities grew in 36 states—but in only six of those states did the public comprehensive sector also grow. Utah was the only state in which enrollment at comprehensive universities grew more than enrollment at research universities.

This was not, therefore, a shift confined to a few unusually competitive state systems. It was a widespread movement of enrollment toward public research universities.

Some of this change reflects the reclassification of 54 institutions from the comprehensive to the research category between 2013 and 2023. But NCHEMS argues that movement between the categories is itself evidence of mission convergence: institutions are not simply responding to enrollment pressures but also repositioning themselves within the hierarchy.

The problem is not that public research universities are growing or that students should be prevented from choosing them. It is that few states have mechanisms for determining whether that growth advances public priorities or strips enrollment and capacity from institutions with different but equally important missions.

Institutions facing these pressures seek new students and sources of revenue. Regional universities expand graduate education, recruit farther from home, or pursue research status.

Program portfolios are also being remade. Institutions across most sectors are simultaneously adding and eliminating large numbers of programs. Some of this activity reflects legitimate changes in student and workforce demand. But creating programs is also a strategy for attracting new student populations and entering markets historically served by other institutions.

The chart’s most striking feature is not simply the number of new programs but the scale of the churn. Institutions are committing resources to new offerings while eliminating others, even though research cited by NCHEMS suggests that relatively few newly launched programs become sustainable over the long term.

Taken together across a state, these individual program decisions can substantially alter institutional missions without anyone first determining whether the resulting mix of programs effectively serves students, communities, or workforce needs. Student choice may determine which institutions grow, but it cannot by itself ensure geographic access, an appropriate distribution of workforce programs, or adequate educational opportunities for populations whose participation in college is least assured.

The UK: England’s recruitment cascade

England provides a clearer view of how enrollment redistribution works and shows that it can begin even before demographic decline arrives. England’s population of 18-year-olds is expected to grow modestly until around 2030 and then fall substantially during the following decade. Yet institutions lower in the hierarchy are already losing ground.

A HEPI report, which I previously covered at On EdTech, shows that between 2016 and 2025, acceptances at higher-tariff universities—essentially, more selective institutions—grew by approximately 27 percent, compared with growth of about 7 percent across the sector. HEPI estimates that roughly 70 percent of the growth at higher-tariff institutions resulted not from expansion of the overall system but from the redistribution of students who, under earlier enrollment patterns, would probably have attended other universities.

This produces a cascade. Higher-tariff universities recruit further into the attainment range historically served by mid-tariff institutions. Those institutions respond by reaching further into the pool traditionally served by lower-tariff universities. Institutions at the bottom of the hierarchy experience the cumulative losses.

Demography helps determine the overall size of the student pool, but it does not determine which institutions bear the losses. Institutional power, recruitment behavior, and public policy help determine how those losses are distributed.

What does redistribution mean for student success?

Both reports recognize that system-level dynamics can affect students, but neither follows the argument very far into students’ experiences and outcomes. Neither provides direct evidence that redistribution has already worsened access, affordability, support, or completion. The implications discussed below should therefore be understood as risks that policymakers and institutions need to investigate, not as conclusively established effects.

What might happen when students and resources move up the institutional hierarchy?

Access is about more than the number of available seats

Institutions are not interchangeable suppliers of college places. NCHEMS finds that community colleges and regional universities draw far more heavily from nearby populations than research universities do. The median student travels 10 miles to attend a community college and 13 miles to attend a public bachelor’s or master’s institution, compared with 39 miles to attend a public research university.

Many community college students face constraints related to employment, income, caregiving, and transportation. If a rural or small-town community college weakens or closes, growth at a flagship hundreds of miles away does not replace the opportunity that was lost. A technically available college place is not meaningfully accessible if a student cannot afford to relocate, leave a job, or set aside family responsibilities.

Redistribution up the institutional hierarchy may also direct students toward more expensive options. Even when financial aid offsets some tuition differences, relocation, housing, and forgone earnings can make the total cost substantially higher, limiting access or imposing heavier debt burdens on students.

The support squeeze at both ends

Greater access to a more selective institution can benefit students, and admitting a broader range of applicants should not automatically be characterized as predatory. But recruiting students is not the same as supporting them through completion.

If an institution rapidly recruits students whose academic, financial, employment, or family circumstances differ from those of the populations it has traditionally served, its advising, scheduling, financial aid, and academic support systems must change accordingly. Without that investment, expanded access can create a gap between the students an institution admits and those it is designed to support.

At the same time, institutions that lose students also lose tuition revenue. They may reduce course sections, advising, tutoring, and student services precisely when their remaining students have substantial support needs. Redistribution therefore creates student-success risks at both ends: capacity strain at growing institutions and resource erosion at shrinking ones.

Program churn can create student risk

Institutions under pressure frequently respond by creating new programs and entering new markets. But expansion is not free. New programs consume startup funding, marketing capacity, faculty time, and leadership attention before they generate revenue—and some will never enroll enough students to cover their costs.

This can create a destructive cycle: enrollment loss produces a search for growth; the search for growth diverts resources from existing students; weaker services make retention more difficult; and declining retention adds to the original financial pressure.

Program churn also creates direct risks for students. When institutions close programs, students can face discontinued course sequences, changes in degree requirements, credits that no longer apply, delayed graduation, or pressure to transfer. Institutions may provide formal teach-out plans while still leaving students with fewer course options and less predictable routes to completion.

The student-success question is therefore not only whether a new program attracts students. It is also whether the institution can sustain that program long enough for its students to complete it—and protect them if the program closes.

Conventional success metrics can conceal system failure

Redistribution can also make institutional metrics misleading. Suppose students with stronger academic preparation move from regional universities to a flagship university. The flagship’s completion rate may remain high. But the regional university loses both enrollment and revenue while continuing to serve students with greater support needs, and its completion rate may fall.

Those numbers might appear to show that students have chosen the more successful institution. But some of the difference may reflect student sorting and resource redistribution rather than changes in educational quality.

We measure success institution by institution, even when the forces shaping it operate across an entire system.

Two reports, two policy responses

If redistribution threatens access, student support, and program continuity, allowing every institution to pursue growth independently is not an adequate response. NCHEMS and HEPI both argue for greater system-level intervention, but they differ over how far that intervention should go.

NCHEMS calls for states to manage the missions of public institutions more deliberately:

HEPI goes further, recommending institution-level recruitment caps intended to prevent higher-tariff universities from protecting their enrollment at the expense of the rest of the sector.

That proposal makes the underlying trade-off unusually visible. Protecting institutional diversity may require limiting some forms of expansion. But doing so could also restrict opportunities for students who would benefit from attending a more selective university.

A system problem needs a system response

Neither unrestricted competition nor rigid enrollment caps provide a satisfactory answer. The first allows individually rational decisions to produce damaging collective effects. The second risks becoming paternalistic—protecting institutions from competition, restricting student choice, and assuming that policymakers can determine in advance where students should enroll.

Individual institutions nevertheless have responsibilities. Leaders and governing boards should understand how their missions are changing, whether expansion is based on genuine new demand or the redistribution of existing students, and whether their support systems can serve the populations they are recruiting. Institutions losing enrollment must also examine how contraction, program closures, and service reductions affect the students who remain.

But institutional action is unlikely to be enough. Institutions are being asked to compete for students and revenue, and it is unrealistic to expect them to voluntarily sacrifice their own interests to preserve the health of the wider system. The central problem described by NCHEMS and HEPI is created by the interaction of institutional decisions, not simply by the failure of individual institutions to make better ones.

Nor is the problem necessarily a lack of policy tools. States and systems already conduct mission reviews, approve programs, develop strategic plans, collect extensive data, and make funding decisions. The problem is that these activities are often fragmented, weakly enforced, or focused on individual proposals rather than their cumulative effects. A new program may make sense for one institution while contributing to unnecessary duplication across a state. Enrollment growth may strengthen one university while undermining access and capacity elsewhere.

A system response does not require freezing institutional missions or assigning students to particular colleges. It does require making changes in enrollment, programs, and missions more explicit; evaluating their combined effects on students and communities; and judging institutional growth by more than the number of students recruited. Institutions pursuing new populations should be expected to demonstrate that they can support those students, while systems should pay closer attention to what is lost when other institutions weaken.

Meaningful student choice requires more than allowing students to apply anywhere. It also requires preserving viable, affordable options for students who cannot relocate or absorb higher costs. Governments cannot treat every enrollment decision as an isolated transaction and then act surprised when the combined result undermines access, institutional stability, and student success.

The answer is not to preserve every institution in its current form. It is to recognize that colleges and universities perform different public functions and that market competition alone will not necessarily preserve the functions students and communities need.

The enrollment cliff will not simply reduce the number of students in higher education. It will redistribute students, resources, and educational opportunity. Whether that redistribution advances student success—or leaves the most vulnerable students with fewer viable choices—will depend on whether policymakers continue to let the market drift or begin treating the system as a system.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

It is officially the dog days of summer here in Utah. I am taking shelter in my cool basement TV room, watching a lot of cricket.

Starting apprenticeships is easier than completing them

In the US, you can’t throw a rock right now without hitting a post or announcement declaring apprenticeships the next great hope for helping young people enter the workforce. For example, this from a recent article titled “Here’s Why Momentum for Paid Apprenticeships Is Surging.”

Recent Australian data suggest that we temper some of this enthusiasm. The US is currently celebrating its ability to get people into apprenticeships; Australia reminds us that creating an apprenticeship is different from sustaining one long enough for someone to complete it.

Data from the National Centre for Vocational Education Research (NCVER) show that, across all occupations, Australian apprenticeship completion rates have declined over the past decade. The four-year rate has fallen particularly sharply, from 58.1% for the 2012 cohort to 42.9% for the 2021 cohort. Six-year completion has been more stable but also remains below its 2012 level.

But the aggregate trend conceals an important divide. Six-year completion among trade apprentices has improved for every cohort since 2015, while four-year completion among non-trade apprentices and trainees has declined for three consecutive cohorts.

Trade apprenticeships include occupations such as motor mechanic, carpenter, electrician, cook, and hairdresser. Non-trade pathways include clerical and administrative workers, community and personal service workers, sales workers, managers, and machinery operators.

The occupational pattern raises an uncomfortable question about incentives. Did the new funding and incentives available for non-trade occupations during the pandemic create durable training pathways, or did they encourage employers to register workers in apprenticeships that neither party had a particularly strong reason to complete?

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

InstructureCon came to Louisville this year under the theme “Education in the Making” at an interesting moment: two and a half months after a high-profile cybersecurity incident, and one year after a conference full of flash but short on delivered product. The biggest question was whether Instructure would take the data breach head-on in front of its own customers, or manage around it.

In May I argued that Instructure was treating a platform crisis as an engineering problem: no named executive owned the response, the technical briefings were lawyered and script-read, academic continuity went unmentioned, and customers and partners were being communicated at rather than with. A users conference is where that argument gets tested in front of the people who lived through it.

Instructure passed by a wide margin. The incident was addressed directly in the opening keynote, given its own slot the next morning, and surrounded by security and privacy panels—one of them moderated by a customer CIO. Nobody appeared defensive. In a room where anger would have been the easy default, the company was appreciative of a community that had every reason to still be angry. If anything it overcorrected. There is a point at which continued contrition stops reassuring a room that has already decided to stay, and Instructure ran past it. This audience extended its grace back in May. What it needed by Thursday was a reason to be excited about next year.

It did not really get a compelling one, and there was a pattern. Product announcements kept getting deferred: Don’t worry, you’ll hear that tomorrow morning at the product keynote. Oh, you’ll hear more at the Solutions Spotlight. But then the Solutions Spotlight tried to cram Canvas, Mastery, and Parchment into 40 minutes with thin demos. I gave the company that feedback directly, and the response was a follow-up meeting yesterday to make the case that there is more product substance than I saw.

But at the conference, there were only a handful of announcements and live demos, and the biggest one was for a future product not even part of Canvas.

There is a four-pillar strategy underneath all of this, and on paper it holds together. But a strategy is a claim about what the product and services work adds up to, and that claim lands only when the work is visibly underway. Kudos on genuinely handling the breach; poor marks for generating momentum.

Below: the breach response, the continuity roadmap behind it, and the executive question it exposed; the IgniteAI update and the pricing deadline that moved; the product story and the strategy that sits above it; the partner and platform work; and Foundry, along with the governance question already trailing Project Athena.

The Breach Response: Head-on, and then some

The customer organization keynote framed the May incident and the response to it as a turning point rather than a technical failure, with transparency named as the operating principle. Instructure reports meeting with more than 365 institutions and roughly 500 conversations since the event, and it is clear that they were listening. The top ask on continuity planning was gradebook access. The second was communication—both from the company, and help for institutions and faculty to reach their own students.

Measure that against the specific things that were missing in May. At the time I wrote that no named executive owned the response, that the academic continuity issue—schools delaying finals, reworking assignments—went unmentioned, that higher ed’s trade press was being deflected to a static page, and that ecosystem partners had no briefing of their own. All of that has changed. CEO Steve Daly took a named role of responsibility and has kept it. Continuity is now the organizing frame rather than the omission. Partners got their own track. And the company met the room with appreciation rather than message control.

Hardening and continuity

Chief Architect Zach Pendleton handled the technical half. The framing was that attacker economics have changed due to AI advances—exploit timelines have compressed to a fraction of what they were seven years ago, a version of the argument I made in June about why EdTech moved into the target zone.

Remediation for Instructure runs along three lines: reduced attack surface, stronger identity controls including a step-up authentication layer on top of existing SSO, and faster detection.

The academic continuity roadmap is the part institutions strongly asked for. Instructors get a dashboard reminder to back up their courses. Administrators get bulk gradebook exports, entering an early adopter program with no upgrade required, followed by course roster exports so an institution can reach its own students when Canvas is unavailable. Further out and explicitly uncommitted: read-only access during outages, offline grading, and a customer-controlled failover instance.

My read: the roster and gradebook downloads landed better than anything else that morning, which did not surprise me. They address the exact pain institutions felt in May and June, and they carry a message a typical technology company does not send: we are not going to protect and control information that is already yours. That is a real concession of leverage, and the room heard it as one. The caveat is at the far end of the list, where institutions that do not want to depend on Instructure during an outage need the independent failover instance—planned for the coming months on AWS, but described without much detail.

The executive team question

But good intentions, transparency, and a real desire to improve are not the same thing as making the organizational changes required to operate differently. There are signs of change here and signs of stasis.

Start with the changes, which are real and only partly trace to the breach. AI is reshaping how software gets built, and Instructure had already concluded that product and technology should not run as separate organizations. The company is creating a combined chief product and technology officer (CPTO) role and looking outside for the hire, on the reasoning that development cycles and product approaches now differ from anything the industry faced in the cloud era. The breach accelerated the timeline for these changes. CTO Michael Lysaght is departing, with Pendleton adding interim CTO duties until the CPTO arrives. Chief Product Officer Shiren Vijiasingam moves to lead Foundry, a new organization funded and set up by Instructure (more on that below), once that hire is in place. Instructure also changed two of its outside advisory firms over the advice it received during the incident (euphemism alert).

One detail helps explains the technology change. After Instructure’s September 2025 ShinyHunters breach through Salesforce—not a Canvas product issue—the company ran a security audit on its own products and built an 18-month remediation plan. After May, the CEO directed an all-hands effort that led to closure of the known issues in eight weeks. The gap between those two responses is not a resourcing story. It is a judgment story, and judgment is the job.

So the attitude has changed, but has the team beyond the CPTO-based reorganization? It is the executive team’s job to listen to its market and know its own audience. For the cybersecurity incident, what customers objected to was not so much the vulnerability as the response to it. What I criticized in May came down to Instructure not really hearing what customers wanted to see delivered.

My read: Instructure has changed course since June and has proven it is hearing its customers. What I cannot yet see is whether enough has changed to sustain the new attitude. The executive team is not turning over much, and the problem of misreading the audience needs is broad. I see more genuine change in attitude than I do in building an executive team that will still behave this way a year from now. Judge it the way I intend to: by whether the listening holds, and whether Instructure delivers what customers need going forward.

IgniteAI: Shipped, extended, still unproven

IgniteAI Agent shipped in March, and the conference additions were incremental. Bidirectional MCP support means an institution’s own agents can act on Canvas while the Ignite agent can act on partner tools, with Wooclap as the first working integration. File upload lets the agent turn a whiteboard photo or document into Canvas content, and a prompt builder adds wizard-style scaffolding for common workflows.

My read: our spring report (premium only) called the summer conversion of the agent from free to paid the thing to watch, on the reasoning that it would be the first real signal of whether agentic AI has a business model in higher ed. That conversion slipped a month and now resolves into the new tier structure, with Canvas Next as the tier enabling agentic usage this fall. Too early to judge uptake, and worth continuing to watch.

The Product Story: Strategy on slides

Four pillars without the delivery

Instructure does have a strategy, embedded in the four pillars described above. A companion slide maps AI onto this strategy more broadly.

In briefings the company added a picture of the stack underneath: the core applications, a data access platform, an AI fabric of APIs and bidirectional MCP connections, and the intelligence engines on top of that.

I believe that the long-term thinking at Instructure thus far has been more strategic than its competitors, particularly around agentic AI and how it can and likely will change the user experience. Issues like not tying the AI agent to specific parts of the current UI workflow, and how this might enable future LMS deployment not even requiring the standard left column / top menus / dropdown options paradigm of the past two decades.

On paper this is a real strategy. Where Instructure has fallen short is in meeting current customer needs that give confidence in how the LMS will improve the educator and student experience. Real delivery issues. Weeks earlier at Fusion, D2L organized its conference around four pillars of its own, and the reason that worked is that more tangible advancements were already shipping and the pillars explained what they added up to. Instructure certainly has more product delivery offered this year than last, but the level of detail provided and the actual customer usage lags what I saw with D2L.

You can read that two ways. Instructure is not leading on delivery, even while making the bigger bet with the agentic approach, and I heard multiple customers ask for more concrete improvements to core workflows, even small ones. But there is now enough shipped to judge customer acceptance against, which was not true a year ago.

My read: strategy documents are cheap, and this one is better than most. The test of a vision at a users conference is whether customers can watch it run, and this year they were asked to take it on faith too often.

What shipped, and what is close

• Available now: IgniteAI Study Tools—flashcards, summaries, and practice questions generated from instructor-sanctioned content—in the top Canvas tier only. Notebook, for cross-course note-taking, across all tiers and switched on by default in August. A rebuilt learner dashboard, all tiers.

• Coming this fall: portfolio enhancements across all tiers, and an early adopter program for the block content editor, where block-based authoring goes to everyone but the AI efficiencies are reserved for the upper two tiers.

• Rebuilt: Knowledge Chats, renamed from LLM Assignments. It began as a high-stakes AI-graded assessment type and came back as a low-stakes formative check-in—course-grounded, instructor-set talking points, no new gradebook column.

• In the paid tiers: accessibility analytics for institution-wide remediation, and learner dropout-risk prediction.

The Platform Play: Partners, apps, and the LTI cleanup

The least-covered thread of the week was the partner and platform work: a partner portal where vendors maintain product details and configure their API and LTI settings, Canvas Apps as the customer-side hub to discover and manage every integration, and usage and billing data so partners can see how their integrations actually get used.

Two things in it matter more than the plumbing. Partners will be asked to disclose their own product AI features in the portal, so institutions can see what AI is running inside the tools they install. And Instructure is deprecating LTI 1.1: its own integrations move to LTI 1.3 by the end of this year, with partners encouraged—though not required—to follow by the end of the academic year. That ask went out at the partner summit on the conference’s first day, under a framing of shared stewardship of the platform.

There is a second story running underneath the platform work, and it is a commercial one. Instructure is monetizing its partner ecosystem far more aggressively than it used to—a higher cost to exhibit than at competing conferences, paid access to APIs, and a new co-sell arrangement offered in more than one revenue-share flavor. Partners I spoke with are, on balance, happier with Instructure than they were a year ago, and a number of them are trying the co-sell program cautiously rather than enthusiastically. I am not going to detail terms here, but the direction is worth naming, and there are signs D2L is moving somewhat the same way. The economics of LMS partner ecosystems deserve their own treatment, and I plan to come back to them in a future post.

My read: The practical caveat is that an app store is easy to announce and slow to populate, and that an ecosystem-wide migration on a voluntary timeline is the kind of ask that slips. The larger caveat is the posture. Openness toward partners was central to how Canvas won in the 2010s, when a cheap and easy ecosystem was part of the argument against Blackboard. What is taking shape now looks less like that Instructure and more like the Blackboard of the 2000s, when the platform’s gravity was something partners paid for. Back to the future.

Foundry and Athena: The real news, and an early governance problem

The biggest news at InstructureCon was structural rather than product. Instructure Foundry is a standalone organization with its own funding, chartered to build in markets adjacent to the core. Our spring report argued that the most likely near-term disruption in this market would come from a vendor changing how it operates rather than from any single AI feature, and Foundry is an attempt at doing exactly that.

The structure is the interesting part. Investors provided incremental funding so Foundry does not compete for the money core Canvas, Mastery, and Parchment need—an answer to the standard objection that innovation gets paid for out of the products customers already buy. Teams are small and AI-native, and go-to-market is decided case by case: an accreditation application would need a different buyer and possibly its own sales organization.

Foundry’s first project is Athena, an AI tutoring agent that connects to Canvas but does not live inside it. What Instructure argues that separates Athena from D2L’s Lumi Learner Mode and the rest of the category is persistence: rather than generating study material on demand within a course, it maintains a longitudinal model of what a student has actually mastered and adapts as that map fills in. Less a study tool, in the pitch, than a tutor who has known you for a year.

Hinds Community College in Mississippi is implementing this fall alongside a pre-and-post research study run with Instructure’s research team. The company is deliberate about calling Athena a project rather than a product: it enters beta now, with general availability targeted for early 2027 and fewer than fifty institutions testing this fall. It will be sold to institutions on behalf of their students as a separate line item rather than a Canvas tier.

My read: the governance question will be interesting to watch. Because of the learner-centric consumer concept involved, Athena appears to place student usage outside direct administrator control as things are currently defined. I do not believe all control needs to sit in faculty and admin hands, but Instructure will have to navigate the different end-customers of its products, and that navigation is part of the experiment.

This might be the sharpest tension of the conference: a company that spent two days rebuilding trust on institutional partnership introduced its marquee new offering on a path that somewhat routes around institutional control and is outside of the core product.

Worth noting on the other side: Instructure reports students themselves preferring institutional delivery, on the reasoning that they did not want to risk being accused of cheating with a tool their school had not sanctioned, and the company’s own read is that faculty skepticism rather than student uptake is the harder problem.

The new category of learner-centric study tools deserves more room than a conference post can give them, and I plan to come back to the learner-facing AI question across vendors.

Putting It All Together

I came to Louisville with a big question about handling the cybersecurity incident, and Instructure answered it. On transparency this is a materially better company than the one I wrote about in May, period. The breach was addressed head-on, continuity is being built, partners are being brought back in, and nobody was defensive.

The second question, the one about momentum, was not as clear. There is a coherent strategy on paper and a thin year of delivery beneath it, and the one genuinely new product came out of Foundry, a separate organization.

There will be a lot to watch with Instructure over the next six to twelve months, particularly around the impact of the CPTO-based reorganization, whether the exec team continues with the new attitude, and customer acceptance of the company’s emerging AI-enabled enhancements.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

Blackboard’s plan of post-bankruptcy reorganization went into effect at the end of February, four and a half months before the Building Blackboard Together users conference on July 13–15 in Dallas. That the company pulled off a real conference in that window is the first thing worth saying. In March I wrote that this event would be the first visible readout of what the new ownership structure would mean, and that a key question was whether we would see stabilization or repositioning. The answer is stabilization, executed well.

That distinction carries more weight than it sounds. The message this conference had to land was we’re back. The message it could easily have landed instead was we’re alive—a company going through the motions because not having a conference would have looked worse. Blackboard cleared that bar, and the clearest evidence was people rather than product. Attendees were engaged, several announcements landed well, and the company is visibly hiring senior leaders (I joked at one point that I was at the Parchment users conference). I previously noted that bankruptcy voided the old executive agreements and let new ownership rebuild the leadership framework from scratch. This is what that looks like.

At the same time, this is a company in a holding pattern, and there was not a ton of meat in the announcements. Matt Pittinsky—Blackboard co-founder, later Parchment CEO, and the incoming chief executive—cannot start until November because the restrictions from Instructure (owner of Parchment and employer of a few lawyers) took longer to resolve than I expected. Our spring report put the problem this way: what Blackboard lacked heading into mid-2026 was a visible forward strategy that customers, prospects, and competitors could read, and the stretch through this conference and Pittinsky’s arrival would set the tone for the next few years. What the conference showed was a deliberate handful of features, delivered cleanly, rather than an answer to what comes next. That restraint is defensible. It also spends time the company does not obviously have.

It is worth being precise about where Blackboard sits, because the headline share number understates the problem. In competitive selections, the real contest has been Canvas versus Brightspace for years, with Blackboard’s role reduced to persuading customers to stay put—true before the bankruptcy and true today with the most notable exception with Centennial College. What changed in February means that Blackboard has its best chance in a decade to change that market situation: no debt, a single product-line focus, new capital, new ownership. They are not Canvas and they are not Brightspace. They are also a company with 1,200 institutions and 16.5 million learners that is, for the first time in years, fresh and focused on teaching & learning.

Below: the core LMS work, Ally, the AI portfolio, the Cursive acquisition, security, and the refreshing argument made from the conference stage against the curse of credentialism.

Core LMS: The Ultra endgame

The end of Original

Learn Ultra course adoption is past 86% globally, and the end of Blackboard Learn Original is more or less a fait accompli (as it should have been years earlier). The remaining Original courses stop running at the end of this year, with conversion to Ultra and a read-only playback environment following. For a transition that has run the better part of a decade, and consumed an enormous amount of institutional attention along the way, the finish line is close enough to plan around.

What shipped, and what is close

The roadmap sessions were a long list of everyday improvements, almost every item carrying a provenance: community votes, Ideas Exchange counts, or Early Adopter feedback. This is where Blackboard continues to deliver.

• Navigation and layout: full-page views with less clutter, an improved course switcher, and full-width course pages later—work that also lays the foundation for dark mode.

  

• A rebuilt gradebook, in beta now and shipping very soon: more rows and columns in the viewport, streamlined direct grading, and an integrated panel for outstanding grading tasks.

• A new content editor, entering opt-in beta shortly, with reorganized menus, better tables, and improved keyboard navigation. Roughly 1,300 Ideas Exchange requests sit behind it.

• A rubric overhaul: central management, export and import, printable rubrics, and multiple rubrics per assignment—a long-standing request that matters for programmatic assessment.

• Quality-of-life work: a student to-do list that respects accommodations and release conditions, group management improvements, and a privilege letting instructors temporarily lift anonymous grading under system controls.

Outcomes moves into the core license

The one announcement that changes what customers get rather than how it looks is outcomes. Outcome and competency alignment, longitudinal trend data, and mastery insights are now included in the core Blackboard license rather than sold alongside it.

My read: this is a company finishing a Dilbert-esque decade-long platform transition while its competitors are continuing to move forward, and there is no way around that. But the backlog-clearing does a second job. After eighteen months in which customers had every reason to wonder whether the company would still have a real development organization, shipping the specific things people voted for answers that question more credibly than a strategy slide would. The pricing move is the more interesting one, because it runs against what Instructure and D2L are both doing with new, higher pricing tiers. Blackboard went the other way, folding outcomes into the core license and shipping PDF auto-tagging at no additional charge. In a retention year that is a rational trade. It is not a posture that survives AI compute costs indefinitely, however.

Ally: The part that is genuinely ahead

Scale, and the investment behind it

Ally has been embedded in institutional workflows for years and now runs in roughly 1,000 institutions. Two numbers from the conference matter more than that count: Blackboard doubled the size of the Ally development team this year, and 60% of Ally institutions run a competitor’s LMS.

PDF auto-tagging goes general

The news was PDF auto-tagging, now generally available after an early adopter program that drew more than 70 institutions against a target of 20. Untagged PDFs are the most common accessibility failure in course content, and remediating them has traditionally meant Acrobat Pro licenses and document-by-document work. Auto-tagging does it inside the Ally instructor feedback workflow at no additional cost under the existing license, with the instructor reviewing and approving the tags before the file returns to the course. This announcement was the highlight of the conference keynotes.

The rest of the roadmap

The rest runs in the same direction. Coming very soon: an AI alt-text assistant for images inside PDFs, and an enhanced reader alternative format with word emphasis, a reading ruler, and dark mode. Further out: bulk handling of decorative images, guidance for improving already-OCRed PDFs, and rebuilt accessibility reporting.

The accessibility work also runs outside Ally. Audio descriptions for video are available now under the core license, and the Learning Object Repository carries accessibility scoring, so a fix to a shared object flows to every course using it.

My read: this is the one part of the portfolio where Blackboard is not playing catch-up, and the timing is favorable. Our spring report argued that the April 2027 ADA Title II date turns accessibility from a documentation exercise—VPATs and attestations—into an operational product requirement, and that vendors without scalable remediation tooling face higher churn risk. The two gaps we named as largest were untagged documents and audio description for video. Blackboard has now shipped both. A tool that removes a license cost and a manual workflow at the same time is an easy internal sell, and this is the piece of the portfolio least dependent on what happens in November.

AI: A steady portfolio and one quiet bet

AVA, now official

Blackboard reports 10 million interactions with its AI Virtual Assistant (AVA) to date and roughly a 50% acceptance rate on AI suggestions, a rate the company says has held for two to three years. AVA now gets a dedicated panel inside the course, deliberately separated from human messaging, plus course-grounded flashcards, on-the-fly quizzing from course documents, and a rubric check that gives students pre-submission feedback while showing the instructor the full revision trail.

Scholar, announced thin

The genuinely new item is Scholar, a learner-facing AI study tool launching with a select cohort this fall, built by a team drawn largely from outside EdTech. Blackboard framed it around students who struggle with executive function and who already use AI through dozens of competing study apps, then said noticeably less about the product than about anything else on the stage. The overlap with D2L’s Lumi Learner Mode is worth noting.

My read: our spring report credited Blackboard with the most developed deployment of AI tools in the market and the greatest customer adoption of them, with genuine institutional learning behind it. The question we posed for 2026 was whether it would convert that latent strength into momentum. On this evidence, not yet. The AVA additions are sensible and incremental, and they land in a market where Instructure is betting on a platform-wide agent and D2L is betting on AI as an accelerant for content at scale. Both of those are arguments about where the product is going. This was a feature list.

Cursive: Buying a position on authorship

The one place Blackboard spent new money was academic integrity. The company acquired Cursive Technology and launched the product as Cursive by Blackboard.

The pitch opened by dismissing the three existing approaches. Plagiarism detection was built to catch copying, not generation. AI classifiers are statistical guesses whose accuracy degrades with every new model and which are biased against non-native speakers. Blue books work but do not scale, and were described from the stage as a retreat rather than a policy. Cursive instead captures a continuous record of the writing process—keystrokes, pace, revisions, time on task—from first draft to submission, producing something an instructor can review rather than a probability score. Processing runs on the student’s device, and Blackboard says what a student writes stays private during creation. The company reports use at more than 100 institutions.

Cursive is also LMS-agnostic, similar to how Blackboard handles Ally. Blackboard’s own materials list Canvas, Brightspace, Moodle, Google Docs, and Microsoft 365, though today it connects through a browser extension rather than a native integration. That gives the sales team a legitimate reason to open conversations at institutions running competitors’ platforms. A company that badly needs to defend its own base now has something to sell into everyone else’s.

My read: the framing as authorship verification rather than detection is smart positioning and probably the right pedagogical instinct. It is also going to be tested. Our spring report argued that the vendors who pull ahead over the next two years will be the ones that help institutions decide where AI belongs—formative feedback, accessibility remediation, course design, learner support—and where it does not, naming surveillance overreach specifically. Cursive sits on the line between those two lists. Continuous keystroke capture is privacy-sensitive regardless of where the processing happens, and the distance between “proof of effort” and “surveillance” is a matter of institutional policy, not product design. The on-device architecture is a real answer to part of that concern. Whether faculty senates and student governments accept it is a different question, and one that will not be settled by a conference session.

Security: Process without specifics

Blackboard ran a dedicated privacy and security session with the CISO, the privacy and trustworthy AI officer, associate general counsel, and the chief of staff. The content was framework and certification: ISO 27001, 27017, and 27701; FedRAMP Moderate since 2020; a DoD Impact Level 4 provisional authorization added last year; annual SOC 2 reports available under NDA; signatory to the EU AI Pact. That is all well and good if you want to be the EU of EdTech, but in the real world the question is about the platform’s security and the company’s ability to respond to what is guaranteed to be an increase in AI-enabled attacks. No one can claim full immunity, and over-reliance on badges can get in the way of genuinely rethinking platform security.

Sitting in that room a week after D2L used its own trust track to make architectural claims that a customer could evaluate—tenant isolation, egress monitoring—the contrast was hard to miss. Both companies are answering the same market question, raised by the same competitor’s breach.

Labaree and the Case Against Credentialism

The opening keynote went to David Labaree, the Stanford education historian whose books—How to Succeed in School Without Really Learning, Someone Has to Fail, and A Perfect Mess—argue that American education has substituted credentials for learning, that public schooling carries an unresolvable tension between equal opportunity and individual advantage, and that the roughly 4,700-institution sprawl of US higher education is a feature rather than a flaw.

It was wonky, and better suited to me than to a ballroom. It was also an unusual but refreshing topic for an EdTech company to put on its main stage. Vendors in this market sell into the credentialing machinery—assessment, analytics, outcomes reporting, transcripts—and platforming a sustained argument that the machinery has crowded out what it was supposed to measure is not the safe keynote choice.

Pittinsky on the LMS at 30

The Day 2 fireside put the incoming CEO on stage with Lev Gonick (ASU’s Enterprise CIO and a Blackboard independent board member) under an explicit constraint: Pittinsky is still in a cooling-off period and did not speak to Blackboard’s forward direction. What he gave instead was the argument from his LMS at 30 series at On EdTech, written before he was recruited to the CEO role, and his follow-up working drafts. The original theory of change was to make it “obscenely easy” for instructors to put courses online and let the second-order effects follow. Thirty years on they have not, and he spread the blame around: vendors, his own company’s acquisition strategy, RFPs that rewarded deeper features inside existing surfaces, and institutions that do not volunteer for transformation. The system is still a course management system, and on his framing it has not yet earned the L.

On AI he was deliberately deflationary—an ordinary technology that can do extraordinary things, not a ghost in the machine—and treated demographics as the actually disruptive force, including the roughly 41 million US adults holding college credit and no degree. His candidate for AI’s first real application is assessment rather than content generation: address the integrity fear and the path clears for everything else, while feedback indexed to specific learning objectives tightens the link between a credential and demonstrated mastery. He called himself a radical incrementalist, with Parchment as the model.

My read: at a conference where the product story was deliberately modest, the most interesting things Blackboard said were about learning itself. Both stage arguments were too wonky for a ballroom and heavily US-centric, and I do not want to over-read a speaker booking. But it is good that the conference spent this much time with this topic framing, and on the critique of credentialism.

Putting It All Together

In March I wrote that the financial reset was complete and the strategic reset was still to come. Four months later, that remains the accurate summary. I also said to watch three things at this conference: the product roadmap, the sales tone, and whether the message was stabilization or repositioning.

The roadmap is modest and mostly backlog-clearing, with real strength in accessibility and one targeted acquisition. The tone was customer-focused rather than defensive, which is not nothing for a company that spent a year being used as a competitive talking point. And the answer to the third question is stabilization, chosen deliberately and delivered competently.

The risk is straightforward. Blackboard has five months before its actual strategy begins, in a market where Instructure and D2L are both making public arguments about where the LMS goes next. What matters in the next two years is client retention, read through the renewal cycle over the back half of this year and early next year. If it holds, a debt-free Blackboard is a credible competitor with room to invest. If it does not, well then I have some interesting future posts to write.

What I can say is that this conference did the job it needed to do. Blackboard is not Canvas and it is not Brightspace, and it is not going to be either of those things next year. But Blackboard is back. That is a real change from where this company was twelve months ago, and it earns the next year of attention.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Erratum: Corrected Lumi Pro reference.

At this year’s D2L Fusion users conference, I saw many Canadians finally appreciating the full heat of Phoenix (by the way, we’re having a real heatwave right now that is five degrees hotter). But I also saw and heard a company with a strong strategy, but too often that company creates its own barriers.

For the first time since the late 2010s, D2L has a coherent product strategy rather than a collection of features. The last time was the move to the cloud and the ground-up UI redesign that introduced the modern Brightspace experience—a clear thesis about where the product needed to go, with sustained investment behind it. Since then, D2L has added a lot of stuff: AI features through Lumi, the Creator+ authoring toolset, the H5P and Course Merchant acquisitions. Individually useful, but with no obvious thread tying them together.

While we were starting to see coherence last year, this year the thread was unmistakable. D2L now describes its product strategy in four pillars—generative AI transforming teaching and learning, high-quality content creation (native tools along with Creator+, H5P, and the new D2L Createspace), product-market fit for employee and compliance training, and continuous reinvestment in core Brightspace—with trust positioned as the foundation underneath. The real strategy is in how the first two pillars combine: D2L is going after scaling engaging course content as its sweet spot, with AI as the accelerant rather than the headline. The implied customer is one that skews towards coordinated course design and top-down decision-making—institutions and systems that deliberately are looking to scale course redesigns and impact—more than laissez-faire institutions. The development investment lines up with the strategy, and my conclusion coming out of conference season is that D2L currently has the most strategic product and services plan of the major LMS vendors, period.

But this is still D2L, and the company has a chip on its shoulder that shapes how customers perceive it. That posture too often makes the company defensive, and gets in the way of scaling beyond the small but healthy growth it has today.

One example from the conference is D2L’s repeated claims of greater-than-50% win rates that extend beyond what any independent market coverage supports and without backup data. And D2L’s aggressive approach to the Canvas cybersecurity incident and competitor user conferences—direct contacts to institutions during these events, described to me in multiple conversations—left more of a bad taste in potential customers’ mouths than intended.

Both examples reflect the same instinct: managing the perception rather than letting the strategy and the results speak for themselves. And that’s a shame, because this year the strategy is worth talking about. Below, I describe the major initiatives—core Brightspace investments, the Lumi expansion including Learner Mode, D2L Createspace, the ecosystem play, and the trust positioning—and what each means for D2L and the LMS market.

Core Brightspace: Unglamorous but crucial

The description of continuous-reinvestment showed up with several core product improvements this year, and notably with release dates on the slides—a response to feedback (including ours) about vague timelines at last year’s conferences.

The Content Experience—D2L is no longer calling it “new,” with justification, as roughly 70% of customers now use it—gained an immersive full-width viewer this month, with a collapsing table of contents in September and sequential-learning visibility (locked future modules visible to learners) by year end. Outcome alignment moved from a multi-click, multi-upload process to a single curriculum screen across an entire module set—meaningful for competency-based and accreditation-driven programs.

The rest of the core investment reads as an everyday-friction list, and I mean that as a compliment. In many regards, the more the LMS fades to the background, the better for its customers.

• Quiz annotations (scheduled for August, behind a config variable): learners can bookmark questions to revisit and strike through answer options they’ve ruled out—built by an intern, Benjamin, after his own semester as a student.

• Zero-point quiz questions, ending the 0.001-point workarounds for information-gathering questions and academic integrity pledges.

• Group assignments at scale: 3,000 learners per group and 1,000 groups per category (up from a 200-group limit), with delegated marking and multi-evaluator workflows extended to group work.

• Distributed administration: sub-org admins can manage their own campus’s navigation, themes, homepages, and announcements without touching the rest of the institution—previously an institution-wide-default-only proposition.

None of this is flashy, and that’s the point. This is the pillar that answers whether a vendor is still investing in the product its customers use every day, and D2L’s answer was more convincing than most.

The AI Extensions: Lumi grows a learner side

Before the announcement, D2L shared usage evidence for the existing Lumi portfolio of AI tools, with hundreds of thousands of educator-created items, and 98% of sessions citing course material. Lumi Insights also extended from visualizing quiz data to suggesting interventions—edit the problematic question, revise content with AI assistance, or generate personalized emails to the specific students who struggled, in one click.

The commercial context, which we covered in our (premium only) spring report, comes from D2L’s own earnings disclosures: Lumi annual recurring revenue (ARR) grew from roughly $2 million to more than $3.5 million over the final quarter of the fiscal year ending in January—against total company ARR of $219.8 million. The attach rate is above 40% among new higher ed customers, but CEO John Baker has been explicit with investors that AI functionality is not yet showing up consistently in RFPs. Lumi is real and growing, and it is also still under 2% of the business. Lumi Pro as a paid tier is the kind of premium AI packaging we argued this pricing-compressed market would force.

D2L is clearly positioning AI as a method to scale course creation and redesign, describing a 50% reduction in time to build a course (this is not a verifiable claim, but the positioning is important).

The marquee announcement was Lumi Learner Mode (beta this fall), a dedicated study space inside each course that belongs to the student. Learners can highlight content and ask for explanations or alternative examples, save responses into a personal notebook with their own annotations, and launch self-directed knowledge checks—including Socratic open-ended questions where Lumi nudges thin answers toward specificity. Lumi Tutor, previously a floating button, gets a permanent home here, as shown below.

Two design choices matter more than the feature list. Knowledge check results are visible only to the student—instructors see aggregate engagement data, not individual interactions—a deliberate choice to preserve student ownership. And responses are anchored strictly to course material with citations, refusing questions outside it, an approach D2L says it has tuned for two years on the instructor-facing tools.

The pedagogical framing is the most direct answer I have heard from an LMS vendor to the existential question of whether LMS-based AI tools matter when students have ChatGPT. D2L’s argument, supported by its Digital Promise research partnership, is that AI with structure and boundaries accelerates learning while unstructured AI use degrades it—so D2L’s answer to counterproductive tools is a structured one students actually want to use. It also sharpens a strategic contrast in the market: where Instructure is betting on a centralized agent that reasons across the whole Canvas platform, D2L appears to be more focused on embedding AI into discrete teaching and learning workflows—two different architectural bets.

But the room’s reaction was noticeably muted—no visible applause for the announcement D2L billed as something never seen in an LMS. Part of that is likely from the presentation: the introduction didn’t explain enough for customers to know how to think about the new tool. In one conversation I had at Fusion, a customer asked whether Learner Mode was competing with Course Hero or something different—a question that D2L will have to address. Even if the company gets the design right, there is still a risk if institutions buy Lumi partly on the student-facing narrative and students don’t adopt it—a risk D2L acknowledged when I raised it. And that risk is not hypothetical: the dominant campus pattern for LMS AI remains hedging and pilots, with vendor keynotes and roadmaps running well ahead of deeper institutional adoption.

D2L Createspace: The fullest (optional) expression of the strategy

D2L Createspace was announced at last year’s Fusion, but there were a lot more details presented this year behind the tool: a full content management system inside Brightspace—in D2L’s words, “not a repository”—where organizations create, version, collaborate on, and synchronize reusable learning content across hundreds or thousands of courses. The tagline is “create once, improve everywhere”: edit the source version, then sync the change to every course where the content lives, with control over where updates apply and support for master/gold course workflows.

To understand where D2L Createspace fits, it helps to see how D2L’s content tools now layer (my understanding based on what I heard):

• Native Brightspace tools—the Content Experience, document templates, standard assessments—are the base every customer gets.

• Creator+ is the paid authoring add-on with richer elements: accordions, tabs, callouts, styled interactives.

• H5P provides the deeply interactive content types—games, interactive video, drag-and-drop—with AI-powered Smart Import to transform existing materials, plus a refreshed UI and WCAG 2.2 AA alignment this year, timely given the April 2027 ADA Title II compliance deadline facing institutions.

• Lumi Remix sits across these, using AI to transform imported content, apply Creator+ elements, and translate.

• D2L Createspace is the management layer above all of it: where content built with any of these tools gets versioned, shared, and deployed at scale.

Learning services

Services are a key part of this content strategy. D2L’s Learning Services team has used AI for two years to modernize clients’ stale course content at scale—including a “SCORM scraper” that automatically uplifts legacy SCORM courses into rich HTML, work that used to take weeks. Note that D2L makes a related but different claim here than the one above: not build time, but internal course development costs cut by more than half, along with measurable completion gains from modernized courses.

Back to the Aussie future

For those of us with long memories, this is a familiar idea. Australia’s HarvestRoad Hive and the federated learning content management system (LCMS) concepts of the 2000s promised much the same thing: centralized learning content repositories with reuse and governance across courses. Those ideas were ahead of their time and did not in the end become established systems, and to a degree the market fragmented into LMS-native tools, authoring systems, and in the case of large-scale online programs, often custom development.

D2L Createspace somewhat revives the LCMS idea but makes it part of the LMS toolset. On one hand, this simplifies the EdTech ecosystem, but on the other hand it would create more dependence on a single vendor.

D2L Createspace is part of core Brightspace at no additional cost—but core does not mean universal. It enters opt-in early access in August and further rolls out through 2026. And there are significant pieces still being built between now and full availability. D2L reports that the closed beta was the largest in company history. And the embargoed news was a co-development partnership with University of Maryland Global Campus (UMGC), whose skills-forward use case—breaking full courses into atomic, skill-tagged “skill packages” that can be searched and reassembled when a new industry partner signs—ships with the product.

My read: D2L Createspace is the fullest expression of the four-pillar strategy but not a required element of it. On paper there is a lot of possibility for improving course design and updates at scale, but there are risks involved. The idea had a lot of value in 2005, and the addition of GenAI to multiply the volume of institutionally-owned content adds even more value. And for what it’s worth, D2L Createspace and the UMGC video received the most reaction during the Solutions Spotlight.

The Ecosystem Play

The quieter strategic release was Brightspace Apps, an app-store experience for LTI integrations rolling out this summer. Partners publish integrations; admins see the requested data permissions explained in plain language, install in a few clicks, and control deployment scope—institution-wide or a specific department. LTI 1.3 only and existing integrations unaffected. The companion Brightspace Builders program gives partners and institutional developers the documentation and publishing path with much better information presented to institutions when browsing third-party apps.

It is worth noting that Brightspace Apps acts somewhat as a counterweight to the D2L Createspace approach. D2L is simultaneously making it easier to connect third-party tools and more attractive not to need them for content.

D2Link, D2L's integration product connecting Brightspace to enterprise systems like Workday and Oracle, grew from 30 to 70 clients year over year, with the company reporting significantly-reduced implementation timelines. Very few institutions are switching LMS vendors right now, so lowering the cost and drama of migration is one of the few levers available to unlock the selections that do happen.

Trust by Design and the Canvas Breach

The conference’s framing theme was “innovation with purpose, built on a foundation of trust,” with a dedicated trust track running throughout Fusion, including sessions with the CTO and SVP of Engineering. D2L was candid at the executive briefing that the trust track was re-framed and emphasized in response to the Canvas cybersecurity incident.

D2L’s engineering analysis of the breach—a thesis built from public disclosures, not inside knowledge—is that profit-motivated attackers were, in the CTO’s phrase, “logging in, not breaking in”: XSS leading to cross-tenant access with elevated privileges.

D2L’s position is that a similar attack on Brightspace would have failed and been detected quickly, based on two architectural claims: complete security isolation between tenants, with every support or integration action requiring independent per-tenant authentication, and real-time data egress monitoring. To its credit, the CTO paired the claim with appropriate nuance—“you cannot be smug about this”—and described ongoing hardening including reduced administrative access and AI-assisted code scanning. The analysis remains internal; customers are asking for the findings, and D2L has not decided on a sharing approach.

My read: D2L put more emphasis on the “we are more secure” direct argument during analyst day than in main sessions, and the descriptions were credible. I had an offline discussion, getting more in-depth descriptions, and there is real understanding behind the team’s leadership. But there are two caveats—in this age a cybersecurity hack could occur in any system, meaning that the incident response is at least as important as architecture; and poor messaging, tone, or timing can undercut the actual claims.

Putting It All Together

As I step back from the individual announcements, I saw much more coherence in strategy. Core Brightspace got unglamorous, daily-friction investment with dates attached. The announced Learner Mode is perhaps the most pedagogically explicit answer to the ChatGPT question (although still a bet). The content strategy is cleaner across native tools through Creator+ and H5P up to D2L Createspace, with services added on top. And the ecosystem work should make Brightspace easier to connect to, not just harder to leave.

Which brings me back to where I started. With the most strategic product and services plan of the major LMS vendors, D2L does not need inflated win-rate claims or too-aggressive customer contacts to make that case. In fact, those instincts can invite skepticism about a story that increasingly stands up on its own.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

I'm writing this a little earlier than usual as I'm heading to Idaho for the weekend in search of trout and garnets. So, what reading did I manage to squeeze into a shortened week?

The hidden costs of capacity

Universities have a lot in common with ambulance services. That was my conclusion after reading David Oks' excellent post on the cost of ambulance rides. Oks argues that the often absurd cost of an ambulance ride—which can run to many thousands of dollars and is frequently not covered by health insurance—is an unintended consequence of the economics of the industry and the way reimbursement works. What struck me wasn't really the economics of ambulances. It was how easily the same logic explains some of the most persistent debates about the cost of higher education.

Although the post is ostensibly about ambulances, it's really about the difference between transaction-intensive organizations, where costs scale with each service delivered, and capacity-intensive organizations, where most costs are incurred before any service is delivered. Ambulance services and universities are both capacity-intensive organizations, which goes a long way toward explaining why both struggle with the economics of pricing their services.

One historical detail particularly caught my attention. Until the 1960s, at least in the United States, ambulance services were often provided by funeral homes as a loss leader. Hearses and ambulances did double duty—which, unsurprisingly, was not a particularly effective emergency medical system.

At first glance, this sounds like a quirky piece of medical history. But it illustrates a challenge that many public services share: paying for capacity is much harder than paying for individual transactions.

As emergency medicine advanced, a different kind of ambulance service became necessary—one equipped with increasingly sophisticated technology and staffed by trained professionals. These services are expensive, but not because of corporate greed or some sinister private equity conspiracy. Most ambulance companies operate on razor-thin margins and generate little profit. The challenge lies in the underlying economics.

Programming Note: I have been asked by quite a few people about coverage of the LMS conferences going on this month. Due to tight schedules (one ending on a Friday, another starting on a Monday; and to be honest also the World Cup and British Open), I plan to release conference reports from D2L Fusion, Building Blackboard Together, and InstructureCon next week, after all three are complete.

Update 22 Jul: I removed an erroneous description of UPCEA having the most conference-heavy revenue for the three groups mentioned below. OLC has a higher reliance on conference revenue than does UPCEA.

Speaking of conferences . . . many nonprofit organizations around higher education run on convening. Associations and technology consortia bring the field together at annual conferences, and for many of them those events are not a side activity—they are among the largest revenue lines keeping the organization running.

Which is what made 2020 and 2021 such a threat. When conferences stopped, a core revenue engine stopped with them. We are now far enough past that to ask a fair question: how did these organizations fare, and how have they recovered?

Answering it is harder than it should be. These groups—associations, consortia, and more broadly policy organizations, think tanks, and foundations—all file public Form 990 returns if located in the US, but the filings are scattered and rarely lined up side by side. So this is a first pass at doing that comparison view. ProPublica makes these forms readily available, and I compiled the most recent filings for this group—generally tax year 2024 or 2025—supplemented with local IRS XML filings where available. The goal is not to reduce any organization to a few numbers, but to establish a common financial baseline and ask basic questions: What revenue comes in? How much goes out? How much cash is held? And how does executive pay track with scale?

A recovery and a reckoning

Two organizations I have covered recently are worth considering with recent financials provided by ProPublica.

EDUCAUSE is a useful place to start, because its conference is central to its finances—registration and booth-rental income together run higher than what it collects in membership dues. Its revenue fell hard and bottomed in 2022, but it has climbed for three straight years since, with expenses comparatively steady and a 2025 surplus of roughly $2.8 million. It remains below its earlier-2010s peak, but the direction is unmistakably up.

The Online Learning Consortium (OLC) is the harder case. Its revenue did not decline as much as did Educause, even though it is even more conference-dependent—conference and workshop revenue is more than half its total. But that dependence did prove costly, as OLC entered Chapter 11 bankruptcy at the end of 2024, driven not primarily by debt or falling revenue but more by disputed penalties on pre-Covid hotel contracts whose room and catering minimums had become impossible to meet in a post-Covid conference market. Note that the 2024 filing shows a dramatic decrease in expenses, leading to surplus of $266 thousand, but that came after three years of losses in the $475 - $898 thousand range. OLC continues to operate post-bankruptcy, with one change being that the spring conference is now virtual. The fall conference remains in-person with virtual options. This was more of a reckoning.

How these organizations spend their money

Let’s look more broadly at nonprofit organizations to place the conference-heavy organizations in context. The expense chart below makes the differences in organizational models more visible.

One caveat before the chart. I grouped these into associations, technology consortia, and think tanks mostly to keep the visuals legible. Those groupings—and, honestly, which organizations appear here at all—reflect my choices as much as any natural structure in the field. So read the categories as scaffolding, not as a finding. What they do usefully show is relative scale and the fact that very different financial models sit side by side; I would not yet defend them as the right taxonomy.

Three patterns stand out. First, Arnold Ventures is in a category of its own: its latest filing shows roughly $194 million in grants, plus about $17 million in general expenses. That is the financial profile of a major grantmaker, not simply an operating policy organization. Second, most associations and technology consortia are primarily people-and-operations businesses: salaries and wages, followed by general expenses, account for most spending, while events, travel, and other operating costs are meaningful but smaller components. Third, the larger policy organizations, including AEI, CAP, and New America, operate at a different scale from many membership associations, with substantial staff-driven budgets but without anything approaching Arnold’s grantmaking volume.

Internet2 is also distinctive among the technology consortia because of its operating scale and large general-expense component. That is a useful reminder that this group is not financially homogeneous: a membership association, a shared-network operator, a policy think tank, and a grantmaking foundation may all influence higher education, but they do so through very different financial models.

It is not surprising that a large component of expenses for nonprofit organizations is based on salaries and compensation. And the question is not just about total salaries but looking at top executive pay compared to organization size (measured in revenue).

The relationship is real but loose. As you'd expect, leaders of larger organizations tend to be paid more, and the cloud drifts up and to the right, but not too tightly. At any given revenue level the spread in top pay is wide, and the biggest budget does not buy the biggest paycheck: Internet2 has the largest revenue in the group by a comfortable margin, yet its top compensation sits well below AEI's, the highest in the set. A few organizations land noticeably above the trend for their size, ACE most visibly, and a few below it. The color adds a second dimension: blue points ran an operating surplus in their latest filing and red an operating deficit, and the deficits are scattered across the size range rather than clustered among the highest or lowest payers.

Putting revenue and expenses together

Another view that might be useful is to consider both revenue and expenses for both Educause and OLC.

The waterfall chart below shows where the money actually comes from and where it goes. On the revenue side, the conference story is right there in the two tallest bars after memberships: registrations ($7.1M) and booth rentals ($4.6M) together bring in more than membership dues ($8.9M)—which is what it means to say the conference is central to the finances. The rest arrives in smaller streams (contributions, investment income, subscriptions, program services), giving Educause a reasonably diversified top line. On the expense side, the shape is typical of a people-and-operations business: salaries and wages ($12.6M) are more than half of total spending, with operations, general expenses, events, and a modest grants line making up the remainder. What's left after everything nets out is the roughly $2.8 million surplus—the bottom line of the recovery described above.

OLC's waterfall tells the same conference story in more concentrated form, and it carries a caution. One revenue bar dominates: conferences and workshops ($3.6M) is more than half the roughly $5.95 million total, with membership dues a distant second. There is far less diversification here than at Educause—when a single line is that large a share, a few bad conference years can be devastating. The expense side reflects the same model, with events and travel ($1.3M) a heavier component relative to size.

There are other association-based conferences, of course, and I don’t intend for this first view to be comprehensive in nature. But it might also be interesting to look at UPCEA. Where Educause spreads its conference income across registrations and booth rentals, UPCEA's single Conference line ($3.2M) is nearly half of its roughly $7.3 million in total revenue on its own—the most conference-concentrated organization of the three, and well ahead of membership dues ($1.7M), its next largest source. The rest of the top line is a long tail of smaller streams (consulting, contributions, webinars, sponsorship, program services). The expense side is the familiar people-and-operations shape—salaries and wages ($3.7M) lead, followed by events and travel ($1.5M), a proportionally heavy line that tracks the event-driven model. It nets to a healthy $767 thousand surplus, which is the fair coda to the recovery question: UPCEA came through 2020–21 and is now comfortably in the black.

There are more lessons to be learned from looking at nonprofit financial filings, and I plan to keep adding data, refining the expense categories, and building out these organization-level and over-time views. If there's an organization or a question you think belongs in the comparison, reply to this email and let me know.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

Apparently, much like zucchini, surveys are suddenly everywhere. I can barely keep up with all the reports flooding my inbox. Which ones were interesting, or infuriating? I refer to this weekly post as IRTW, so this week the "I" can stand for either.

The strategy gap

Almost every week brings another story about financial stress somewhere in higher education. Reading this year's Inside Higher Ed/Hanover Research survey of Chief Business Officers (CBOs), it's hard not to conclude that we'll be reading many more, and that they're likely to become increasingly grim.

The surprising thing isn't that CBOs underestimate the challenges facing their institutions—they don't. They recognize declining enrollments, demographic change, and mounting financial pressures. What surprised me was how often they stopped short of embracing the kinds of strategic changes those challenges would seem to demand.

The quantitative results alone show that CBOs are deeply concerned about enrollment.

• 46% said declining enrollment was the top financial risk facing their institution over the next five years.

• 46% also said demographic shifts would significantly affect their institution between now and 2030.

• Growing enrollment was the top-ranked strategy (44%) for improving their institution's financial stability.

The write-in comments reinforce this picture. Many CBOs recognize the need for a different strategy based on sharper institutional focus and differentiation.

Yet neither this concern nor the strategic thinking reflected in the write-in comments translates into bold choices around delivery models or reaching new student markets. Instead, most CBOs continue to prioritize investment in physical infrastructure.

Which is how we end up with over-investment in physical facilities, as I described in a recent post about the State of Facilities in Higher Education.

The emphasis on physical infrastructure isn't an isolated finding. The same pattern appears in the survey's treatment of teaching and learning. Despite enrollment concerns and repeated discussion of efficiency, CBOs show only limited enthusiasm for expanding online and remote delivery.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

I am hunkering down in my air-conditioned office while Utah swelters in record-breaking heat. Before Phoenix-based Phil and all the folks who were at D2L's Fusion this week jump in to tell me to hold their beer and opine about real heat, let me tell you what I read this week.

The missing variable in the AI hiring debate

AI adoption intensity matters in predicting post-adoption hiring trends.

Over the past year, higher education has been consumed by debates about AI's impact on hiring, especially entry-level hiring. Unfortunately, much of the evidence has been surprisingly weak. Many studies estimate which jobs are exposed to AI rather than examining what firms actually do. Exposure is a useful thought experiment, but it is speculative, freezes AI capabilities at a particular moment in time, and tells us little about how organizations respond.

New research from the Ramp Economics Lab avoids that imprecise approach and has some encouraging news. The central finding is surprisingly simple: AI spending alone changes very little. What matters is the intensity with which AI is adopted.

Instead of estimating exposure, the authors use observed AI spending from more than 21,000 firms linked to employment records. The paper's most important contribution is identifying a threshold effect. Low-intensity AI adoption had no measurable impact on hiring. High-intensity adoption increased overall employment by about 10% and entry-level hiring by roughly 12%.

The authors argue that high-intensity firms aren't simply buying more ChatGPT licenses. Their higher spending is correlated with more sophisticated AI tools—coding agents, APIs, and workflow integration—and, perhaps more importantly, with organizations capable of redesigning work around AI. There is a lesson here for higher education. Buying AI licenses is adoption. Redesigning advising, assessment, curriculum, and administrative workflows around AI is intensity.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

I’m currently at the first of three LMS user conferences, D2L Fusion, to be followed by Building Blackboard Together next week and then InstructureCon the week after. Expect conference notes and discussions later this month.

In the meantime, I wanted to provide two updates on enrollment analysis resources.

International Enrollment Addition

The interactive enrollment analysis tool, available exclusively to On EdTech+ subscribers, continues to evolve thanks to feedback from our readers (thank you). A big functional addition is the inclusion of international enrollment data, which is particularly important given the current turmoil and enrollment changes in this area.

The Institution Dashboard now has a breakout of the number and percentage and year-over-year changes of nonresident students compared to domestic, for both undergaduate and graduate levels, based on IPEDS Fall Enrollment data. This can be shown by sector, or state, or by selected institution and peer groups.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

Over the past 12 months, the higher ed community has faced a whirlwind of rulemaking from the Department of Education (ED), making some of the biggest changes to education policy in at least two decades. And as Politico Pro described yesterday, there is more to come.

A major factor impacting colleges and universities is not just the scope of changes but also the speed of changes — the amount of time between seeing draft rules for negotiated rulemaking (NegReg) sessions and the final rules being published.

Has the timeline for rulemaking been compressed? And if so, at what cost?

Confirmation of speed

People have long lamented the slow, laborious pace of ED’s rulemaking process. Put changes on the federal agenda, collect initial public comments, publish draft rules and conduct NegReg sessions that often take months, take the revised rules and publish a Notice of Proposed Rulemaking (NPRM), collect final public comments over a period of 1-2 months, revise those rules and then publish Final Rules.

The short answer is that yes, the major rules from the Trump II Administration are far quicker than the Biden, Trump I, Obama II, and Obama I Administrations.

To see whether this speed is genuinely new, I mapped every major Department of Education higher-education rule going back to 2009, grouped them by administration, and measured how long each one took to travel from the first Negotiated Rulemaking session (where ED and negotiators work through draft regulatory language) to a published Final Rule. The pattern is unambiguous. Across the rules I examined, a typical rule took about 17 months to make that trip, and the fastest anyone had managed before now was roughly nine months. The three Trump II rules — RISE (student loan limits) at about seven months, and the two AHEAD rules, Workforce Pell at five months and Institutional Accountability at six — are not merely quick; they are the three fastest rules in the entire 2009–2026 record, and they clear the old floor by a comfortable margin. Some of that was forced, since the One Big Beautiful Bill Act set a statutory July 1, 2026 effective date that pushed ED to move fast. But forced or not, the result is a compression of the rulemaking calendar with no precedent in the period I looked at, and it came alongside an equally unusual run of consensus in the negotiations themselves.

The stacked bar chart is the clearest way to see it. Each bar is a single rule, and its height is the full Neg-Reg-to-Final duration in months, split into two pieces so you can see where the time actually went. The lighter lower segment is the stretch from the first negotiating session to the NPRM, and the darker upper segment is the stretch from that NPRM to the Final Rule. The shaded background bands mark the five administrations, so you can read left to right and watch the cadence change by era.

Trump II is the only era that is short top and bottom. What jumps out is that the red Trump II bars on the right are not only the shortest overall, they are short in both segments at once: ED did not simply rush the comment-to-final step while taking its time up front, or the reverse — it ran the entire pipeline fast. The long bars elsewhere get long in different ways, which is part of the point. State Authorization in 2016 (34 months) stretched almost entirely in the lower segment, because ED sat on negotiated text for roughly two years, while the 2019 Borrower Defense rule stretched in the upper segment, bogged down between proposal and finalization.

There is a strategy

Regardless of whether you like or dislike the policies of each of the rules, what we are seeing with the Trump II Administration is unprecedented in speed. But there could be an argument that this record should have an asterisk — Congress passed OBBB with an unusual level of detail, essentially writing much of the resultant regulations, and they set aggressive timelines.

Rather than an asterisk, however, I would argue that what we are seeing is a considered strategy designed to get fast results.

The first step is getting Congress (through a lot of strong-arming) to actually legislate with specific policy details. The second step is picking NegReg negotiators who, while they purposely are chosen to represent divergent points-of-view, are personally inclined to listen and actually negotiate. The third step is choosing federal negotiators and moderators who are inclined to make hard decisions quickly with the express aim of reaching consensus (meaning the outcome rules are largely written even before the NPRM). And the fourth step is to very quickly review and respond to public comments in an all-hands-on-deck manner.

There is a cost of this strategy, and not just from forcing colleges and universities to scramble to implement the final rules. There are mistakes that are made in the final rules due to the rush, ones that expose ED to litigation risks. Congress set the clock, but the most litigable choices — e.g., how professional programs are defined, whether undergraduate certificates count — were ED's own, not Congress's.

But more on that in a future post.

I would expect this speed strategy to continue for at least the next two years, and there will be consequences.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

After yesterday’s post about the Department of Education (ED) final rules on Institutional Accountability, I had a few updates worth sharing. One is on specific examples of the specific years for the first program cohorts to be evaluated; and two is sharing ED’s slides as presented at NASFAA today.

The official final rules are now available at the Federal Register.

What are the first cohorts that will face rule determination?

I mentioned yesterday that there were two delays built into the Final Rules impacting when the first programs will be officially judged, with the basic cadence described in this infographic.

There is complexity in effective dates and exemptions, and accordingly I had an email exchange this morning asking for clarity on specific dates for first program cohorts, so here is my reading of how the July 1, 2027 rule effective date plays out, working backwards.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

Tomorrow, the new Department of Education (ED) institutional accountability rules become official. For the first time, the federal government is putting a broad program-level earnings test at the center of Title IV eligibility—broad as in for nearly all academic programs. The unofficial final rules came out yesterday, which covers all of the substance, but the formatting and page / footnote numbering will change tomorrow with the official version.

I have covered the reality of Earnings Premium measurements and the details of the accountability concept, and it is time to take stock of what has changed and what hasn’t with this week’s release.

What Has Changed

The primary changes from the public comment period (i.e., the new information in yesterday's draft) are primarily about schedules. In essence the effective date has been pushed back, but not by a single amount and not for everyone at once. There are really three dates to track.

• The piece that binds institutions to the framework—the program participation agreement (PPA) language under which a college agrees to be subject to these rules—takes effect 60 days after publication, or around September 2026. That is only about a two-month delay from the original July 1, 2026 target, so the legal commitment arrives almost on time.

• For most programs, though, the accountability machinery itself (i.e., the earnings premium measurements that can actually cost a program its loan eligibility) is effective July 1, 2027, a one-year delay.

• And for the tipped occupations (where more than 50% of workers receive at least $100 in tips), cosmetology most prominently, the first determination that can fail a program does not arrive until July 1, 2029, a two-year delay.

Hence "pushed back one to two years" is accurate, as long as you understand that it is really three different dates.

The Cadence

To see why those dates land where they do, it helps to walk through how this metric travels from program completion to a consequence. The cadence runs in six steps:

• Students complete a given program in Year X.

• Graduates earn income for a given tax year in Year X+4 (the rules are based on fourth-year earnings).

• These earnings are reported to the IRS, nominally the following spring, in Year X+5, when returns are filed.

• Institutions report their program completer data to ED by October 1 of Year X+6.

• ED matches and collates the records and publishes the dataset in February of Year X+7.

• The determination—pass or fail, and the consequences that follow—takes effect July 1 of Year X+7.

None of this is current; the metric is always looking several years into the past.

For most programs, with a July 1, 2027 effective date, the first dataset produced under the new rule will not publish until February 2028. That one-year delay matters because the rule is built on six-digit Classification of Instruction Codes (CIP6), while every dataset we can see today—College Scorecard, PPD2026—is four-digit (CIP4). Think of CIP6 as where students actually get degrees or certificates, such as Occupational Therapist Assistant or Radiologist Assistant, whereas CIP4 is the broader grouping of Allied Health and Medical Assisting Services.

That means we have a strange transition year ahead. The rule is real, the legal framework is arriving, but institutions being held accountable still will not be able to see the program-level CIP6 results that matter most. We will be blind for an additional year on what the CIP6 framework actually does to specific programs.

The tipped occupations are a special case, because the data and the consequences now run on different starting calendars. The earnings data for these tipped programs will first be reported in the February 2028 dataset, but the first determination that can actually impact one of these programs will occur July 1, 2029 (the two-year delay).

There were some other adjustments in the final rules highlighted by ED that narrow rather than expand the rules, including appeals schedules and loan eligibility exemptions.

In a nutshell, the final rule is official tomorrow, but with one-to-two-year delays from what has been discussed for the past year.

What Hasn't Changed

Everything else. Every flaw I have described in the Earnings Premium is now a fact of life (and, I am sure, the subject of future lawsuits). There were no changes to the underlying concept: earnings measured four years after completion, compared against survey-based aggregations that stand in for an earnings threshold (the median earnings of 25-34-year-olds with only a high school diploma for undergraduate programs, and a more complex bachelor's-earnings formula for graduate programs). Commenters pressed on the age band, the four-year window, and the elimination of the debt-to-earnings test, and ED held its position on all of them. Undergraduate certificates, which the statute never named, remain in the final rules.

 My own public comment, which focused on ED’s analysis of gender-impact concerns, was rejected essentially in full. ED repeated the same basic footnote, still without publishing the regression details needed to evaluate the claim, and still using a field-of-study control that removes much of the disparity negotiators were concerned about. In ED’s own published data, removing that control changes the estimated gap from roughly zero to about $12,000 per year. I’ll return to that issue separately, but the short version is that ED chose not to add transparency.

What to Expect

I'm glad ED recognized that an effective date two days after the first view of the final rule text would have been foolish. Colleges and universities have a one-to-two-year reprieve, depending on the program.

Most institutions aren't prepared to evaluate their programs at a strategic level, and that's the real shift. By strategic I mean that program review can no longer be delegated down to the deans and the colleges as a routine academic exercise. Once a program's earnings outcomes can cost it access to federal loans—and with that its enrollment and its revenue—the question stops being academic and becomes one of institutional risk that belongs with presidents and boards. This is a new era of institutional risk, and most institutions' governance hasn't caught up to it.

We've seen a preview of this in the state-level reviews of program viability, where systems and legislatures have started pruning programs on enrollment and break-even finances. The earnings rule extends that same logic to every institution that touches Title IV federal aid (loans and Pell Grants alike), on a federal, ROI-based measure that no campus controls. Whatever an institution thinks of the metric, it now has to plan around it.

The institutions that are best prepared will not be the ones waiting for ED’s first CIP6 dataset in 2028. They will be the ones using their own student records now to understand which programs are exposed, which ones are safe, and where governance needs to change before the first determinations arrive.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

The best thing I read all week was the order prohibiting fireworks in my state this coming July 4 holiday. Given the ongoing drought and fires, that is a relief. But what about EdTech?

In my reading this week, I found myself thinking that AI has taken two steps forward and one step back. We're asking better questions about learning and work, but we're still letting hype shape investment.

Where does AI belong?

For most of the past three years, we've been asking the wrong question about AI and learning. We keep going back and forth about whether to ban it or embrace it and, if we do embrace it, how to use it in a course.

This is the argument made by a pair of engineering faculty members from the University of Zagreb, Croatia in a new article in which they make the case for what they term placement. After reviewing a broad range of recent research on the impact of AI, they come down in favor of its use—but only in certain places within the learning process.

The critical question, then, is how to decide where AI should be used in a learning experience so that it does all of those good things, rather than the kinds of things that can harm student learning, such as simply providing the answers.

To help instructors answer that question, the authors develop a six-stage framework for learning: Prime, Probe, Point, Attach, Strengthen, and Test.

They argue that AI belongs in some stages but not others. The first and last stages (Probe and Test) should remain AI-free, while AI can be used as scaffolding through the middle stages.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

Update 26 Jun: As is on cue, OpenAI has now agreed with the US government to delay & stagger the release of its GPT 5.6 models. Same story as Anthropic, but without the antagonism, and it reinforces the analysis below.

A few weeks ago, in an On EdTech+ post titled The Resistance Is Real, I argued that the AI resistance on display at ASU+GSV wasn't a passing annoyance. It's likely to be one of the major EdTech stories of the next year or two. Some of that resistance rests on weak arguments and outdated assumptions. But not all of it — there are legitimate concerns underneath, particularly about AI's impact on deeper learning and on students' future job prospects.

That resistance has mostly been B2C: students and faculty reacting as end users. Two weeks into the Anthropic Fable 5 saga, I think the B2B version is emerging, even if it won’t look on the surface like the B2C resistance. It will show up as procurement friction, institutional hesitation, regulatory pressure, and pushback against American-dominated AI infrastructure.

Get ready for AI sovereignty to become an EdTech issue.

The Fable 5 Nutshell

The short version of the Fable 5 story is that for several months Anthropic touted its new Mythos class of AI models as having a new level of risk for cybersecurity. Two weeks ago, Anthropic released Fable 5, the first Mythos product release, with safeguards to supposedly remove the cybersecurity risks along with biological weapons development and distillation attempts of other AI models.

Based on my two-day experience with Fable 5, I agreed with most analysts who called it a meaningful leap in capability, significantly beyond OpenAI's GPT 5.5 and Anthropic's own Opus 4.8. Within days, the U.S. government imposed export-control restrictions on Anthropic's most advanced models, citing national security concerns partially triggered by Amazon’s testing. Anthropic responded by pulling access to Fable 5 and Mythos 5 for all customers, arguing that the directive made it impossible to comply cleanly while continuing to operate the models.

There are real arguments to have on the merits. Maybe Anthropic was right to be cautious with the injected safeguards. Maybe the Trump Administration was right to intervene. Maybe the government overreached based on an incomplete grasp of how probabilistic AI systems actually work. Those debates matter — but for institutions and EdTech vendors, the takeaway is simpler: access to a major AI capability that customers had already started using was pulled back.

Not because a customer misused it. Not because a university changed its mind. Not because a vendor failed to renew a contract. The model was effectively taken away because Anthropic and the US government each showed they had both the power and the willingness to decide what AI capabilities could remain available.

That will likely get the attention of organizations building strategy around AI, even if the immediate issue gets resolved, with Fable 5 re-released, in the near term.

Anthropic has already shown that it will change model behavior, adjust prompts behind the scenes, downgrade model access, and make unilateral product decisions in the name of safety. And that Anthropic leadership believes it is solely positioned to determine. The US government has now shown that it may treat advanced AI models as export-controlled strategic technologies, with access subject to national-security judgment. In its sole discretion. Agree or disagree with either side — the combined lesson makes it viscerally clear that customers do not fully control the AI capabilities they are beginning to depend on.

Enter AI Sovereignty

In a technology context, sovereignty is about control. Who controls the infrastructure? Where does the data go? Which legal regime applies? Who can inspect the system, change it, or turn it off?

For AI, that set of questions is broader than it was for cloud computing, where the focus was mostly about data residency. AI sovereignty is about control over the models, the data path, the hosting environment, the allowable uses, the continuity of access, and the ability of institutions to rely on those capabilities over time.

Two angles matter for EdTech.

The first is national sovereignty. Non-US governments, universities, and systems of higher education will increasingly ask how much of their AI strategy can safely depend on American companies and American policy. What happens during the next trade dispute? What happens if the US decides a capability used in advising, assessment, research, or student support is too sensitive to export? What happens if a model is available one semester but restricted the next?

This is where the Fable 5 saga becomes bigger than Anthropic. It gives non-US institutions a concrete reason to ask whether American AI can be trusted as infrastructure. That doesn't mean they'll all abandon the leading AI tools — in many cases they won't have realistic alternatives. But it could mean that procurement rules, national AI strategies, and institutional risk reviews may start looking for different answers: local hosting, country-of-origin requirements, model transparency, open-weight options, or preferred support for national and regional providers — France's Mistral, the UAE's G42, and the national AI pushes now underway from India to the Gulf.

The second angle is organizational sovereignty, and this one applies inside the US as well.

What happens when a college builds operational processes around AI — not just classroom experiments, but advising workflows, tutoring, accessibility support, research assistance, student services, compliance, and administrative automation? What happens when an LMS, assessment platform, or student-success system embeds AI based on one model provider, and the vendor's pricing, roadmap, and revenue projections all assume that access will continue?

Here, sovereignty is not a geopolitical issue. It's an operational-risk question: is there trust that AI capabilities will continue to be available, who controls these decisions, and what happens if policies changes? Put another way — is the institution buying a capability, or taking on a dependency?

Cloud Computing and Data Residency

EdTech went through a somewhat related version of this with cloud computing.

In the 2000s and early 2010s, many non-US institutions were wary of cloud-based EdTech, fearing that student data could pass through US data centers or be exposed to government surveillance. Privacy rules, data-domicile requirements, and procurement expectations slowed cloud adoption, particularly in Europe. Vendors could insist that cloud hosting was secure, scalable, and better than local infrastructure — but that didn't answer the additional questions of where does the data live, and which government can reach it?

One inflection point came when AWS opened its Frankfurt region in 2014. The AWS Ireland data center had been open for a few years, but there was no acceptance of that providing a viable and long-lasting solution to the cloud data residency issue for EU countries. AWS Frankfurt had a much bigger impact. It gave vendors and institutions a practical and believable answer they'd lacked: European data could be hosted in Europe under a more acceptable and viable operating model. Cloud EdTech didn't become frictionless, but the infrastructure changed the nature of European higher ed procurement decisions.

Seen in this light, AWS Frankfurt wasn't only a relief for nervous buyers. It was a win for the vendors who built it and who adopted it. Amazon turned a procurement objection into a competitive advantage and took business that hesitant or US-only competitors couldn't. Similar changes came from EdTech vendors ready to adopt its usage. The lesson for AI isn't merely defensive. Whichever vendors give institutions a credible answer to "who controls this capability, and what happens if it's withdrawn" will gain a competitive advantage.

Assuming that AI sovereignty becomes a much bigger EdTech issue, AI will need its own AWS Frankfurt moment. The analogy isn't perfect — cloud sovereignty was mostly about data location and legal access, while AI sovereignty is about something harder: control of capability.

There's also a risk that EdTech answers a legitimate question with a rearview-mirror understanding of AI itself. Much of the sector is still operating from an AI conversation two years out of date — chatbots, cheating, syllabus statements, generic privacy worries — and procurement rules built on those assumptions could slow adoption without solving the sovereignty problem. Cloud went through a similar pattern: real issue, understandable fear, some blunt early regulation that delayed adoption, and eventually new infrastructure compromises that largely unlocked the market. AI sovereignty looks likely to follow the same arc.

The open question is what those compromises will be. Local model hosting? Open-weight models? Regional AI clouds? Vendor-neutral AI gateways? Contractual guarantees of model continuity? Viable back-up options? Government-backed national models? Some combination of all of them? Whatever it is, I doubt the current "just plug in the best available model and move fast" phase stays acceptable in every market.

What This Means for EdTech Vendors

This matters for vendors in ways that are easy to underestimate.

AI is no longer just an end-user tool sitting outside institutional systems. It's starting to be built into products. The LMS isn't just adding a chatbot; advising tools aren't just adding better search; assessment products aren't just adding writing detection. Vendors are rebuilding around AI-assisted workflows — and institutions may find themselves depending on capabilities they never separately procured or fully understood.

So adding AI features is no longer just a product-design decision. In some markets it will be seen as a procurement risk. Institutions may start asking which model provider is being used, where prompts and outputs are processed, whether data is retained for training, whether the feature can be disabled, whether a local or open-source model can be substituted, and what happens if the underlying model is withdrawn or restricted.

Vendors will want to market AI as a capability layer. But I suspect that institutions will also evaluate it as a dependency layer. That complexity could slow down adotpion, but it could also provide competitive advantage for whoever answers it first.

Put It on the Radar

I don't think AI sovereignty will dominate EdTech procurement in 2026. The immediate story in higher education will still be faculty resistance, student use, academic integrity, workflow redesign, and vendor messaging. It will be fascinating to watch the July LMS conferences try to thread it: promoting AI capabilities while navigating the B2C resistance minefield.

But I'd add the B2B question to the radar.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Upgrade to the On EdTech+ newsletter.

What caught my eye and interest this week?

Some interesting reads on EdTech and higher education, but also a few spectacular catches at the Women's T20 World Cup.

A lot of what I read this week left me thinking about the dangers of assuming someone else will solve your problem—whether that is employers waiting for colleges to produce skilled workers or institutions assuming future growth will justify today's investments.

Everyone wants skilled workers—nobody wants to create them

The Open University Business Barometer 2026 report offers some fascinating insight into the inconsistent and uncommitted way many businesses think about employee training. The report is naturally focused on the UK and is ostensibly about skills shortages and the potential of adults Not in Education, Employment, or Training (NEETs) to help fill those gaps. But what it reveals about training, reskilling, and upskilling has much broader applicability.

Reading the report, I kept coming back to a simple conclusion: everyone wants skilled workers, but fewer organizations seem willing to invest in creating them. The headline statistic is that 57% of employers in the UK report a skills shortage, and 42% expect it to worsen over the next five years.

Right off the bat, there seems to be a contradiction. Unemployment is higher than it has been in recent years, which should make it easier to hire the people you need. Yet 43% of employers report hiring fewer people over the past twelve months—presumably because they can't find the skills they need.

These skills shortages have led to real and serious impacts for employers, including increased workload and turnover and lower morale.

Given those shortages, you might expect employers to be aggressively investing in developing talent. That is exactly what they say they are doing.