I'm obsessed... with coding using AI.

(Sorry for the clickbait-ish subject line.)

But seriously... coding with AI is getting ridiculous. It’s becoming incredibly easy and incredibly challenging at the same time.

On one hand, I’m barely writing any code myself. AI can do it 10x to 1,000x faster than I can — especially since I’m a beginner. Tools like OpenAI Codex or Google Jules can even run in the background and fix things while I’m off doing something else.

On the other hand... making a usable website or app is way more than just "coding." There’s version control. Databases. Tests. Security. Scalability. Usability. Architecture (different kind from the one we are used to). Deployment. And more.

All of these are things I’m just starting to wrap my head around.

And it kind of gives me a unique lens into what our profession might look like once non-engineers can "do structural engineering" with AI.

Let me explain.

Coding with AI vs. Engineering with AI

An amateur AI-enabled coder like me can now get a draft web app up and running. But...

• It’s going to have security flaws

• Trying to add features will probably break things

• The setup might not be right for the job

• And when things do break, I won’t know how to fix them quickly

Similarly, in the near future, an amateur using AI might be able to spit out a draft framing plan. But...

• They won’t understand how the details fit together

• They won’t know why some things work and others don’t

• They won’t catch subtle judgment calls

• Renovation? Good luck deciphering the as-built (if they even exist)

• And of course , they won’t (and shouldn’t) be stamping anything

Things will evolve as the tech advances...
But we’re still far away from AI replacing the full depth of what an engineer actually does.

It’s one thing to go from zero to a decent-looking prototype.
It’s another to make that prototype robust, secure, and usable.

Same in our world.

Getting from a blank sheet to a polished SD set might soon be doable by anyone with AI.
But turning that into an actual buildable CD set? That’ll still take experience and engineering judgement.

A Small App Experiment: ACI 318-14 vs. 318-19

With that said, I’ve been doing a lot of experiments lately.

Here’s one that actually worked, using Gemini (deep research) + Firebase Studio + Jules.

Remember how ACI 318-19 updated the shear equations and suddenly you needed a lot more reinforcing? (I talked about it here)

I built a little app to compare the cost difference with ACI 318-14.

I added two more options too:

• one where you boost flexural steel to get shear to work,

• and another where you increase member depth instead.

Then it compares all three using different f'c values to see which is most economical.

You can even click in and see the full line-by-line calc.

Check it out here and let me know what you think.

Shear Check ACI 318-14 vs ACI 318-19

Took a lot of trial and error to get that experiment running and working.

I’m thinking of putting together a running log of all my experiments… wins and fails. Perhaps something will spark an idea for you too. Stay tuned.

Alrighty, that’s it for now.
Have a great weekend!

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

I’ve also been experimenting with using AI to generate memes based on the email I just wrote… but it’s kind of like asking the Terminator to write jokes. They’re not funny (at all), and it has to explain what the meme even means.

But somehow, with that context, the whole thing becomes slightly more humorous, in a dad-joke-ish kind of way.

Here is one (scroll down for explaination):

AI’s explanation:

My thought… yes of course, a hardhat and blueprints in the background will present the complexity of using AI in structural engineering!

Happy Friday! Andy from Back of the Envelope here.

Last week, we talked about how ChatGPT works at a fundamental level [link].

It's an extremely sophisticated pattern recognition system, constantly predicting the next word (or token) to generate human-like responses.

One of the limitations of this approach is that it's still making educated guesses — great for writing and reorganizing texts, but not so great for math and engineering.

But that started to change over the past 6 months as researchers improved both the models and how we interact with them.

In this email, I want to pull back the curtain a little more… to explore what’s changed and what it means for us structural engineers.

Let’s go!
(Estimate read time: 2 minutes and 30 seconds)

Not long ago, if you asked ChatGPT to design a beam, you’d get something that sounded smart... but wasn’t quite right.

It might get the math wrong. Or it would pick a size using Fy × Z and forget to mention that it skipped over lateral bracing or deflection checks.

Why? Because it was trained to predict the next word, not to "reason."

Think of it like a student with really really good memory who's seen a lot of problems and solutions, but hasn’t actually learned to think through them step by step.

That was the case… until researchers came up with a clever workaround.

Before we continue, a quick note:

If you are working on a steel building and you haven’t heard of Durafuse, you might want to check it out.

It is a proprietary steel moment frame connection that offers high ductility and fast recovery after an earthquake.

And they’ve got your back with full design team support related to the connections: calcs, plan check comments, RFI responses, and shop drawing reviews.

Click here to learn more: https://go.sehq.co/durafuse

How Does "Reasoning" Work with Prediction?

In simple terms: the model first plans what it’s going to do, then walks through the problem step by step, checking itself along the way.

For reasoning-capable models like ChatGPT o3, Grok 3 (click on “Think”), or Gemini 2.5 Pro, this happens behind the scenes.

But you can catch a glimpse of the thinking process.

Grok 3 “Thinking”

What’s happening is that the model is still making predictions — but now it’s doing it “out loud.” And by talking to itself, it builds a kind of short-term memory to keep track of what it’s already figured out.

By producing that ongoing internal dialogue, it gets better at structured, step-by-step logic.

Try It Yourself

You can simulate this even in non-reasoning models by starting with a prompt like this:

It’ll generate a series of guided prompts that you can use to simulate an AI thinking process.

(By the way, companies like OpenAI have already started integrating some thinking behaviors into non-reasoning models like GPT-4o. So you’ll likely see more and more of this.)

Limitations to Keep in Mind

One limitation is speed. Thinking takes time. It’s processing more steps and building that internal chain of logic.

Another is that it’s still probabilistic, not deterministic.

If you ask the same question twice with slight variation, you might get two different versions of a similar answer… because it’s still relying on statistical prediction.

For example:

….will often give two very different outputs.

So in practice, it’s useful for checking your work — but harder to trust unless you know what to expect.

It’s kind of like working with a junior engineer. Over time, you build up trust based on consistency and quality.

With AI, though, your “junior” might change from prompt to prompt.

This is also where tools and agents come into play.
We’ll dive into that in future emails.

Until then, enjoy experimenting.

Have a great weekend!

Andy Lin
Back of the Envelope/Structural Engineer HQ

PS.

If you’ve got a minute, could you help me out by filling out this 1-minute survey about moment frames? Thanks! [link]

Happy Friday! Andy from Back of the Envelope here.

Today I want to share three things that have had a huge impact on my life recently.

Each one carries its own weight. But when you hold them together, something shifts.

The first two apply to just about everyone. The last one will hit different if you're a parent.

I could go deep on each one, but since we’ve only got a few minutes, I’m going to keep it tight. Let me know if any of these resonates and I’ll happily dive deeper in a future email.

Let’s go!
(Estimated read time: 3 minutes 0 seconds)

1. The 90-Second Emotional Response

Ok, I’m going to be openly vulnerable for a second here...

I’ve struggled with managing anger and frustration.

Not usually with work or engineering-related things. But with life stuff… and parenting.

If you know me in real life, you’ve probably never seen this side of me, because most things just don’t get to me.

But every now and then, something cuts through — and when it does, it really messes with my mind.

I’ve tried the usual stuff: meditation, breathing, relaxation techniques. Some helped, but it never quite clicked.

Until recently, when I was listening to an audiobook called The Let Them Theory by Mel Robbins (highly recommend, by the way), I stumbled across the "90-second emotional response rule."

The concept comes from Dr. Jill Bolte Taylor, a neuroanatomist:

That stopped me in my tracks.

What this means is that when something really pisses me off, my brain floods my body with adrenaline, cortisol — all that fight-or-flight stuff.

But it only does it for roughly 90 seconds.

After that... it’s on me. I can either keep replaying the story that triggered me, or I can let the storm pass and shift.

Which explains why the whole "just breathe" advice doesn’t seem to work when we’re in the heat of the moment. Because the chemicals haven’t dissipated yet.

So now, when something triggers me, I wait.

I count 90 seconds.

Then I choose.

And that leads into the second concept.

2. We Only Have About 4,000 Weeks to Live

The average life expectancy in the U.S. is about 77 years.

77 x 52 = 4,004 weeks.

If you’re like me, you’ve probably already gone through half of those.

I made a spreadsheet to visualize it.

Each dot = 1 week. Each row = 1 year. The blank dots? That’s what’s left.

Seeing time like this… it changes things.

It reminds me how short life is. It helps me catch myself before I spiral.

Why stay pissed over something small when I only get this many weeks total?

The 90-second rule becomes easier to practice when I know how precious each moment is.

And that brings me to the third concept.

3. By the Time Your Child Turns 18, You’ve Spent ~90% of the Time You’ll Ever Spend with Them

Let that sink in for a second.

Back of the envelope math:

• First 18 years: spend every day with them = 18 * 365 = 6,570 days

• After 18: maybe I see them every other week for 30 years = 52/2 * 30 = 780 days

• Total: 6,570 + 780 = 7,350 days

• Time spent by 18: 6,570 / 7,350 = 89.4%

Nearly all the in-person time I’ll ever have with my kids... happens before they’re 18.

I added this onto the same chart from earlier. A little rectangle showing how much time I’ve got left until all of them turn 18.

…Every dot matters.

To bring it all together:

• When something or someone pushes our buttons, remember the 90-second rule. Let the chemicals run their course. Then choose our response.

• Zoom out. Life is short… especially when measured in weeks. Most of the little stressful events won’t matter in a month, let alone a year.

• And if it’s our kids testing our limits... remember how little time we actually get with them. That perspective alone is rewiring everything for me.

Thanks for reading!

Andy Lin
Back of the Envelope/Structural Engineer HQ

PS.

Speaking of time — if you want to spend less of it stressing over moment frame connection design and the chaos it can bring during construction, check out DuraFuse Frames.

It’s a code-approved moment frame connection.

And they’ve got your back with full design team support: connection calcs, plan check comments, RFI responses, shop drawing reviews — the whole shebang.

Here’s that link again: https://go.sehq.co/durafuse

PPS.

I remember dropping off my son on his first day of school. He looked worried, so I asked, “What’s wrong?”

He said, "How long do I have to go to school for?"

I smiled and said, “Until you’re 18."

He paused, nodded quietly, then asked: "You’ll remember to come get me when I’m 18, right?"

(not a true story)

Good morning! Andy from Back of the Envelope here.

By now, you've probably heard about the M7.7 earthquake that struck Myanmar last week. My heart goes out to the tens of thousands affected 😥.

Just like the email I wrote back in 2023 about the Turkey earthquake, this one is tough to write. So many lives lost… and many loved ones left behind.

But, just like I said back then: we're structural engineers.

We talk about failures. Earthquakes. Collapses.

Not to sensationalize, but to learn.

And to make things a little safer next time.

Most of the failures near the epicenter make sense: older buildings, soft stories, weak construction.

But one failure... stood out.

It happened 685 miles away — in Bangkok.

A modern, under-construction high-rise: the State Audit Office tower.

Let’s dive in.
(Estimated read time: ~4 minutes)

Quick disclaimer: The official investigation is still ongoing. Facts are still trickling out, and there’s a lot of speculation floating around. I’ll do my best to separate the signal from the noise.

My heart dropped when I saw this video:

X/Twitter video link

and this:

Reddit video

…the speed at which the building came down is terrifying.

And the fact that it’s so far from the epicenter? That’s what really shook me.

source (https://x.com/engineco16/status/1905627964785434643/photo/1)

So I started digging.

What we know so far

Confirmed facts:

• 33 stories, 449 feet tall

• Budget: ~$63M USD

• Still under construction, but topped out in April 2024

Likely facts
(based on a Chinese technical article that was later taken down — but archived. Link below):

• Structural system: prestressed flat slab (no beams) with exterior concrete columns and core walls

• 500 KSC (≈7000 psi) concrete for vertical elements

• 350 KSC (≈5000 psi) concrete for slabs

source: archived Chinese article

Reasonable inferences:

• The building likely followed local Thai code, which references recent ACI standards

• Bangkok is generally considered a low seismic risk zone (no nearby active faults)

• But the city sits on deep soft soil, which can amplify long-period seismic waves

• This soil amplification effect was supposedly added to the seismic design maps in recent years (link to the research at the end)

And now... speculation

According to a LinkedIn post (link below), the recorded acceleration response spectrum in a nearby station shows a peak at around 1.2 and 1.8 seconds (see yellow and blue lines below, which represent acceleration in the east-west and north-south direction).

That’s a bit different from a typical response spectrum where short periods dominate.

Let’s do a quick back of the envelope calc:

At that period, the spectrum suggests the building may have experienced ~2 m/s² acceleration.

Yet, if designed per code (assuming that’s what the brown line represents), the target design acceleration might've been closer to 0.6 m/s².

That’s more than 3x the design force!

So what does that mean?

It suggests the building may have experienced seismic demands well beyond what it was designed to handle.

And if the core walls weren’t designed for high ductility, and if the lateral displacements were large, the gravity system (columns and slab joints) may have failed due to deformation incompatibility.

Once that happens, collapse can become a cascade…

source (https://www.bangkokpost.com/thailand/general/2993072/collapse-criticism-unfair-state-audit-office-chief)

Of course, this is just a hypothesis. The truth may look very different as more data comes out.

But I’d like to point out one thing:

In the aftermath of disasters like this, it's easy to reach for blame:
design flaws, construction errors, corruption, cost-cutting.

And you’ll see that all over the web.

But sometimes, even when everyone follows the rules, nature changes the rules.

The Bangkok collapse reminds us that codes aren’t guarantees. They’re approximations built on what we know.

And events like this? They expand what we know.

That’s how we get better.

Alrighty that’s all for now. Thanks for reading!

Yours truly,

Andy Lin
Back of the Envelope/Structural Engineer HQ

PS:

Speaking of earthquake resiliency, if you haven’t heard of Durafuse, it is a proprietary steel moment frame connection that offers high ductility and fast recovery after an earthquake.

Click here to learn more: https://go.sehq.co/durafuse

PPS:

Links mentioned above:

• Chinese technical article (use Google Translate to convert it to English): Archived link

• Seismic hazards in Thailand - a compilation and updated probabilistic analysis: Read here

• LinkedIn post on response spectra: View post

• Google Maps link to the tower: Link

Oops… FYI there was an error in one of the links in the email I sent earlier. Here’s the corrected version.

Happy Friday! Andy from Back of the Envelope here.

Today I want to share a strategy I’ve been testing: using AI and humor to cope with stress.

This came about while dealing with some plan check comments on a project that seemed stuck in an endless loop.

You ever been there?

You’re responding to what should be the final outstanding comment, but instead, the email chain gets longer… and longer… and now it’s weeks later, no one remembers what the original issue was, and somehow one comment has multiplied into three.

It’s frustrating. And it disrupts your other work because now you’ve got to pause, rewind, and dive back into an issue you thought was settled weeks ago.

Well… I don’t have a magic fix for this.

But I do have something that might lighten the load.

A way to inject a bit of levity that helps you feel better about the stress, so you can re-engage with less dread and more energy.

Let’s go!
(Estimated read time: 3 minutes 51 seconds)

Why Humor?

Quick story.

Back in the late 2000s, Conan O’Brien was crushing it on his Late Night show. Eventually, he got the dream gig: hosting The Tonight Show.

I was a huge fan. I watched nearly every episode when it launched in 2009.

It was clear how much he loved doing that show — he even said it was the fulfillment of a lifelong dream.

But not long after, NBC pulled the plug. (2010 Tonight Show Conflict).

It was a mess. Public, painful, political.

And yet... Conan stayed funny.

Even as the walls were closing in, he kept cracking jokes:

He made fake bits about putting the show on Craigslist.

Or “wasting” NBC’s money by buying a giant ground sloth fossil to spray caviar onto a Picasso.

Point is: humor made a painful situation feel less suffocating.

It gave him (and us) room to breathe.

And as I later learned, this isn’t just anecdotal; humor can actually lower cortisol, increase dopamine and endorphins, and balance serotonin.

In other words, it helps your brain chill out.

researched using Grok

How to Use AI to Find Humor in Stress

So, how can we bring this into our day-to-day as structural engineers?

Simple: take something that’s stressing you out, and turn it into a joke.

But unless you’re a stand-up comic or an unusually witty person under pressure, that’s not easy to do...

…until now (thanks to AI).

Here’s the prompt I use with ChatGPT (or any AI chatbot):

That’s it.

You’ll get something unexpected, and potentially hilarious.

And if you want to take it further, use Suno.com to actually turn it into a song.

Other prompt variations I’ve tried:

• “Write a funny, self-deprecating song about me (a structural engineer) based on the following situation…”

• “Write a stand-up comedy bit about this scenario…”

• “Write a comedy film plot starring Will Ferrell, Adam Sandler, and Steve Carell…”

You get the idea.

The goal isn’t to fix the problem. It’s to step outside of it for a moment and see it from a funnier, less personal angle.

That shift alone can help you return to the task with fresh eyes and more resilience.

Quick side note:

Speaking of resilience, if you haven’t heard of Durafuse, it is a proprietary steel moment frame connection.

But unlike most systems that require major repairs after a severe earthquake, Durafuse only needs a replacement of the bottom fuse plate.

In other words: Faster recovery, less downtime.

Click here to learn more: link

A Word of Caution (Real-Life Example)

AI-generated humor is great, but one caveat: humor is subjective.

What makes you laugh might not land the same with others… especially the people involved in the stressful situation.

I learned this the hard way.

One morning, I was feeling overwhelmed by the daily chaos of getting my kids to school on time. (Parents, you know the drill.)

So I asked ChatGPT to write a funny song about it:

I also used Suno and turned it into a song (link at the bottom).

I thought the song was hilarious.

My daughter... did not (I accidentally showed it to her. I know, big mistake.)

Turns out:

1. What feels like lighthearted fun to a stressed parent might feel like mockery to a sensitive kid.

2. Satire and sarcasm don’t always land with younger brains.

3. No one likes to be the butt of the joke, especially not when they’re already struggling.

4. She wanted to be taken seriously, and this felt like I was brushing her off.

(Wait… how did a structural engineering newsletter turn into a parenting confessional?)

Anyway… lesson learned. Be mindful about sharing your AI-generated comedy.

It’s a tool for you — to help you process and cope.

Most of the time, it’s best to keep the joke to yourself.

Let it do its work — so you can laugh, recharge, and move forward with grace.

Alright, that’s it for now. I hope this little trick helps you carry a bit less stress and a bit more joy into your work.

Now if you’ll excuse me, I’ve got a daughter to apologize to.

Until next time,

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

For parents out there with pre-teens, you might enjoy this song.

Good morning!

Andy from Back of the Envelope here.

Today, I’m predicting the future.

Well… kind of.

With AI evolving at breakneck speed, do you ever get that weird mix of excitement and unease?

Those are the kind of questions I've been asking… and I've been putting a lot of thoughts into this.

And today, I want to share some of those thoughts, because the better we understand what’s coming, the better we can position ourselves ahead of the curve.

Let’s go!
(Estimated read time = 3m 20s)

1. The Relentless Progression of AI and the Rise of the “Pseudo” Engineer

Remember last week when I shared how NotebookLM helped draft a concrete repair recommendation? And then I joked that AI could also write a song about it?

Well… turns out, that wasn’t a joke.

I took those lyrics, fed them into the latest ChatGPT model (GPT-4.5) to enhance them, and then used Suno v4 (song-writing AI) to generate a fully produced J-pop-style song— with vocals, music, everything.

And now, I can’t get this concrete repair track out of our heads!
(link in the bottom — check it out after reading the article)

This got me thinking…

AI is progressing at a ridiculous pace.

First, it was just text generation. It got really good, and at times, almost mind-blowing.

Then there are images, coding, audio and video — each rapidly improving.

Now, the hot topics are AI agents (AI doing real tasks like deep research and compiling reports) and robotics (have you seen the NVIDIA’s “Blue” robot from yesterday!? Link below also).

What does this mean for us?

It seems that the barrier to entry for any profession is getting lower, at least at the conceptual or schematic level, if we use AEC language.

There are already some AI-assisted software for architecture and land development that can generate building layouts rapidly (links below).

It’s only a matter of time before a generative AI tool for structural engineers shows up.

Maybe it produces rough framing concepts.
Maybe it suggests column placements.
Maybe it lays out preliminary lateral systems.

The generated result may be conceptual and most likely inaccurate, but convincing enough to an amateur.

And that’s the shift we need to be aware of.

Imagine this:

An architect or developer runs an AI tool that spits out a structural concept before talking to an engineer. They show up with AI-generated structural plans, thinking they’ve already done some of the work, and ask:

"Hey, since we already have a concept and schematic design, can we skip those phases and get a lower fee and start with design development or construction document?"

Sounds far-fetched? Maybe.

But we should be thinking about it now, because AI-generated content is going mainstream.

Soon, we won’t just be designing… we’ll be verifying, guiding, and correcting.

And that changes the conversation.

The question isn’t "Will AI replace us?"

More like, "How will we differentiate ourselves from what AI can do?"

2. The K-Shaped Economy: Enhancing vs. Replacing

Got this idea from entrepreneur Shaan Puri (worth a read - link below).

Here’s the gist:

A lot of people think AI will replace jobs.

But that’s not quite right.

AI doesn’t replace jobs.
It replaces tasks, or job skills.

And that’s an important distinction.

Think of Our Job as a "Bundle of Tasks"

Instead of seeing our jobs as one big, fixed thing, imagine breaking them down into a collection of tasks.

Take the RFP phase, for example. We:

• Review the RFP.

• Identify structural concerns.

• Communicate clarifications.

• Define the scope.

• Estimate and develop fees.

Each of those can be broken down even further.

Some tasks, like summarizing an RFP or identifying key requirements, can already be enhanced with AI.

But most of the tasks we do?
AI can’t replace them — yet.

How This Connects to the "K-Shaped" Future

Source: Shaan Puri

Right now, most structural engineering tasks fall into three categories:

1. Unaffected – AI can’t help or replace.

2. Enhanced – AI can assist but not fully replace.

3. Replaced – AI can automate.

Right now, most of what we do is in the "Unaffected" category.

But over time, as AI improves in computer vision (understanding what's on screen and in drawings) and agentic workflows (e.g., AI doing things on our computers for us), more tasks will shift from Unaffected to Enhanced, and eventually to Replaced.

Which brings us to this:

The Future Value Shift

In addition to continually keeping up our engineering knowledge and expertise, one of the highest-value skill in the next 5–10 years will be managing AI tools.

Because when AI starts taking on more and more tasks, the engineers who know how to integrate AI into their workflow will become more valuable than those who ignore it.

Put another way, embracing AI isn’t about giving in to technology.

It’s about using it to level up our expertise.

Ahh, there’s so much more to explore, like “how do we embrace AI”?… but this email is getting too long so I’ll save it for another one.

Hopefully this sparks some ideas and gets you thinking more about it (ideally with more excitement than worry 😆).

Until next time,

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

If you haven’t heard of our sponsor, Durafuse, I put together a quick, 30-second read on how this moment frame connection benefits us structural engineers.

Take a look and let me know your thoughts!

P.P.S.

Here are the links I mentioned if you want to check them out:

• Concrete Strong (Suno AI Song)

• NVIDIA's Blue Robot

• Hypar

• Testfit

• Shaan Puri’s K-Economy Article

Happy Pi Day! Andy here from Back of the Envelope.

Today, let’s talk about how to use AI to augment and supplement your structural engineering knowledge and expertise.

This is especially useful for informational tasks — like minor concrete crack repair.

Let’s dive in!
(Estimated read time: 3 minute)

Quick heads-up!

In case you missed it, I’m running a bi-monthly $50 giveaway—and you can enter by doing one of these three things: Check it out here

I’ll be compiling entries Sunday morning (3/16) and drawing a winner. If you’ve referred someone to the newsletter, hit the Click to Share button at the bottom to check their status — only confirmed sign-ups count!

Concrete Crack & Spall Repair — Background

As structural engineers, we all run into this scenario:

An owner or occupant notices cracks and spalls in their concrete building. They want to know: is this a structural concern?

And… what’s the best way to repair it?

Diagnosing cracks involves a lot: shape, pattern, size, spacing, location... you get the idea.

Then comes the recommendation phase: how to patch and repair.

And this is where AI saved me a ton of time…

The Challenge

On a recent job, I had to assess an existing building riddled with minor racks and then recommend a repair approach so the owner and contractor could get a rough order of magnitude pricing.

I’ve done this several times before.

I knew the answers were buried in ACI RAP (Repair Application Procedure) bulletins. And I knew I could use Sika products like Sikagrout and SikaQuick VOH…etc.

ACI RAP

But here was the challenge:

• There are a dozen RAP bulletins, and I don’t have them memorized.

• I wasn’t sure if Sika’s product data aligned with the ACI RAP bulletins or if they conflicted.

I needed a fast way to cross-check these sources.

NotebookLM to the Rescue

If you haven’t heard of NotebookLM, it’s an AI tool by Google (free) that lets you upload documents as "sources,” and then it answers your questions based only on those sources.

It’s different from chatbots like ChatGPT or CoPilot because:

1. It has a large context window – meaning it takes way more information into account when making inferences.

2. It’s set to be minimally “creative” – meaning it sticks to the facts instead of making things up.

Perfect for research-heavy tasks like this.

How I Used NotebookLM for Crack Repair Research

First, I uploaded:

• ACI RAP bulletins

• Several Sika product data sheets

Immediately, NotebookLM generated an overview of the documents.

Then, I started asking questions:

NotebookLM pulled answers directly from my sources with references. This was key. I could double-check each answer instead of blindly trusting AI.

It also refreshed my memory on key details I had forgotten.

Then, I asked deeper questions:

Finally, I had it summarize my recommendation, incorporating the Sika products and tailoring it to my case.

After a few refinements, I had a solid draft to work from.

With several more rounds of reviewing, editing, and back-checking… my final recommendation was done!

AI Saved Me Hours… But It’s Not Perfect

AI can misinterpret information. It can also make incorrect inferences.

So make sure to always verify the sources and make sure you understand them yourself.

Thankfully, NotebookLM makes this easy by showing exactly where it pulled information from.

Other key things to note:

• Some organizations (like ASTM) prohibit their documents from being uploaded to AI tools.

• Never copy-paste AI-generated text without reviewing and understanding it first.

We’re the professionals.

It goes without saying: AI is a tool, not a replacement for expertise.

Want to Go a Step Further?

You could take your recommendation and run it through ChatGPT with a different perspective:

Or, because why not... ask it to generate a rap song about concrete repair.

🎤 "Crack Fix Flow" 🎤

Cracks in the concrete, we fix ‘em right,
Epoxy injection, lock it tight.
Wire brush clean, then seal that line,
Sikadur flowin’, smooth design.

For bigger gaps, we grout that space,
Sikagrout-328 sets in place.
Spalls get patched, no need to stress,
SikaQuick VOH, we clean the mess!

…And that’s a wrap for today!

Hope this helps. Let me know if you have any questions!

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

If you’ve had enough concrete for today and want to talk steel instead, check out Durafuse.

A proprietary moment frame connection that saves you time and saves the owner money when disaster strikes!

Here is the link: https://go.sehq.co/durafuse

Good morning! Andy from Back of the Envelope here.

We've talked quite a bit lately about goal-setting and AI, so today we're shifting back to structural engineering — because, at heart, that's what this newsletter is really about.

Today, we're diving into Open-Web Steel Joists (OWSJ), specifically existing ones.

I recently encountered them on a project, and there's plenty to share.

Let's get into it!
(Estimated read time: 3 minutes)

First Things First: What are OWSJ?

You've definitely seen these before:

source: Vulcraft

They're essentially standardized steel trusses made from angles, bars, or plates.

They're called "standardized" because the Steel Joist Institute (SJI) sets clear guidelines and specifications that manufacturers like Vulcraft follow.

For example, when you use software such as RAM Structural Systems or RISA Floor, it'll spits out joist sizes like "14K6" or "20LH05."

The designation "14K6" means a 14-inch deep, K-series joist designed to meet specific load requirements: in this case, for a 14' ft span, it’s 825 plf for safe factored loads, and 550 plf based on deflection limits.

(In other words, the “6” at the end by itself has no specific meaning other than being a label.)

SJI Load Tables

Manufacturers must follow these load criteria, which gives us engineers confidence when we call out a designation.

What this also means is that the actual web and chord member sizes can vary depending on who built them and when.

The Tricky Part: Existing Joists

Here's where things can get complicated.

In existing structures, even if the joist designation is clearly marked on as-built drawings, you usually won't know the exact member sizes.

Maybe they are shown on the original shop drawings?

But chances are... the owners don’t have them.

So, what's the game plan?

First, figure out your new loading conditions.

Next, consult the SJI historical load tables to gauge if you're close.

Finally, decide whether to bypass the existing joists altogether and just add new members, or, add reinforcements to existing members.

Reinforce or Add New?

Reinforcing existing joists can quickly become complex.

On my recent project, I didn't end up going the reinforcement route — but I did plenty of research.

I'll share some good information I found (so you don't have to do as much digging when you encounter these on your next project):

• Historical Load Tables – (already mentioned above) Essential data for evaluating existing joists.

• Historical standards & specifications

• Retrofitting Webinar (YouTube) – One-hour SJI webinar on retrofitting existing buildings.

• SJI Technical Digest 12 – A detailed guide (paid resource but looks promising).

• Joist Reinforcement Tool – A free Excel tool by SJI with helpful reinforcement calculations.

screenshot from the SJI reinforcement tool

You'll also likely have to measure these joists yourself (or delegate to someone comfortable climbing ladders).

SJI provides a measurement form to help guide your efforts.

Also, supposedly, if you don't have the designation but have the measurements, you can send them to SJI and they can help you identify them.

Don't Overlook Your Fee!

Hopefully, this insight helps you clearly communicate to your clients the substantial effort involved in evaluating existing framing systems.

When drafting your proposal, make sure to clearly account for tasks such as:

• Field visits and detailed measurements

• Evaluating existing joists using historical data

• Determining whether reinforcement or adding new members is more suitable

• Designing and detailing necessary reinforcement solutions

Sometimes we underestimate how involved a project can become until we're knee-deep in it, especially if it's something we haven't tackled before.

Thinking ahead and clearly communicating this effort upfront can save everyone headaches (and unexpected costs) down the road.

Plus, being transparent about these efforts from the start will help demonstrate your expertise. 

Alrighty, that's all for now. If you have any other helpful ideas, please let me know!

Talk soon,

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

Durafuse is currently testing a pretty innovative solution that could genuinely change the game.

If your project involves potentially retrofitting existing moment frames, you might want to talk to Jared — just fill out the quick form at the bottom of this page: Durafuse Frames.

P.P.S:

How do steel joists apologize after an argument?

.

.

.

They rebuild the truss.

Yay it’s Thursday! Andy from Back of the Envelope here.

It’s only been a month since I wrote about DeepSeek’s reasoning model, R1… and wow, a lot has happened since then.

So today, I’m keeping the "goal setting" chat short (just enough to get you on track), before curing your FOMO with a rapid-fire AI update.

Let’s go!
(Estimated read time + 2 minutes)

Before we dive in:

I put together a quick, 30-second read on how Durafuse Frames benefits us structural engineers.

It’s still a bit draft-y, but take a look and let me know your thoughts!

The Quarterly Goal Setting Cycle

I’ve hinted at this a few times before. Even built a little tool (the 12-Week Plan) to test the idea.

The core concept?

Dr. Andrew Huberman suggests framing goals within a macro timeframe of 12 weeks.

Instead of setting a vague “This is my year to do X” resolution (which, let’s be real, often fizzles out by February), you set a clear, focused goal for 12 weeks — essentially a quarterly sprint.

Why 12 Weeks?

The problem with year-long goals:

• Too much time → We trick ourselves into thinking we have forever.

• Lack of urgency → January to October? Meh, we’ll “get to it.” November? Holiday mode. December? Full surrender.

The 12-week cycle fixes this:

• Urgency kicks in → A visible finish line = motivation boost.

• Faster feedback loops → Progress checks = dopamine hits that reinforce effort.

• Habit formation window → Research shows it takes ~66 days (~2 months) to lock in a habit. A 3-month sprint sets you up for lasting change.

• Natural rhythms → Business cycles, seasons, and academic terms — all naturally follow quarters, making this potentially easier to integrate

Putting It into Practice

Here’s how Huberman breaks it down:

1. Pick ONE major goal. The fewer, the better. (Talked about this a few times before in previous emails.)

2. Define success. What does “done” look like at the end of 12 weeks?

3. Time-block it. Decide exactly how many hours per week you'll commit, and on which days. (12-Week Plan helps visualize how this adds up.)

4. Track weekly. Studies show regular check-ins double your odds of success.

5. Review & adjust. After 12 weeks, reflect: What worked? What didn’t? Tweak and repeat.

And that’s it. Simple, powerful, and (most importantly) doable.

Now, Let’s Talk AI

Since DeepSeek R1 dropped on 1/20, the AI world has been on fire. Here’s what’s new, in chronological order:

• ➡ OpenAI released GPT-o3-mini – better reasoning, free (with limits).

• ➡ Perplexity added "Thinking Mode" – also free (with limits).

• OpenAI launched "Deep Research" – a web-searching, 30-60-source-scanning, mega-detailed research tool (paid).

• ➡ Perplexity added its own "Deep Research" – slightly less detailed but way faster (and free!).

• ➡ xAI released Grok 3 – now with “Thinking” and “Deep Research” modes (free, limited).

• Anthropic dropped Claude 3.7 Sonnet (yesterday) – supposedly insane at coding (paid).

(➡ means some of the things worth for you to try right now)

And in just a few hours… OpenAI is hosting a livestream. What’s coming next!?

Wild times.

Alright, that’s it for today. Talk soon!

Andy Lin
Back of the Envelope/Structural Engineer HQ

Happy Thursday! Andy from Back of the Envelope here.

We’ve been on a deep dive into the science of goals lately. In case you missed it, check out what we’ve covered so far here.

Today, let’s unpack another fascinating topic from Dr. Huberman: The Myth of Accountability.

Or, put another way: Should I tell the world what I’m about to do so they can keep me accountable?

Let’s dive in!
(Estimated read time: 3 minutes).

Quick side note:

In case you missed my last email, Durafuse Frames is now sponsoring Back of the Envelope.

I’ll share more about them later, but if you’re unfamiliar, Durafuse is a proprietary steel moment frame connection.

But unlike most systems that require major repairs after a severe earthquake, Durafuse only needs a replacement of the bottom fuse plate.

In other words: Faster recovery, less downtime.

Here’s a short video if you’re curious: link

(If you’d like to know more, hit the “Contact” link on their website and mention that Andy from Back of the Envelope sent you!)

The Traditional Take on "Accountability"

We've all heard it: Tell someone your plans, and you’ll be more committed.

The theory is that by broadcasting our goals, we create external pressure to follow through.

For example, a while back, I announced that I’d share everything I know about goal-setting, AI advancement, and how both can benefit structural engineers.

By doing that, I felt like I had to deliver—otherwise, I’d be letting you down. That kind of external pressure can work.

But this “tell the world” strategy only works under certain conditions…

Huberman’s Twist: The "Accountability Myth"

According to research, sharing our goals too soon can actually backfire.

Why?

Because when we announce our intentions, our brain’s reward system gives us a premature dopamine hit...

Meaning, we feel a false sense of accomplishment—before we’ve even started.

The result? We become less motivated to take action.

At first, this seemed quite counterintuitive. So I did some digging. Here’s what’s happening under the hood:

1. Our brains crave progress. Dopamine, the ‘feel-good’ neurotransmitter, spikes when we achieve something.

2. When we tell someone our goal and get positive feedback, our brain registers it as a mini-success. Dopamine is released—even though we haven’t actually done the work.

3. Dopamine follows a "prediction error system."

   • If the actual reward exceeds expectations, dopamine surges.

   • If the reward is less than expected, dopamine drops.

Now, apply that to goal-setting:

• When we only talk about our goal, our brain assumes progress is happening.

• But when reality sets in and we haven’t actually done anything, dopamine levels plummet.

• That drop makes it even harder to stay motivated.

This is why announcing big plans—like a New Year’s resolution—feels great at first but makes it harder to follow through later.

TL;DR: Telling the world your goal before taking action can sabotage your chances of success.

Finding the Balance

Here’s the upside: Dopamine isn’t just about reward—it also fuels effort.

Studies show that motivation is strongest when rewards are tied to actual progress, not just intention.

In other words, if we shift our focus to effort over outcome, we stay engaged for the long run.

Quick Counterpoint

By the way, this doesn’t mean we should never share our goals.

The key is timing and who we share with.

• Timing: Keeping goals private until we’ve made real progress protects intrinsic motivation.

• Who we share with: Find people who give constructive feedback and track progress, rather than just offer praise.

Putting It into Practice

That was a lot—so let me summarize:

✅ Start Solo: Work on your goal before talking about it. Delay sharing until you’ve taken action.

✅ Break goals into effort-driven steps.

✅ Selective Sharing: Share progress with people who will hold you accountable in a productive way, not just celebrate the idea.

✅ Stay Focused: Regularly remind yourself why this goal matters to you.

By understanding how our brains respond to goal-setting, we can tailor our strategies for better outcomes.

So next time you’re tempted to announce your big plans, consider keeping them to yourself—at least until you’ve got momentum.

And that’s all for now. Thanks for reading!

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

By the way, if you haven’t already, could you please fill out this 1-minute Durafuse survey? It’ll really help me understand how familiar you are with the system. Thanks!

P.P.S.

What’s the easiest way to keep your New Year’s resolution to read more?

.

.

.

Turn on the subtitles on your TV.

Good morning! Andy from Back of the Envelope here. Got some exciting news...

The newsletter just landed its first sponsor! 🎉

(I’ve known Jared from Durafuse for a long time, and they have been generous enough to let me experiment with this for a couple of months to see how it goes.)

Don’t worry, nothing’s changing - just a small snippet added to the usual content.

What This Means for You:

1. You’ll get insights into a system that potentially makes life easier for engineers while delivering real value to owners.

2. I’ll be able to give back to some of the readers who share the newsletter regularly and those most engaged in the SEHQ Pro community -- something I’ve wanted to do for a long time.
   I’ll dive into more details in a future email, but for now, I’d love your help.

Could you please take 1 minute to answer a quick survey? 

I want to get a sense of how familiar you are with Durafuse Frames before we kick off this experiment.

👉 Take the 1-Min Survey

Appreciate it—and thanks for being part of this journey!

Andy Lin
Back of the Envelope/Structural Engineer HQ

Gooood morning!

Andy here from Back of the Envelope.

The last two weeks, we dipped our toes into AI (because, well… DeepSeek had everyone buzzing).

Now, let’s get back to cracking the science of goal-setting because we still have that New Year’s resolution that we need to accomplish.

Quick recap:

So far, we’ve talked about the importance of "priority" and "lofty" goals. We also covered how defining specific, measurable actions can make all the difference.

Today, we’re tackling something that feels like it should work but doesn’t.

It’s called The Post-it Fallacy.

Essentially, why sticking your goals on our fridge, mirror, or laptop does nothing (and what to do instead).

Let’s go! (Estimated read time: 2 minutes).

The Post-it Fallacy

Here’s the classic move:

We write down a goal on a sticky note.

Slap it on the mirror. The fridge. The edge of our computer screen.

The idea?

Seeing it every day will remind us, keep us on track, motivate us.

Except… it doesn’t.

Why?

Because our visual system is working against us.

Our brains are wired to tune out things that stay the same in our environment.

Whether our fridge is cluttered with papers or that sticky note is the only thing on it—it will eventually become invisible to us.

Our brains adapt. They cancel it out. Like background noise.

This is why “reminder” sticky notes don’t work. They stop reminding us.

The Science: Habituation

It’s wild to think that our best intentions—to keep our goals top-of-mind—actually fail because of how our brains work.

This is all thanks to a psychological process called habituation: The more we’re exposed to a stimulus, the less we notice it.

Studies on warning signs (like workplace safety reminders and hand-washing posters) show that repeated exposure makes people pay less attention over time.

Even switching up the color or format of these signs doesn’t help.

Why?

Because our brains are smart. They recognize it as the same thing and tune it out anyway.

And our Post-it note is no different. Even if we move it around, our brains will still adapt.

So… What Does Work?

Huberman suggests a simple fix:

Write your goal down—again and again.

Every day. Or at least every few days.

This reactivates your brain’s engagement with the goal.

You can either use a notebook. Or write on a new sticky note (different color each time) and place it somewhere different every time.

The key is active engagement—not passive exposure.

Write it. Re-write it. Change it up.

That’s what keeps your brain paying attention.

And that’s it for today. Hope that helps!

Next time, we’ll dive into another popular fallacy—the idea that telling people your goals makes you more accountable.

(Spoiler: It might actually backfire… with some nuances.)

Stay tuned.

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

How do Post-its apologize for being ineffective reminders?

.

.

.

"I'm sorry, I just couldn't make it stick in your mind!"

Good morning! Andy from Back of the Envelope here.

This week, I got sick 🤧, so I didn’t write as much… but remember last week we talked about DeepSeek and its implications? (link)

I warned about using the official version since the data flows through Chinese servers… but since they’ve kindly open-sourced the model, it’s only a matter of time before it’s widely available in the U.S. (Deploying the full model takes a lot of resources)

Well, looks like Perplexity (an AI-powered search engine) is on it, and it didn’t take them long.

From the founder:

(Right now, they only allow five free uses per day for non-paying users though. Hopefully, more free uses will come as other providers follow through.)

What can you do with it?

Pretty much anything that requires some thought. 

For example, you can prompt something like:

• “What are the broader implications of Trump’s recent executive order on [insert policy you care about]?” (Use this list: Federal Register - Executive Orders)
  

• “Tell me the difference between ASCE 7-16 and ASCE 7-22 wind load provisions and how they impact structural engineers. Present the info in a table if possible.”

And there’s more news on the AI front…

OpenAI just dropped “GPT o3-mini”—an even more capable model (free, logged-in users can try it in limited doses).

They also introduced "Deep Research" (not widely available yet) but essentially, you ask a challenging question, and the AI will go search the internet and think through everything for up to 30 minutes, then develop a detailed research report based on its findings 🤯.

AI is moving fast. It really feels like we’re stepping into the future.

Do you feel the same way? And do you have any burning questions about AI or anything you’d love to learn more about?

Hit reply and let me know.

Andy Lin
Back of the Envelope/Structural Engineer HQ

Gooood morning gang! Andy from Back of the Envelope.

By now you've probably heard of the Chinese AI called DeepSeek—it's everywhere in the news.

If you haven't been keeping up with AI trends, you might be wondering: What is it? Why is it a big deal? And how does it impact us, structural engineers?

Don't worry—I got you! Let’s break it down in three minutes (maybe 4).

Here we go!

Quick primer on ChatGPT’s GPT-4o and GPT-o1 models

Before diving into DeepSeek, let’s make sure we’re on the same page with ChatGPT.

Unless you've been living under a rock, you’ve likely heard of ChatGPT—and probably even used it.

But did you know ChatGPT offers different AI models? GPT-4o and GPT-o1 (plus smaller variants like GPT-4o mini and GPT-o1 mini, which aren’t as powerful).

GPT-4o is the model most people are familiar with—it’s available to everyone, free or paid, and has been around for about a year.

GPT-o1, on the other hand, is the premium “thinking” model, only accessible to paid users.

What’s cool about the reasoning model? It actually “thinks” before responding.

For example, the other day I needed to calculate wind loads for rooftop solar panels using ASCE 7-16 (Sections 29.4.3 & 29.4.4). It was a bit tricky, so I grabbed the relevant text/image, pasted it into ChatGPT, and used o1 to walk me through it.

Then, I had it generate a simple web app, which led to me making this (link):

(Of course, it wasn’t as simple as just asking—it required debugging and a lot of back-and-forth. More on that another day.)

Using the less-capable GPT-4o produced a much more error-prone result.

The takeaway? The “thinking” model is getting really smart.

Now, let’s talk “cost” and API.

Cost & API

Most people know ChatGPT as a chatbot—you type, it responds. For users like us, it’s either free or $20/month for more features (https://openai.com/chatgpt/pricing/).

But there’s another side: API access.

This allows developers to integrate ChatGPT into apps, websites, and other backend systems. OpenAI charges based on usage—because running these models requires serious computing power.

Ever seen a website say “Powered by AI”? It’s likely using an API from OpenAI, Anthropic (Claude), or Google (Gemini).

I recently also added AI-powered action planning to 12week.app to test it out.

Toggle the setting, and a button will come up that can generate an action plan using API.

Under the hood, I’m using GPT-4o (cheaper, less capable). Each request costs me about $0.01.

If I used GPT-o1 instead (better results), it’d cost about $0.06 per request.

That might not seem like much, but for companies running millions of queries, it adds up fast.

Deepseek vs. ChatGPT

Now, here’s where things get really interesting.

Last weekend, a Chinese AI company released the R1 model, which rivals OpenAI’s o1 reasoning model—but at a fraction of the cost.

How much cheaper?

28 times cheaper. 😲

So instead of $0.06 per request, it would cost $0.002—cheaper than even GPT-4o.

And here’s the kicker: It’s open-source. Meaning, anyone can build infrastructure around it and create their own ChatGPT equivalent.

Let’s put that into perspective:

Imagine someone releasing a structural analysis tool as powerful as ETABS or RAM Structural System, but for hundreds instead of thousands of dollars per year.

Now imagine it being open-source—allowing you to tweak, expand, and customize it for your projects.

That’s the kind of shift Deepseek is bringing to AI.

What does this mean for us?

In short: More powerful AI tools at lower costs.

AI models that “think” will become more affordable, driving increased competition and innovation.

And with OpenAI’s upcoming o3 model, we’ll see even more powerful tools emerge—helping us develop custom solutions for our workflows.

Imagine being able to build your own structural engineering tools, even with minimal coding skills. That’s where we’re headed.

Exciting times ahead!

Let me know what you think — reply to this email or join the discussion at sehq.pro!

Andy Lin
Back of the Envelope/Structural Engineer HQ

P.S.

One important caveat with DeepSeek: it's based in China, and if you use their web interface, your queries pass through Chinese servers. It also comes with built-in censorship on certain topics.

Not trying to wade too deep into geopolitics here, but personally, I wouldn’t rely on it in its current form for most things.

That said, there are U.S. companies working on adapting the model for local deployment. I’d wait for that.

P.P.S

PROMPT:

Write me a 20 word song about how cool structural engineers are. including some steel design concepts.