Do you know what the difference is between Chat and Cowork? I’ve got a video walkthrough at the bottom too 👇

The function kind of similarly, but they are inherently different from each other and provide different solutions, depending on what you're working on.

Chat

Everyone is probably familiar with chat. This was the first thing that came onto the scene and was popularized. It's the interface that virtually every LLM tool uses today.

You chat back-and-forth asking questions getting answers and getting it to produce things for you.

If you're not careful, these can become extremely lengthy and depending upon the model that you're using, costly.

Cowork

Cowork is related to this and aesthetically looks about the same.

You toggle between chat and cowork by simply select in one of the other. As of a recent update, this works almost identically whether you're in ChatGPT or Claude.

ChatGPT causes it work, however, instead of cowork.

Here's what it looks like in Claude:

Cowork selection in Claude

Here's what it looks like in ChatGPT:

Work selection in ChatGPT

Claude selection makes a little more sense as it's right next to where you're already typing and looking.

Prediction: ChatGPT will update theirs to do exactly the same thing. I think it's a little out of place way up at the top of the screen.

When to Use Work / Cowork

Work (or Cowork) is intended to be used when you need the LLM model to actually do something for you - create/edit files, upload things to your Google Drive, draft and email, etc.

You’ll give it additional permissions so it can do work on your behalf.

Now the interface is the same - it’s a chat window just like we’re used to.

But you can also add a project or a local folder to the chat thread, allowing the tool to work in that projector folder rather than just in the chat window.

I've used coworker quite a bit myself and creating websites as well as code snippets to run on my computer.

This is allowed me to automate some very boring tasks in a much faster manner than going back back-and-forth with the LLM model every single time.

Here's the quick video walk-through I did on this topic. Hope it's helpful for you:

Cheers, Eamonn

I still hit Claude's usage limits every now and then, even doing everything in this list. So this isn't a magic fix, but hopefully it’ll be helpful for you.

I'm on the $20 Pro plan most of the time. I've bumped up to the $100 Max plan twice when I was deep in a coding project and didn't want to wait on the 5-hour reset window, then downgraded a couple weeks later. That's the cost of playing right now, and if $20 a month feels like a lot, the amount of stuff I've automated with it has paid for that many times over.

Here are my top 4 tips followed by the full video with the other 7…

Use Projects, For Real

If you're not using Projects yet, this is the one thing on the list that matters most.

A project is the folder, your chats are the files inside it — and every chat in there reads the project instructions before it starts working. That means separate topics get separate chats instead of one giant sprawling thread that's part Google Sheets question, part business plan, part "wait what was I asking again."

PDFs Are Expensive. Plain Text Isn't.

I attach PDFs to projects all the time; I've got one with franchise documentation I reference constantly. But every time the model has to parse a PDF, it’s very expensive.

If you can, use a markdown file, a Word doc, or copy-pasted text…same information, way cheaper to read. I still use PDFs when I need to, just know what it's costing you.

Match the Model to the Job

Opus is the expensive thinking model. Sonnet is the one I use for almost everything. Haiku is the cheap one. People sleep on it, but it’s great for a quick formula tweak or a single code snippet.

You can also swap models and dial the "effort" level mid-chat, which I do constantly. Just check what's selected before you hit enter. I've fired off a simple prompt on high-effort Opus left over from the day before more than once.

Pay-as-you-go Option

One more thing worth knowing: your usage pools across the web app, desktop, Cowork, and Claude Code — it's all coming out of the same bucket. And if you're regularly bumping the ceiling, there's a pay-as-you-go credits option buried in Settings → Usage that gives you wiggle room without committing to Max.

Here’s the full video walkthrough with the rest of my tips:

Claude built me a webpage. Then I shipped it — for free.

I've been getting a lot of emails from brands and agencies asking about product reviews. So I had Claude build me a media kit page. One HTML file, all my rates and info in one place.

Here’s the site I built.

Here’s the github repository where it lives.

Here’s the video walkthrough:

So I walked through the full publishing cycle: Claude builds it, GitHub stores it, Cloudflare Pages hosts it.

Github and Cloudflare Pages are both free to use for this.

The Setup Is Simpler Than It Sounds

The workflow is basically: build in Claude → push to GitHub → Cloudflare watches GitHub and auto-deploys.

That auto-sync part is the thing. Once you've connected Cloudflare Pages to your repo, every git push to main triggers a new deployment. You're not manually uploading anything. You just commit your changes and the live site updates.

Netlify works the same way. I've been using it for years and it's still great. Cloudflare Pages is just another option that's worth knowing about. Same idea, slightly different dashboard.

If you've never set up GitHub before, that's an extra step I didn't cover in this video.

Let me know if you’d like to see more of the setup/basics stuff.

One HTML File Is All You Need

The site I shipped was one index.html file. That's it…well, that and a headshot I added later.

Claude-generated HTML is almost always clean enough to deploy directly. I didn't touch the code. I just reviewed it in the browser locally, liked what I saw, and pushed it.

The Performance Thing Was Genuinely Surprising

Here's a bonus I wasn't expecting. I ran a Lighthouse report on the live site and got a score of 89 at first. Not terrible, but I expected more since it was just a single HTML file

The culprit was a Google Font import Claude had added. One deleted <link> tag later, I committed the change, pushed it, and re-ran the report for a nice 99/100.

Dropping a web font saved 2 full seconds of load time on mobile. Good thing to know for any Claude-generated page you're putting in front of people.

Have a great one!

Cheers, Eamonn

The most powerful AI model anybody had ever shipped to the public lasted a few days, and then the US government switched it off.

Anthropic (the folks who make Claude, the tool I use every single day) put out a model last week called Claude Fable 5.

It was the public, guardrailed version of something even more capable they'd been testing with security researchers. The early reviews were kind of bonkers. People handed it big, messy coding jobs and watched it power through them.

One company said it found and fixed hundreds of security holes in its own software.

Then last Friday, Anthropic got a letter from the Commerce Department.

The government decided foreign nationals couldn't have access, and the cleanest way to comply was to shut the whole thing off for everyone.

Anthropic says it's a misunderstanding and they're working to turn it back on. Maybe they will. I don't know.

A lot of people spent last week spun out over a tool that, for most of them, was live for about a long weekend.

Meanwhile Opus 4.8, the model I use to run real parts of my business, the one sitting right there in my account for $20 a month, didn't move an inch. It does everything today it did the day before Fable showed up. Which is a lot.

I've been using this stuff a while now and it's crazy how what's available today on the lowest LLM models is light years ahead of what the biggest ones could handle a year ago.

We get desensitized quickly.

Don't worry so much about the "rug-pull" of Fable 5. The "regular" models are still incredibly useful. Haiku, Sonnet and Opus all have their place in my weekly workflow.

Talk to you next time. Have a great one.

Cheers, Eamonn

Learn More of Your Computer

Before you go off and learn the next shiny AI tool, learn your own computer.

I’ve been using the terminal a ton since upgrading my daily computer, and I’ve been using Claude to build little pieces of software to help out with mundane things I’ve got to do every week.

LLMs are making people lazy. I’m not exempt, and I want to encourage both myself and you to learn just a bit more about how the software we’re using actually works.

Here’s my full take in video form:

The terminal isn't scary, it's the foundation

In the video I poke around the command line with some basic operations

If you've never touched it, the terminal is just what we had before the graphical stuff you're used to. Finder, folders, double-clicking, all of that sits on top of this.

It's basic stuff. But once it clicks, you actually save some time and feel like a hacker at the same time.

And once you're comfortable there, you can build small programs that do little pieces of work for you.

Programs vs Prompting

Here's the part I really want you to get. You can absolutely set up an automation inside Claude Code or Cowork, put it on a schedule, and let the AI run your task every week.

That’s an option, it’s just not the best one a lot of the time.

Every single time the AI runs, you're burning tokens. Claude is notorious for cost and for people running out of tokens.

So instead, I used Claude Code to help me build an actual program once: a little QuickBooks integration that takes my weekly coffee orders, cleans the data, and uploads them as invoices.

Now I just type coffee import, hit enter a few times, and eight invoices land in QuickBooks.

It Does Take Work

I'm not gonna lie, this was a lot of iteration.

The program itself was the easy bit. The hard part was the QuickBooks integration: the authentication, the reference data, getting it all to sync up.

I had Claude walk me through all those pieces before it worked.

And "error proof" only goes so far. Feed it bad data and you'll get bad invoices. It's only as smart as what you give it.

But this is the way I think AI actually helps small businesses. Not by running the AI on a loop forever — by using it to build the little tools once, then getting out of the way. Learn the computer first. The rest gets a lot easier.

Hope that's helpful. If you've got a repetitive thing in your business that a small program could kill, reply and tell me — happy to go deeper on how I built this one.

Cheers, Eamonn

Yes, you should still know how Apps Script, and programming in general, works…but this is how you need to be using LLMs to help you leverage these tools:

When You Shared Too Much, Too Loosely

I had years of messily shared Google Sheets and Docs.

I’d initially shared a ton of them directly with people instead of through folders.

And I needed to remove some of those people all at once…

Doing it manually wasn't really an option so I asked ChatGPT to write me an Apps Script, and one prompt later, I had something that worked.

The full demo, including the ChatGPT conversation and the script running live, is in the video.

One Prompt, One Script

The premise was simple: build a script that takes a folder (or a list of folders), finds every file inside, and removes a specific email address from all of them.

It worked out of the box with just 83 lines of code.

But before I ran it on anything real, I asked a couple of follow-up questions. What happens if that email isn't shared on a particular file? And what about everyone else who has access? I needed those people untouched.

It had already covered these situations.

The script uses file.removeViewer and file.removeEditor so it catches both access types and it only targets the email address you tell it to.

A Few Things Worth Knowing Before You Run It

The folder ID is just the string in the URL after /folders/. That's what you paste into the script. If you need to remove someone from multiple folders, you add them as a comma-separated list.

Got multiple people to remove? Same deal — instead of a single email constant, you just make it a list. The AI figured that out when I asked, and updated the script accordingly.

One thing I'd recommend: keep the logger messages in.

When the script runs, they print out exactly what it's doing: which files it processed, which ones it skipped, who got removed.

If something looks off, you copy that output back into ChatGPT and ask it what went wrong. I've had to do that on almost every Apps Script project; that’s just part of coding.

It Was a Little Glitchy, but It Worked

Fair warning: when I tested it the first time, the sharing UI still showed the old email address.

It looked like it hadn't worked.

But it turns out it was probably a browser cache/Google Workspace glitch rather than something wrong with the code - a browser refresh cleared it up after at first I thought something hadn’t worked.

So if you run it and something looks stuck, check the logs first, then refresh before you assume it failed.

This is the kind of thing that would've taken me forever to do manually; now it's just a script I run when I need it.

What LLM do I Recommend?

ChatGPT, Claude, Gemini…any of them can write this.

Coding for LLMs is easy; they are all excellent at it - especially small projects like this.

If you've got a task that's mind-numbingly repetitive, there's probably a script for it. You just have to ask.

Use whatever tool you’re most comfortable with and get good at it before mixing up the tooling for different uses.

Cheers, Eamonn

Using Claude and GitHub from iPhone

I'm very impressed with how Claude Code integrates with GitHub on my iPhone to make pull requests. Review code and merge changes all without touching my desktop.

Now I don't think this is going to replace desktop computing by any means.

However, I've had a real good time with very minimal issues for the past week using the Claude Code <> GitHub integration.

How it Works

Download the Claude iPhone app.

Download the GitHub iPhone app.

Open the Claude iPhone app and select Code from the sidebar menu. This opens up a version of claude code accessible on your phone.

Click the little chat button and the top right to create a new session.

Sessions in Claude Code operate much like chat message threads if you're familiar with the regular chat screens for Claude, Gemini or ChatGPT.

The first time you use this, you'll need to connect your Claude and GitHub accounts. Claude will install its extension on your GitHub account, which enables it to read and write according to whatever you give it access to.

Once you've done this, you'll now see a little plus icon above the message bar. Click this, and you can select whatever repository your referencing in this session.

Now it's just like any other chat. Ask it to do or fix whatever you need to.

In my example above, I asked it to make some optimizations to a simple HTML webpage product and create a PR request for that.

It should create a PR by default. This allows you to review the changes over on the GitHub app before you merge them into the live codebase.

For something like my project with very low stakes, I didn't even leave the apps. I reviewed everything quickly, tested the preview link that was created via my Netlify connection, and made the changes go live.

Watch the short video above to see it in action.

I was very impressed with this.

I can do all this stuff from a command line. But I don’t live in that environment every day. This really minimizes the friction to create proper code updates even for a peasant coder like me.

What do you think?

Did I just go full nerd on you or is this something you’re interested in using?

Let me know if you have specific use cases you'd like to see me discuss or make a tutorial about.

I read every email response.

Thanks so much for your time and hope you have a great week!

Cheers, Eamonn

Hey {{first_name|there}},

Opus 4.7 is out — here's what actually matters

Anthropic dropped Claude Opus 4.7 today, and if you're running anything serious in Claude: coworker stuff, long projects, document-heavy work…this is the one you'll want. Same price as 4.6 ($5 in / $25 out per million tokens), but it WILL consume more tokens.

It’s available across all the Claude products and the API.

I pulled up the announcement, watched it do its thing, and made a video walking through the parts that caught my eye.

The memory piece is the real upgrade

I've got one Coworker project that has ballooned way out of proportion. Directories inside directories, chats referencing other chats, and I keep having to nudge Claude: "hey, go review the whole project before we keep going." It usually does fine, but I’ve had to ask for things that I wish it’d just check (maybe I just don’t trust it enough?)

Opus 4.7 is supposed to be better at file-system-based memory across long, multi-session work. Which is exactly the thing I've been running into. Less upfront context from me, more picking up where we left off. If that holds up in real use, it's the upgrade I care about most. Not the benchmark bumps.

xhigh effort and an ultra-review command

There's a new "xhigh" effort level sitting between high and max. It's another slider on the reasoning-vs-tokens tradeoff. In Claude Code the default is already bumped up to xhigh, so out of the gate you're burning more than before unless you dial it down manually.

There's also a new /ultra-review slash command in Claude Code that does a careful pass on your changes looking for bugs and design issues. As more of our code gets written by the model, having the model review it like a real reviewer would is going to matter more, not less.

Don’t sleep on Haiku

Opus eats through usage limits faster than anything else in the lineup. If you're just asking quick questions, use Haiku. For everyday stuff, Sonnet 4.6 is still my daily driver. Opus 4.7 is for the heavy-lift office work, the sprawling projects, the coding that actually needs the horsepower.

Also worth noting: Anthropic is openly saying this isn't their best model. There's a thing called Mythos Preview sitting above it that they're holding back for cybersecurity reasons. Good marketing, whether you buy the rationale or not.

Try it, but pick your moments

If you're on Pro or Max, Opus 4.7 is already there ready and waiting.

Talk soon,
Eamonn
AI strategy and automations

Free Week of Claude Cowork

If you want to check out Claude Cowork, they gave me 3 free guest passes for you to check out for a week.

If you’re on a paid plan, you have access to this already, so just download the desktop app and start checking it out.

If you aren’t bought into Claude yet, here’s your chance to check it out.

Problem is, I only have 3 codes to giveaway…

And not all of you are going to get this email as soon as I send it…

Here's my plan.

If you need a code, click yes on the poll below.

On Friday, April 3rd, I'll have a drawing from all of the entries and pick three of them at random to get the code.

New From Claude

Several pieces of notable news and releases from Claude this week.

In no particular order, here are my top 3.

And I'd also like to know what you'd like me to cover in depth first.

1. Projects in Claude Cowork - We can now have dedicated projects with their own memory in Claude Cowork.

2. Claude Code Architect - Anthropic has released its very own course and certification for Claude code architects.

3. Claude Dispatch - You can now text Claude Cowork from your phone and it'll do stuff on your computer remotely. Here’s a quick short I did when it was announced.

Just Released

My latest video on Claude Cowork goes through setting up and using it for the first time.

It took me longer than it should have to try out Cowork, as it allows me to use Claude to do stuff on my local computer.

Check it out here if you missed it 👇

A Real Claude Cowork Project

I'm editing and releasing a video on Claude Cowork project that I set up last week. It should be available within a couple days.

Make sure you're subscribed to the YouTube channel to grab it when it's released.

Now tell me below which topic you want me to go into more detail on first. 👇️ 

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

The Free Voice-to-Text App I'm Using to Write This Newsletter

Project instructions are fine. But if you've ever copy-pasted the same prompt into three different Claude projects, you may be ready to build a skill instead.

Skills are different. A skill teaches Claude how to do a thing, not what to do in a specific project.

You write it once, upload it, and Claude can reference it across any project or chat. No more copy-pasting. No more hoping Claude picks up on context.

Why I Finally Built One

Most of these articles began as a video where I walk through how to do something, like build skills.

The article version takes a lot less time to publish when I can have Claude write a draft from the video transcript that I just spent a lot of time working on.

So to make this skill, I downloaded transcripts from a handful of my YouTube videos, threw them into Claude, and asked it to write a voice skill based on how I actually talk.

It came up with stuff I wouldn't have thought to include and in a lot more detail than I would have written off the cuff.

The output wasn't perfect. I had to edit it. But it gave me a real starting point way faster than writing from scratch.

What Actually Goes into a Skill

It's a markdown file: just text with some formatting.

The part Claude sometimes gets wrong is the front matter at the top. You need three dashes, then name: and description: in lowercase, then three more dashes to close it out. That part has to be there or it'll throw an error when you upload it.

---
name: eamonn-cottrell-voice-skill
description: my custom voice skill
---

Once that's sorted, you make a folder, put the .md file in it and zip the folder.

Then you go to Claude and upload it under Customize → Skills.

Claude will show you the parsed title and description so you know it read correctly.

And don't worry if you mess something up, it will not work properly and it will give you an error. That happened to me a few times. That's how I figured out the syntax mistakes.

From there, you can reference it explicitly in a prompt or just mention it in your project instructions so Claude knows to check it automatically.

One Small Thing Worth Knowing

You don't have to tell Claude to use a skill every time. If your project instructions say something like "reference the voice skill when writing articles," it'll go grab it on its own.

One more new thing:

Thanks for reading!

Also, I'm contemplating a new podcast where I would have a conversation with, potentially…you.

I'm looking for founders and CEOs of AI and tech companies who need an audience for their business and/or product. 👇

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

The Free Voice-to-Text App I'm Using to Write This Newsletter

Most voice-to-text tools cost money.

Handy is the exception, and I’m using it right now…

Why Bother with Voice-to-Text?

If you're still typing everything from scratch, you're leaving speed on the table.

Speaking is roughly 3x faster than typing for most people, and for first drafts, rough notes, or capturing ideas on the fly, it's been worth it for me to start using.

Most of the good voice-to-text apps charge a monthly fee. And the free built-ins from Windows (Win+H) or Mac (F6) work, but they're clunky and not particularly customizable.

Handy sits in the middle: genuinely good, completely free.

What Makes Handy Worth Installing

It gets out of your way. From first launch it to actually recording took me about two seconds. No account or onboarding, which I appreciate.

The keyboard shortcut is sweet. I set mine to Ctrl+Shift+Alt+Spacebar. Hold it down, start talking, release to stop. That's it.

Audible feedback you can toggle. Handy plays a chime when it starts and stops recording. I leave it on. There have been a few times where I didn't quite hit the keyboard shortcut correctly and it did not record anything.

Model flexibility. Since I only speak English, I'm running the Parakeet V2 model which is fast and accurate. But Handy lets you swap between several AI transcription models, all downloadable for free. If you need multilingual support or want to experiment, it's easy to switch.

Should You Bother If You Already Use the Built-In Dictation?

It's worth it to try them all out.

The OS built-ins are fine for a quick note. But Handy gives you a custom shortcut that works across any app, better model options, and a cleaner overall experience. It's available on both Windows and Mac.

If you're already on a Mac, the choice is a little more complicated because I did find the built-in Mac voice to text to be quite good.

One more new thing:

Thanks for reading!

Also, I'm contemplating a new podcast where I would have a conversation with, potentially…you.

I'm looking for founders and CEOs of AI and tech companies who need an audience for their business and/or product. 👇

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

MORE AI CONTENT COMING SOON

Hey y’all,

I’ve been throwing more of my time into AI tooling this year, and I will be sharing my progress with you soon. Nothing has increased my understanding of and productivity with spreadsheets, automations, and software in general quite like the AI tooling available right now.

I’m using all the frontier models, but Claude is going to be my primary focus.

I’ve started making more content already on the YouTube channel - check it out here if you aren’t subscribed yet.

It’s the best at coding and technical tasks which is where I think the most gains are going to be had in this space.

My focus is on implementing AI in businesses.

Hit reply and let me know how your business is, isn’t, or should be using AI today.

The Google Sheets Function That Replaces Your Running Total Column

I used SUM to solve this problem for years until I found out about SCAN.

Here’s the official definition of SCAN:

Your Balance Sheet Shouldn't Be This Fragile

I use a running total column in my most popular spreadsheet, which is a personal finance app.

You need a running balance that updates automatically as new rows appear. I've always just had a balance column with a formula that I copy all the way down.

It works, but it's not very elegant. And there's a better way: to write one formula in the top cell. A little help from ARRAYFORMULA and you don’t have to copy anything else down.

What SCAN Actually Does

SCAN walks through an array and accumulates a result, step by step, like a running total on autopilot. It takes three arguments: a starting value, an array to process, and a LAMBDA that defines what happens at each step.

Don't be afraid of LAMBDA. I was for a long time, but it's very powerful and easy to use once you wrap your head around it.

It's basically just a way to define a formula inside of a formula.

Here's the formula that replaces an entire column of manual calculations:

=ARRAYFORMULA(
  IF(Table1[amount],
    start + 
    SCAN(G1,
      IF(Table1[category]="income", Table1[amount], -Table1[amount]),
      LAMBDA(a, x, a+x)),
  ""))

Here’s how it works 👇

The Inner Logic

IF(Table1[amount], ..., "")
The outer wrapper keeps empty rows clean. No amount? No balance displayed.

IF(Table1[category]="income", Table1[amount], -Table1[amount]) 
This is your transformation layer. Income stays positive. Everything else flips negative. You're converting a raw transaction log into a signed array before SCAN ever touches it.

SCAN(G1, ..., LAMBDA(a, x, a+x)) 
G1 is your starting balance (maybe zero, maybe last month's closing balance). SCAN then walks through each signed amount, adding it to the accumulator. The a is the running total so far. The x is the current value.

start + SCAN(...)
Offsets everything by your starting value, so your running balance begins exactly where you need it.

Why This Beats the Drag-Down Approach

One formula, infinite rows. Because it's wrapped in ARRAYFORMULA, it spills results down automatically. New transaction? The balance updates. No dragging required.

Nothing to break. Delete a row and the formula recalculates. Insert a row and it adapts. There's no chain of cell references waiting to snap.

Readable logic. I've been switching to ARRAYFORMULA in all of my spreadsheets lately so that I can have single points of failure instead of tons of different formulas littered throughout each sheet.

Where Else SCAN Shines

Running balances are the obvious use case, but SCAN is quietly powerful anywhere you need cumulative logic:

• Inventory tracking: Start with stock on hand, subtract orders, add restocks. One formula.

• Goal progress: Accumulate daily outputs against a monthly target.

• Cumulative revenue: Show month-to-date or year-to-date totals that update in real time.

• Sequential numbering with conditions: Increment a counter only when a condition is met.

The pattern is always the same: starting value → array → LAMBDA that defines the accumulation rule.

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

Fixing Google Sheets’ Missing Auto‑Fit Shortcut

There’s one Excel shortcut I’ve missed for years: auto‑fit column width (Alt + H,O,I).

In Google Sheets, you can double‑click the column edge. You can right‑click and resize. You can even convert a range to a Table Ctrl + Alt + T) and let it auto‑fit for you.

But there’s no fast keyboard shortcut.

So, I built one.

The Idea

Google Sheets already has the method: autoResizeColumn().

All we’re doing is wrapping it in a tiny Apps Script and assigning it to a macro shortcut.

Two versions:

• Resize current column (based on active cell)

• Resize selected range (auto‑fit multiple columns at once)

That’s it. And, honestly, you don’t need the current column version because the range version works for all use cases.

The multi‑column version uses SpreadsheetApp.flush() to make sure everything actually applies when resizing in a loop — a small but important detail when Apps Script acts inconsistent.

Keyboard Shortcut Setup (High Level)

You can’t bind true custom shortcuts in Sheets, but you can assign macros to:

Ctrl + Alt + Shift + [Number]
(Mac automatically converts to Command + Option + Shift + [Number])

Inside the appsscript.json file, you define:

• A menuName (what it’s called in the Macros window)

• A functionName (must match your script)

• A defaultShortcut

Refresh the sheet, and you’ve got an instant auto‑fit shortcut.

First run will ask for permissions. Approve once and you’re done.

Why This Is Worth It

This is just another area where removing a little bit of friction makes a big difference for me. I don’t like clicking things to execute actions that I need to make often.

And it’s a great entry‑level Apps Script project if you’ve been meaning to get your hands dirty.

Watch the video above where I show you everything you need to do to get this working in your sheet.

If you’d rather skip setup, I packaged the sheet for you so it’s ready to go:

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

• Anthropic’s support page for Claude in Excel.

• Microsoft’s Marketplace listing for the Claude in Excel add-on.

The Tech newsletter for Engineers who want to stay ahead

Tech moves fast, but you're still playing catch-up?

That's exactly why 100K+ engineers working at Google, Meta, and Apple read The Code twice a week.

Here's what you get:

• Curated tech news that shapes your career - Filtered from thousands of sources so you know what's coming 6 months early.

• Practical resources you can use immediately - Real tutorials and tools that solve actual engineering problems.

• Research papers and insights decoded - We break down complex tech so you understand what matters.

All delivered twice a week in just 2 short emails.

Join 100K+ engineers

Claude in Excel: Promising, Polished… and Still Limited

I’ve been testing Claude inside Excel for a few days now, and my reaction lands squarely in the middle: excited by the direction, underwhelmed by the execution.

On paper, this integration checks a lot of boxes. Claude lives directly inside Excel as an add-in, opens a familiar chat sidebar, and can read and act on your workbook. You can ask it to analyze data, clean things up, build summaries, and even generate pivots or charts. If you’ve ever bounced between a spreadsheet and an LLM tab, the appeal is obvious.

But once you start using it for real spreadsheet work, the cracks show.

What Claude in Excel Does Well

The setup is straightforward. Install the add-in from the Microsoft marketplace, open it from the Excel toolbar, and you’re off. Claude immediately understands your sheet context, which is genuinely useful.

A few things it does well:

• Generating dummy data quickly for demos or testing

• Spotting formatting issues, like date columns that came in as serial values

• Summarizing datasets and surfacing high-level insights without much prompting

For beginners or casual Excel users, this feels powerful. You can ask questions in plain English and get something usable without knowing formulas, pivots, or charts.

Where It Falls Short (Fast)

Once you move past surface-level tasks, it’s just not perfect (see my video demo above for examples).

A few pain points stood out immediately:

• Hardcoded results instead of formulas
  Claude often outputs static numbers where formulas should live. That defeats the entire purpose of working in Excel.

• Outdated formula choices
  I saw INDEX/MATCH and even VLOOKUP-style logic where modern functions like XLOOKUP, SUMIFS, or GROUP BY would have been cleaner and more flexible.

• Limited sheet control
  Simple actions like merging across cells or applying structural formatting aren’t fully supported. In some cases, Claude just explains what you should do instead of doing it.

• Heavy rate limiting
  This is the biggest issue. After only a handful of small prompts, some of them simple yes/no follow-ups, I hit Claude’s usage cap. That makes sustained spreadsheet work frustrating and unreliable.

The Bigger Takeaway

Claude in Excel is not a replacement for spreadsheet knowledge.

If you don’t already understand pivots, lookups, summaries, and basic modeling, you won’t know when Claude’s output is wrong, inefficient, or brittle. And if you do know those things, you’ll often find it faster to just build them yourself.

That said, I do like the direction. Embedded AI inside tools like Excel is clearly where things are headed.

There are interesting efforts underway to reduce token usage and improve efficiency in Claude-based workflows, and if those pan out, this could become much more compelling. But today? For my use case, it’s a nice sidebar; not a must-have.

My recommendation for now is:
learn spreadsheets first, then use AI as an assistant, not a crutch.

If you’re a beginner, check out my free 7-day course which covers the basics.

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

Video Walkthrough

WALKTHROUGH
How to build it

For a long time, I’ve been bad at remembering to send birthday wishes to my friends. Sure, I remember some. Sure, I’ve got some on my Google Calendar.

But for those of you, like me, who aren’t relying on Facebook to spam birthday celebrations to your cell phone everyday, this little Google Sheet may be exactly what you want.

Sometimes, simple is best. Here’s what we need:

1. A table in a Google Sheet that I can add names and dates to

2. Every day, the sheet checks for an anniversary in the morning

3. If there is one, I get an email reminding me to text them

4. I’ve also got a third column where I can specify what the date is (birthday, anniversary, etc)

That’s it.

Here’s what you need to do:

Open a new Google Sheet (sheets.new)

Rename the sheet tab, Birthdays.

Add the column headers and text

Convert this to a table (not 100% necessary, but it’s good practice to do. CTRL + ALT + T or CMD + OPTION + T).

Put your email in another cell somewhere in the spreadsheet and create a named range named NOTIFY_EMAIL. This will pull your email address into the Apps Script we’re about to create.

Open up Apps Script (Extension → Apps Script)

Paste the code (below ⬇️) and save.

Run the function setup() once so it’ll set up the trigger to check every morning.

I like to add a date to test that the email part is working. Add a row with someone’s name and today’s month and day. You can pick a random year. Now run the reminderToday() function and make sure you get an email.

Call me crazy, but getting an email from myself like this has already worked better than throwing an all day calendar event in my Google Calendar.

This is already working for me. My friend Mike’s birthday popped up a couple days ago and I immediately shot him a text.

I’ve always had the best of intentions with birthdays and anniversaries, but sometimes I need a little help from technology.

Careful!

The version of the script that i’ve written expects ONLY one table of information to be on the Birthday sheet. If you’re putting more information in the workbook, do it on a new sheet tab.

Too Complicated? Get the Sheet ⬇️

You can build this for free by following the steps above, but if you need the pre-built custom sheet that’s ready to roll… Grab it below.

const SHEET = 'Birthdays';
function getNotifyEmail() {
  return SpreadsheetApp.getActive()
    .getRangeByName('NOTIFY_EMAIL')
    .getValue();
}


function onOpen() {
  SpreadsheetApp.getUi()
    .createMenu('Setup')
    .addItem('Run setup', 'setup')
    .addToUi();
}

function reminderToday() {
  const sh = SpreadsheetApp.getActive().getSheetByName(SHEET);
  const rows = sh.getRange(2, 1, sh.getLastRow() - 1, 3).getValues();
  const tz = Session.getScriptTimeZone();

  const today = new Date();
  const mmdd = Utilities.formatDate(today, tz, 'MM-dd');
  const year = Number(Utilities.formatDate(today, tz, 'yyyy'));

  const matches = rows
    .filter(([name, d]) =>
      name && d instanceof Date &&
      Utilities.formatDate(d, tz, 'MM-dd') === mmdd
    )
    .map(([name, d, reason]) => ({
      name,
      age: year - d.getFullYear(),
      reason: reason || 'anniversary'
    }));

  if (!matches.length) return;

  matches.forEach(m =>
    GmailApp.sendEmail(
      getNotifyEmail(),
      `Reminder today about ${m.name}`,
      `Reminder: ${m.name}'s ${m.age} ${m.reason}`
    )

  );
}

// run once to set up daily trigger at 8am
function setup() {
  ScriptApp.newTrigger('reminderToday')
    .timeBased()
    .everyDays(1)
    .atHour(8)
    .create();
}

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

WALKTHROUGH
When and How to Use ARRAYFORMULA

I'm not really sure why I didn't do this sooner.

I've been using a custom Google Sheet for work to manage the processing of several invoices at one time into Quickbooks.

The details aren't super important for the sake of this tutorial, but the way that I was doing it is…

One of my columns took information from two of my other columns and applied an if statement to it so I could have it automatically enter a specific value based on the info already in those cells.

Here's what I'm talking about:

Column F automatically pulls the store name from another list of items because column D has a “Q” in it.

The reason for this is not important and it's very specific to this particular use case but the application to get a value automatically entered is important.

Here's the code that I had in column F before today:

=IFS(ISNUMBER(SEARCH("C",D2)),"Clinton", ISNUMBER(SEARCH("F",D2)),"Fondren", ISNUMBER(SEARCH("Q",D2)),"Quarter", ISBLANK(D2),"") 

This worked perfectly fine however it was not a way to do it because I had to copy it all the way down for the entire column F.

the alternative is to wrap this up in an ARRAYFORMULA.

Instead of copying it down however many rows an ARRAYFORMULA in one cell allows me to simply write one formula one time and use it for the entire column.

This is good for a few reasons:

1. Performance: One formula processing an array is significantly faster than 41 or 401 individual formulas. Google Sheets calculates the entire range in one operation instead of looping through each cell.

2. Maintenance: If I ever need to modify the logic all I have to do is change the one formula in the 1 cell at the very top and we're good to go.

3. Data Integrity: I don't usually run into errors when I've copied down formulas however it has happened and you always run the risk of it happening if you are copying down a ton of formulas to every cell in that column. One of them could get deleted or modified on accident.

4. Scaleability: If I add new data down to the bottom or if I extend my sheet and add new rows with ARRAYFORMULA I can just account for that and have built into the formula itself an extended range i.e. D2:D.

5. Cleaner Output: Now when I cursor over any of the cells except for that very first one in column F it shows simple text output instead of a formula.

=ARRAYFORMULA(IFS(ISNUMBER(SEARCH("C",D2:D41)),"Clinton", ISNUMBER(SEARCH("F",D2:D41)),"Fondren", ISNUMBER(SEARCH("Q",D2:D41)),"Quarter", ISBLANK(D2:D41),""))

Enjoying these newsletters?

Happy Spreadsheeting!

Enjoyed this issue?
Subscribe below or Leave a testimonial

<iframe src="https://embeds.beehiiv.com/eab8502a-f9ab-4d3f-abc8-30b1abb520a9?slim=true" data-test-id="beehiiv-embed" height="52" frameborder="0" style="margin: 0; border-radius: 0px !important; background-color: transparent;"></iframe>

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.

TOOLS
Make life easier

• Coefficient - Live spreadsheet connections to 60+ business systems

• Lido - Automate your spreadsheets; accurately extract date from PDFs

• TransactionPro - quickly import, export or delete data in Quickbooks

• beehiiv - my choice for a newsletter operating system

• Carrd - free one-page website builder

• Transistor - my favorite podcast host

Do you know what this is?

This is a toggle that lets you hide and unhide rows or columns in Google Sheets.

You can do the same thing in Excel, and it’s very handy, especially when you have a table that’s gotten a bit bloated, but you’re too committed to the data to delete any of it.

There are two ways to hide:

1. Right-click and select hide row/column

2. Right-click, select view more cell actions, select Group column/row

If you’re in Excel, the grouping is in a different place, and you’ve got a sweet keyboard shortcut you can use:

• Find the manual option in the ribbon by clicking Data, then Group or Ungroup from the Outline section

• Or you can be more awesome and just press SHIFT + ALT + → to group and SHIFT + ALT + ← to ungroup.

Enjoying these newsletters?

Enjoying these newsletters? 👇

I ran across an issue a few weeks ago at work while building a custom Google Sheet for a new vendor project.

Part of the sheet was a table of item details that would grow over time.

The nature of what I was doing required that sometimes duplicate item numbers might get appended to the table.

I wanted a quick visual flag if this ever happened, and I landed on conditional formatting.

What the Formula Did

I needed the entire row to get highlighted red if there were ever duplicate item numbers in column A.

In case you didn’t know, there are a host of different rules you can set up for conditional formatting.

Many of the built in rules suffice for simple use cases.

But you can also select “Custom formula is” down at the bottom of the list if you need to homebrew a solution.

That’s what we’re doing here.

And the solution is pretty simple: COUNTIF()

We simply want to count the number of occurrences of the item numbers in column A and make sure it’s always only 1.

=COUNTIF($A$2:A2, A2) > 1

And just like that, we’re done.

By using an absolute at the start of the range $A$2, we lock the start of the equation and then look all the way through whatever row is being evaluated (this goes through all the rows one at a time because we’ve instructed the conditional formatting to go from A2:A500).

It will highlight the duplicate row if there is one because the COUNTIF() formula will return 2 instead of 1.

If we wanted to highlight all occurrences of the duplicated row, we could change the syntax just a hair and write this instead:

=COUNTIF($A$2:A, A2) > 1

Now it’s looking at the entire column every time it evaluates the conditional formatting.

Enjoyed this issue?
Subscribe below or Leave a testimonial

Cheers, Eamonn
Connect on LinkedIn, YouTube & Twitter.