iiinova — where your digital self keeps evolving

Tom Yedwab @tomyedwab.com · 15/09/2026

Both companies issue press releases predicting soda drinks will be responsible for worldwide plague by the mid-nineties and imploring the government to strictly regulate reformulations of RC Cola and to restrict imports of Schweppes.

010

Tom Yedwab @tomyedwab.com · 15/09/2026

It is 1985, the height of the soda/pop armaments race. Coca-Cola reveals some batches of New Coke have been found to contain rat poison. Not to be outdone, Pepsi introduces Pepsi Anthrax, which is immediately banned from sale...

100

Tom Yedwab @tomyedwab.com · 10/09/2026

You know how you have a hole in your vision where the optic fibers go through your retina that you aren't aware of? A small thing can hide there, but not for very long. Maybe we just perceive our moral systems as internally consistent and complete, and that is good enough 99% of the time.

020

Tom Yedwab @tomyedwab.com · 24/07/2026

This reads like someone complaining that Boeing doesn't make smaller jets that seat five, don't fly and can drive on the highway. Global scale social media doesn't work at all like P2P sharing, and trying to combine them into one protocol/app would make the UX much worse.

010

Tom Yedwab @tomyedwab.com · 16/07/2026

Changing minds is not the intent? As leader of a big org he can't remain neutral. My tl;dr: * we will support those who want to use AI in their work * we will respect those who cannot or do not want to use AI in their work * those who just want to argue about what others are doing can leave

001

Tom Yedwab @tomyedwab.com · 14/06/2026

Also got the ol' blog set up with standard.site. This requires learning a lot of different concepts to get off the ground, but seems pretty neat once it's all set up!

standard.site

Standard.site - One schema. Every platform.

Standard.site provides shared lexicons for long-form publishing on AT Protocol. Making content easier to discover, index, and move across the ATmosphere.

000

Tom Yedwab @tomyedwab.com · 14/06/2026

This advice is intended for AI coding enthusiasts but like a lot of these best practices, it stems from pre-existing ideas: AI makes it too easy to skip the most important decisions in a project, the ones that happen right at the beginning. www.arguingwithalgorithms.com/posts/lay-th...

arguingwithalgorithms.com

Lay the cornerstone yourself

As starting projects becomes easier than ever, it's never been more important to pay careful attention to the foundations and not let the AI make all the decisions for you.

110

Tom Yedwab @tomyedwab.com · 20/05/2026

I don't know what I expected, but neither ChatGPT nor Gemini were able to generate a 3D printable OpenSCAD model of a simple VESA mount. The lack of spatial reasoning on display, both for the object itself and the constraints of the printer, is comical.

010

Reposted by Tom Yedwab

Camille Fournier @skamille.themanagerswrath.com · 14/05/2026

Even though I've made this joke myself I actually really hate the "all developers have to be engineering managers now with agents" No, they all need to be tech leads, not ems. Unless you're having to deal with the agents' interpersonal issues and the PM distracting them and getting them promoted and

414422

Tom Yedwab @tomyedwab.com · 01/05/2026

If you find the social media/sleep health studies worrying now, just wait until you read the AI/sleep studies coming in 10 years. Signed, Mr. No time for sleep my 5 hour window just reset

000

Tom Yedwab @tomyedwab.com · 30/04/2026

May be rational in the short run, but the proportion of developers using zero AI tools is rapidly dwindling. You can absolutely set high expectations for contributions without being absolutist and potentially driving away the next generation of contributors.

230

Tom Yedwab @tomyedwab.com · 30/04/2026

This argument is going to age poorly IMO. Are maintainers pairing with newbies to answer questions and show how to be successful in the project? Are the docs beautifully written for beginners in multiple languages? Moreover, if its the person who matters, why are we judging the tools?

560

Tom Yedwab @tomyedwab.com · 08/04/2026

When I read sensational LLM-breaks-out-of-sandbox stories I think of pet hamsters I once had who would eventually break out of any cage through pure, dumb persistence and luck. I was unimpressed then, and I'm unimpressed now.

000

Tom Yedwab @tomyedwab.com · 29/03/2026

At this point if you're writing a compiler for your new language, you may as well offer the option to pipe your errors and code context to an LLM agent. In the past I had too little patience to learn Rust, but it's quite pleasant now that I have an error-to-English translator.

020

Tom Yedwab @tomyedwab.com · 29/03/2026

I believe it's a more complex problem than we initially expected. Wikipedia has so much knowledge, but many articles are too technical for lay readers. When your audience is the entire world, it's a big challenge to meet people where they are re: existing knowledge, political beliefs, and motives.

000

Tom Yedwab @tomyedwab.com · 20/02/2026

Why learn to paint when you can use Photoshop and pay a subscription fee for the rest of your life.

100

Tom Yedwab @tomyedwab.com · 17/02/2026

This week I wrote something a bit more personal, about my history with anxiety disorder and why it's important to push back on the artificial urgency of AI. www.arguingwithalgorithms.com/posts/the-ai...

arguingwithalgorithms.com

The AI Panic Attack

Sharing what I've learned from mental health challenges and the moment we are in today.

010

Reposted by Tom Yedwab

Scott Hanselman 🌮 @scott.hanselman.com · 13/02/2026

You vibed an app, for an audience of one. Add it to Tiny Tool Town. But You know what Tiny Tool Town REALLY needed? *TIKTOK MODE* Why not mindlessly scroll awesome apps rather than videos? Maybe one will help you or someone else! go hit tinytooltown.com and tap "SHORTS" for Shorts/Reels Mode

tinytooltown.com

Tiny Tool Town 🏘️

A delightful showcase for free, fun & open source tiny tools. Stupid-delightful software made with love.

48814

Tom Yedwab @tomyedwab.com · 02/02/2026

But also: being able to delegate tasks while setting clear requirements is *absolutely* necessary for a Senior+ engineer. It's how you scale yourself beyond the output of one person and leverage your team.

000

Tom Yedwab @tomyedwab.com · 02/02/2026

My worst days as a manager were spent stressing about a difficult performance conversation. My best days were celebrating a team win, or covering for someone who needed emergency time off. Thankfully, I don't have to worry about writing an annual review for Claude.

100

Tom Yedwab @tomyedwab.com · 02/02/2026

Sound analysis here: addyo.substack.com/p/the-80-pro.... The only quibble I have is with the oft-repeated idea that delegating to agents feels like being a manager. Respectfully, it does not.

addyo.substack.com

The 80% Problem in Agentic Coding

Managing comprehension debt when leaning on AI to code

100

Tom Yedwab @tomyedwab.com · 30/01/2026

In the time between the weekend when I decided to give it a whirl, and finally getting it running today, Clawdbot got renamed *twice*. I hope #openclaw has better longevity!

000

Tom Yedwab @tomyedwab.com · 14/01/2026

Sharing some background on my values. When I feel a lot of uncertainty about the future of our profession and our industry, it helps to go back to why I chose this career in the first place. www.arguingwithalgorithms.com/posts/my-cod...

arguingwithalgorithms.com

My coding philosophy

Sharing my philosophy of software development and my hopes for the future of our craft.

000

Reposted by Tom Yedwab

Gergely Orosz @gergely.pragmaticengineer.com · 14/01/2026

The Pragmatic Summit: A one-of-a-kind event where you'll hear from (and get to meet) standout folks who have been on The Pragmatic Engineer Podcast, teams who were in The Pragmatic Engineer deepdives - all of them building cutting-edge software, or researching how best to do so.

2253

Tom Yedwab @tomyedwab.com · 19/12/2025

I'm not sure what examples you have in mind but what I am sure of is you get 2, maybe 3 false positives before no one ever looks at the check engine light again. That is the problem with vague alerts, they very easily become background noise.

110

Tom Yedwab @tomyedwab.com · 04/12/2025

One big benefit is that the prompt and work progress is on disk rather than locked in session context and if I hit my usage cap I can switch easily to another agent. I haven't done much with screenshots TBH.

000

Tom Yedwab @tomyedwab.com · 04/12/2025

I don't use detailed prompts with any of my agents. I used to write spec files and prompt Claude Code to implement them, but now I drop a PLAN.md file into the repo root with the desired task info and have the agent update it as it goes. That is all handled for me by the global instructions.

plan.md

PLAN.MD

At Plan.md, we provide expert business consulting, funding advice, and tailored support programs to help your business thrive and secure grants.

110

Reposted by Tom Yedwab

Eric Bailey @ericwbailey.website · 24/11/2025

“These "solutions" are fundamentally inequitable. They build a world of digital and physical segregation disguised as progress. And we users are expected to be thankful and grateful.” #a11y nimerblogs.blogspot.com/2025/08/cong...

nimerblogs.blogspot.com

Congratulations, You've Unlocked 'Expert Mode'! (Whether You Wanted It or Not)

Nimer's blog

062

Tom Yedwab @tomyedwab.com · 13/11/2025

Personally I think it's worse than this: pleasing the marginal user is *the right thing to do* when all you have are early adopters and you're trying to go mainstream. It's when you get to that long tail, low propensity marginal user that you start to damage the product.

030

Tom Yedwab @tomyedwab.com · 06/11/2025

Unpopular opinion of the day: protecting software developers from interruptions in the name of preserving "flow" state is overrated, and potentially harmful for the team as a whole. Deeper dive here: www.arguingwithalgorithms.com/posts/flow-i...

arguingwithalgorithms.com

"Flow" in software development is overrated

Tackling the myth that

100

Reposted by Tom Yedwab

Birgitta B @birgitta410.bsky.social · 15/10/2025

I tried to make sense of "spec-driven development" by looking at 3 tools: Amazon's Kiro, GitHub's spec-kit, and the Tessl Framework martinfowler.com/articles/exp...

martinfowler.com

Understanding Spec-Driven-Development: Kiro, spec-kit, and Tessl

Notes from my Thoughtworks colleagues on AI-assisted software delivery

1279

Tom Yedwab @tomyedwab.com · 03/10/2025

If anyone from YouTube is following: you could save a ton of compute if we all just filled out a form documenting our current opinions and beliefs. The algorithm could more efficiently match us with confirming and affirming videos. Bonus: The form would never need updating!

000

Reposted by Tom Yedwab

Merriam-Webster @merriam-webster.com · 01/10/2025

We are thrilled to announce that our NEW Large Language Model will be released on 11.18.25.

649249438244

Tom Yedwab @tomyedwab.com · 30/08/2025

The distinction between "making" and "consuming" music might be going away. There is always human intent when I as a consumer define what it is I want to consume. Now I don't have to be limited to what artists have chosen to create.

000

Tom Yedwab @tomyedwab.com · 30/08/2025

IMO more accessible music generation would allow me to hone in on my own personal music style, even if the quality is less. Like smartphone cameras: we create 100x as many photos compared to 3 decades ago, moments that professional photography wouldn't have captured.

110

Reposted by Tom Yedwab

Andrew Garrett @andrewgarrett.bsky.social · 06/08/2025

My Berkeley colleagues & I are saddened by the death of our colleague Robin Lakoff. Her 1972 book Language & Women's Place created the modern field of language & gender. She also wrote articulately, passionately & impactfully about Latin linguistics (Abstract Syntax & Latin Complementation, 1968) 1/

49129

Tom Yedwab @tomyedwab.com · 16/06/2025

Databases don't contain truth. They contain data. Context, grounding, interpretation are all delegated to external sources. Give me a big enough dataset, and I can "prove" any proposition you like. Large open datasets exist but using them to derive truth is not easy at all.

000

Tom Yedwab @tomyedwab.com · 22/05/2025

If you've tried agentic AI coding for larger features and got stuck in a "death spiral", I highly recommend this pattern: www.arguingwithalgorithms.com/posts/techni...

arguingwithalgorithms.com

How to keep your AI coding agent from going rogue

Complete agentic LLM coding projects reliably with this proven workflow.

010

Tom Yedwab @tomyedwab.com · 09/04/2025

Wrote up some thoughts about the difficulty in using AI agents effectively in legacy codebases and how we might overcome it: www.arguingwithalgorithms.com/posts/legacy...

arguingwithalgorithms.com

Overcoming the legacy roadblock

Legacy codebases are the bane of AI tools. How can we overcome this?

010

Tom Yedwab @tomyedwab.com · 19/02/2025

We agree that LLMs produce prose at an average (I would argue, slightly above average) ability. Half of writers are by definition below average, and would therefore write better with the help of an LLM.

100

Tom Yedwab @tomyedwab.com · 19/02/2025

Perhaps the great tragedy here is the many mediocre writers who could benefit from LLMs but will reject them due to illusory superiority bias.

110

Reposted by Tom Yedwab

Charity Majors @charity.wtf · 10/02/2025

I have a new blog post up: "Corporate 'DEI' Is An Imperfect Vehicle for Deeply Meaningful Ideals" charity.wtf/2025/02/10/c... I know a lot of people are feeling incredibly scared and demoralized right now. I get it; I am too. (Who knows if my marriage will still be valid in two years?)

charity.wtf

Corporate “DEI” is an imperfect vehicle for deeply meaningful ideals

I have not thought or said much about DEI (Diversity, Equity and Inclusion) over the years. Not because I don’t care about the espoused ideals — I suppose I do, rather a lot — but because corporate…

1228975

Tom Yedwab @tomyedwab.com · 06/02/2025

A write-up of my recent experiment to build and deploy a personal self-hosted task list application using AI tools from scratch in a single week, how it went and what I learned. It is absolutely wild what we can do today with these tools! www.arguingwithalgorithms.com/posts/the-ei...

arguingwithalgorithms.com

The Eighteen Hour Application: The Project AI Made Possible

How I used AI tools to write a personal application from scratch in 18 hours.

000

Tom Yedwab @tomyedwab.com · 10/01/2025

On a technical level I agree with your observations. What I am interested in is whether the focus shifts from fancy GUIs as the end product to data systems that can be connected to LLM frontends. Will we see new startups where the API is the primary value prop, and not just an afterthought?

010

Reposted by Tom Yedwab

Anna Mills @annamillsoer.bsky.social · 01/01/2025

"Having overcome my broader skepticism, I find I’m just shocked at the mental gymnastics militant cynics will execute to avoid admitting that generative AI is doing anything particularly unique or useful."--@mikecaulfield I do see that too sometimes. mikecaulfield.substack.com/p/critical-r...

mikecaulfield.substack.com

Critical Reasoning with AI: How we know LLMs are applying reasoning patterns, and not just reverse image searching

Writing this because I'm tired of having this argument

66716

Tom Yedwab @tomyedwab.com · 06/11/2024

I recommend this 3blue1brown video on the subject. Very accessible but also good for building intuition: youtu.be/eMlx5fFNoYc?...

youtu.be

Attention in transformers, visually explained | Chapter 6, Deep Learning

YouTube video by 3Blue1Brown

010

Tom Yedwab @tomyedwab.com · 29/10/2024

I wrote up a review of Cursor and how I use it for software development, which seems to be a hot (and mildly controversial) topic these days: www.arguingwithalgorithms.com/posts/cursor...

arguingwithalgorithms.com

How I write code using Cursor: A review

A personal review of Cursor, an LLM-powered coding tool.

381