15 min read

March 2026 Bakery

Content that caught my attention all in one go!
March 2026 Bakery

js-notes is my personal knowledge management system. I built it this month to captures articles, podcasts, and videos. It extracts their full text, and enriches them with AI-generated summaries, notes and tags which all becomes instantly searchable to me. This is everything that I added to js-notes in March.

I might consider organizing this by theme next month. Instead this list is chronological. Let me know if you'd prefer another format.

Steve Yegge coined "vibe maintainer" for his approach to reviewing AI-generated PRs in his open-source projects. Rather than rejecting AI submissions, he's built a workflow for encouraging and managing them.

Vibe Maintainer
Some attendees at an AI Tinkerers meetup in early Feb were asking me what it’s like to be the maintainer of a big OSS project where the…

Mike Masnick argues that agentic AI tools could democratize web development and help people escape corporate digital silos — a return to the early web's creative ethos, but this time with AI lowering the technical barrier.

AI Might Be Our Best Shot At Taking Back The Open Web
I remember, pretty clearly, my excitement over the early World Wide Web. I had been on the internet for a year or two at that point, mostly using IRC, Usenet, and Gopher (along with email, naturall…

Aaron Levie applies the Jevons Paradox to knowledge work: as AI makes individual tasks more efficient, the total volume of knowledge work expands rather than contracts.

Jevons Paradox for Knowledge Work
In the 19th century, English economist William Stanley Jevons found that tech-driven efficiency improvements in coal use led to increased demand for coal across a range of industries. The paradox of course being that if you assume demand remains constant, then the volume of the underlying resource s

If Books Could Kill examines the worst media takes on the U.S.-Israel military strikes on Iran, critiquing how prominent columnists echoed the same flawed arguments used to justify the Iraq War.

Claire Vo went from OpenClaw skeptic (it deleted her family calendar) to running nine specialized agents across multiple machines. A good look at what daily life with AI agents actually looks like.

From skeptic to true believer: How OpenClaw changed my life | Claire Vo
Listen now | Claire Vo went from OpenClaw skeptic to running nine specialized AI agents across her family calendar, inbound sales, kids’ homework help, and more

DHH announces Basecamp to prioritize AI agent integration. After finding native AI features underwhelming, they bet on making the platform agent-accessible instead.

Basecamp becomes agent accessible
In the past 18 months, we’ve experimented with a ton of AI-infused features at 37signals. Fizzy had all sorts of attempts. As did Basecamp. But as Microsoft and many others have realized, it’s not that easy to make something that’s actually good and would welcomed by users. So we didn’t ship. In the meantime, agents have emerged has th…

From an engineer at Anthropic on building multi-agent AI systems for full-stack apps. The GAN-inspired approach to frontend design quality is clever.

Harness design for long-running application development
Anthropic is an AI safety and research company that’s working to build reliable, interpretable, and steerable AI systems.

37signals is treating agent-friendliness as a form of accessibility. Their new CLI lets AI agents interact with Basecamp directly.

Bring your AI agents to Basecamp
Episode from REWORK

Marc Andreessen dismissed introspection as harmful. Jay Michaelson explains why that's historically illiterate and dangerous.

It Matters that Marc Andreessen is Sometimes Really Dumb
Optimizing for wealth is different from optimizing for wisdom. Why can’t powerful tech elites see that?

Product success depends more on company incentives and structural factors than individual engineering quality. Resonates.

Working on products people hate

Patrick McKenzie on how a YC-backed startup allegedly sold fake SOC 2 audit reports. The compliance industry has some wild corners.

Delve into compliance theatre
Delve and compliance theater.

MIT's Heather Kulik on why materials science lacks an AlphaFold-level breakthrough despite AI progress. The quantum mechanics complexity is real.

🔬Why There Is No “AlphaFold for Materials” — AI for Materials Discovery with Heather Kulik
Lessons from a Decade on the Frontier of AI for Science

Boz on writing as a thinking tool. Writing forces loose ideas into clarity. Even writing no one will read makes you sharper.

Writing Is Thinking
Writing is a linear process that forces a tangle of loose connections in your brain through a narrow aperture exposing them to much greater scrutiny

Waymo's co-CEO on going from lab research to 500,000 autonomous rides per week. Twenty years of patience paying off.

The 20-year journey to fully autonomous cars with Dmitri Dolgov of Waymo | Cheeky Pint | Episode 29
Waymo is now doing nearly 500,000 rides a week across 10 cities. Co-CEO Dmitri Dolgov came to the pub to discuss how they moved from scientific research to massive global scaling. He gives a masterclass on the sensor stack (and why you still need Lidar), how they use “Simulation” and “Critic”…

Scott Alexander reframes AI 'hallucinations' as 'shameless guessing.' The framing matters — it changes how you think about fixing the problem.

Shameless Guesses, Not Hallucinations

"bff (build, friction, fix) is the pattern i use to give myself the permission, the freedom, to get started and take the first step. it applies at every level of granularity: a single session, a larger feature, entire projects or dreams. a self-reinforcing fractal pattern. small to wide to wider."

bff: build friction fix
just start.

Roundtable with engineering leaders from Stripe, NVIDIA, Microsoft, Google DeepMind, xAI, Apple, and OpenAI on the future of software engineering in the age of AI.

The Future Of Software Engineering with Anthropic
With Ash from Anthropic and Sivesh from Balderton Capital

Evidence that AI agents may be degrading software quality and long-term development velocity. Worth reading with a critical eye.

Are AI agents actually slowing us down?
As more software engineers use AI agents daily, there’s also more sloppy software, outages, quality issues, and even a slowdown in shipping velocity. What’s happening, and how do we solve it?

Holding two contradictory truths: the Iranian regime is genuinely evil, and the Trump administration's prosecution of the war has been reckless. Given Harris's outspoken criticism of Trump, this is an absolutely insane take.

Moral Confusion About the War in Iran
To think clearly about this war, we need to hold two sets of ideas in our minds at the same moment: the Iranian regime is evil, and the Trump administration is dangerously amoral, corrupt, and incompetent.

World models learn to predict future states by observing video paired with actions. Agents can train in simulated dreams before deploying in the real world.

World Models: Computing the Uncomputable
A Co-Written Essay with General Intuition’s Pim DeWitte

Felix Rieseberg on Claude Cowork — Anthropic's desktop app for agentic knowledge work. Built in 10 days by orchestrating multiple Claude instances.

Why Anthropic Thinks AI Should Have Its Own Computer — Felix Rieseberg of Claude Cowork & Claude Code Desktop
Claude Cowork came out of an accident.

Stratospheric aerosol injection as climate intervention. Drawing parallels to how volcanoes naturally cool the planet by reflecting sunlight. His Lordship Neal Stephenson predicted this in Termination Shock.

Sunscreen for the planet, Out Loud
The world is warming faster than we can cut emissions. Volcanoes are already cooling the planet, with particles that reflect sunlight. Maybe we can too.

How semiotics applies to naming conventions in code. Variable names and class names function as signs that carry meaning.

Simon Willison on OpenAI acquiring the company behind uv and ruff. Is OpenAI acquiring the tools or the talent?

Thoughts on OpenAI acquiring Astral and uv/ruff
Plus GPT-5.4 mini and GPT-5.4 nano, which can describe 76,000 photos for $52

Thariq on the new Claude Code playground plugin — interactive HTML files for visualizing and iterating on problems.

Building effective agent tools requires matching the model's capabilities to its needs rather than maximizing the number of tools.

Feed 1,000 comments to an LLM and ask it to profile the user. Surprisingly effective and deeply unsettling.

Profiling Hacker News users based on their comments
Here’s a mildly dystopian prompt I’ve been experimenting with recently: “Profile this user”, accompanied by a copy of their last 1,000 comments on Hacker News. Obtaining those comments is easy. …

Simon Willison's guide to Git with coding agents. Agents have deep fluency with Git — let them handle it.

The Ghost CLI that's powering this very blog post. AI-native content management.

Developer Beta: ghst cli
Hi everyone, Wanted to share a new ghst CLI tool we’ve been working on - as a developer beta - which allows you to interact with Ghost publications from the command line. In short, this allows you (or an LLM like Claude, or Codex) to automate tasks within Ghost using a set of pre-built tools. Pretty much everything you can do in Ghost Admin, you can do using this CLI. For example: Create/edit posts and publish them Import or export members Download/upload/activate themes Find out which post…

Karynn Ikeda applies semiotics — the study of signs and meaning — to software naming conventions. Variable names and class structures communicate meaning to humans, not just machines.

- YouTube
Bekijk je favoriete video’s, luister naar de muziek die je leuk vindt, upload originele content en deel alles met vrienden, familie en anderen op YouTube.

The nuanced art of knowing when to intervene. Managers who never decide create as little value as those who decide everything.

dein.fr - Charles-Axel Dein
Charles-Axel Dein’s personal website.

A CTO built a markdown-based AI operating system using Claude Code to manage 10 engineers, ship code, and operate at C-level simultaneously.

Some spicy takes from Searls. If your values aren't right and you can't figure out how to deliver impact you are in trouble.

v53 - Pod Freeze
Episode from Breaking Change

Ross Douthat interviews a Claremont Institute fellow arguing white Americans face growing discrimination. The conversation is unhinged.

White Identity Is Galvanizing the Right
Episode from Interesting Times with Ross Douthat

Jewish establishment organizations are failing by focusing on left-wing anti-Zionism while minimizing more dangerous right-wing antisemitism.

Antisemitism Explodes on the Right, Watchdogs Blame the Left
The Anti-Antisemitism industry is myopically focused on the Left -- while some on the Left are starting to sound like the Right.

Jason Fried challenges the bespoke software hype. Most businesses don't actually want to build and maintain their own systems.

The bespoke software revolution? I’m not buying it.
A bespoke software revolution? I don’t buy it. It’ll exist. It already exists. Small consultants and big consulting firms have made custom software for years. It almost always sucks. It’s bloated, confusing, and because the client pays, it’s built in all the wrong ways. Who’s excited about bespoke software? Software makers! Of course t…

Patrick McKenzie on how debt collection operates as a commodified waste stream. Banks sell delinquent accounts as CSV files with minimal documentation.

Understanding consumer debt collections: the underbelly of finance
Episode from Complex Systems with Patrick McKenzie (patio11)

Simon Willison's Pragmatic Summit talk on engineering practices for AI coding agents. TDD as the core reliability framework, working from minimal infrastructure, and treating agents as real collaborators.

- YouTube
Bekijk je favoriete video’s, luister naar de muziek die je leuk vindt, upload originele content en deel alles met vrienden, familie en anderen op YouTube.

Ben Thompson on OpenAI shifting toward enterprise, drawing parallels to Dropbox's evolution and 1980s Microsoft.

(Preview) OpenAI’s Enterprise Pivot, The Rise of Agents and Bubble Counterpoints, Nvidia Changes Its Inference Story
Episode from Sharp Tech with Ben Thompson

Will Maier, growth engineering lead at Stripe, on how November 2025 was the AI inflection point. Relatable.

I Haven’t Opened an IDE Since November — Will Maier
Episode from StaffEng

Steve Yegge on how AI is transforming software engineering. His adoption framework is useful.

From IDEs to AI Agents with Steve Yegge
Episode from The Pragmatic Engineer

We are at a printing press moment.

Why Leonardo was a saboteur, Gutenberg went broke, and Florence was weird – Ada Palmer
Episode from Dwarkesh Podcast

Calling AI 'just a next-token predictor' is a category error — like calling humans 'just next-sense-datum predictors.' Levels of description matter.

Next-Token Predictor Is An AI’s Job, Not Its Species

Computer reviews function as permission slips that tell people what they're allowed to want. They miss how genuine obsession and skill development actually work.

“This Is Not The Computer For You” · Sam Henri Gold
Sam Henri Gold is a product design engineer building playful, useful software.

Scott Alexander argues progressives should stop pressuring thoughtful Republicans to publicly denounce Trump and resign. Removing moderating voices from the inside doesn't help.

Support Your Local Collaborator

Simon Willison's Pragmatic Summit talk on TDD, prompt injection defense, and shipping fast with AI agents. Essential viewing.

52 newsletters, a podcast, 50+ YouTube videos, a completed book, and coaching 200+ engineers — all from two focused hours per day.

2 Hours of Focus Will Put You in the Top 1%
Three ways to find focus time that don’t require waking up at 4am.

DHH presents ONCE for simplified multi-app deployment on a single server.

37signals pivoted ONCE from paid self-hostable apps to open source after limited market interest. I'm running Fizzy instances thanks to this.

ONCE (Again)
The original concept for ONCE sought to sell self-hostable web apps for a one-time fee. That didn’t work. Sure, we recouped the investment on Campfire, our chat app, but that was it. You gotta listen when the market tells you what it wants! And it didn’t seem to want to pay for self-hosted web apps in a one-off way. So we set Campfire,…

Joel Spolsky's classic argument that rewriting software from scratch is a catastrophic mistake. The Netscape cautionary tale is evergreen.

Things You Should Never Do, Part I
Netscape 6.0 is finally going into its first public beta. There never was a version 5.0. The last major release, version 4.0, was released almost three years ago. Three years is an awfully long tim…

Oversimplified solutions to complex problems offered by people with superficial knowledge. Dan Luu at his best.

Cocktail party ideas

Traditional code review is becoming obsolete when agents generate code faster than humans can review it. The shift to specification-driven development.

How to Kill the Code Review
Human-written code died in 2025. Code reviews will die in 2026.

StrongDM's 'Software Factory' where coding agents write, test, and deploy code entirely without human code inspection. Bold.

How StrongDM’s AI team build serious software without even looking at the code
Last week I hinted at a demo I had seen from a team implementing what Dan Shapiro called the Dark Factory level of AI adoption, where no human even looks …

"The definition of poverty is terribleness caused by scarcity." Patrick McKenzie on the dark underbelly of gift card regulation.

The gift card accountability sink
The gaps scammers exploit in gift cards are the result of deliberate policy tradeoffs.