Atlas · Essentials · 25 entries
Essentials
These are my best and most impactful articles and essays. The newer ones (from 2026) have become required reading from teams to boardrooms, and the older entries are the ones that people tell me they love most. I've thrown in a few of my own favorites for good measure. You can also visit my full bibliography for more adventures in writing over the years. If you have a favorite piece that you think is a must-read for this list, let me know!
- ★ Essentials
- 👍 Good read
- 💩 Not worth it
- 🤓 Language nerd
- 😄 Humor
- 📚 Story / fiction
- 🕹️ Gaming
- 🔮 Called it
- 🤡 Whiffed it
-
2026
The AI Vampire
The cost/value dynamics of running AI agents at scale. Why some companies turn token spend into compounding leverage and others turn it into a slow drain.
-
2026
The Anthropic Hive Mind
Anthropic runs as a vibe-driven hive mind — no spec, no plan beyond ninety days, improv at scale. The argument: this is how every successful company is about to have to work. And Golden Ages survive only while there is more work than people.
-
2026
Software Survival 3.0
Steve's account of which software survives in a world where AI writes everything: tools survive if they save cognition — and there are six levers you can pull to make sure yours does.
-
2026
Welcome to Gas Town
The opening essay of the Gas Town arc. The next leap isn't a smarter solo agent — it's orchestrating twenty or thirty of them at once, and Gas Town is the "Kubernetes for agents" Steve built to do exactly that.
-
2025
Revenge of the Junior Developer
The sequel to The Death of the Junior Developer. AI coding is rolling through six overlapping waves toward agent fleets — and the developers who adapt fastest, not the most senior, are the ones who come out ahead.
-
2024
The Death of the Junior Developer
The 2024 essay that opened the junior-developer debate. Chat-oriented programming made senior developers cheaper to keep than juniors to train, and the hiring market noticed before anyone wanted it to.
-
2023
Cheating Is All You Need
Written a few months after ChatGPT shipped, this is the essay where Steve — a programmer with a long, well-earned skeptical streak — went all-in in public. It demolishes the "you can't trust AI code" objection, reframes AI coding as a search problem, and names the "data moat" that the entire retrieval era would spend the next two years chasing.
-
2020
Dear Google Cloud, Your Deprecation Policy Is Killing You
The famous "Besties Forever" rant. Google Cloud breaks customers' software on a yearly schedule and dresses the notice up as friendly — and the essay's quieter point is that backwards compatibility is what keeps platforms alive for decades.
-
2018
Why I Left Google to Join Grab
The 2018 post announcing Steve's jump from Google to Grab. Underneath the news is the idea that outlived it: a company's capacity to innovate tracks how close its people sit to customers — Grab's mantra, "go to the ground."
-
2012
Return of the Mystery Machine Bus
The immediate August 2012 sequel to Notes from the Mystery Machine Bus. Half rebuttal pass, half new material: code as asset versus liability, applications as cities versus military encampments, and action-at-a-distance — the deeper technical disagreement under the conservative/liberal axis.
-
2012
Notes from the Mystery Machine Bus
A long, wine-paced Google+ essay from 2012 that introduces a vocabulary still in use — software liberal and software conservative. Conservatives manage risk; liberals manage speed; the two camps are genuinely irreconcilable.
-
2011
Stevey's Google Platforms Rant
An internal Google+ post from October 2011 that a permissions slip made public. Under the leak is the argument that a company's fate turns on whether it builds platforms or products — and the seven-point Bezos mandate, written out crisply here for the first time, that the industry has quoted ever since.
-
2008
The Universal Design Pattern
The humble name/value pair — written off everywhere as a cheap hack — argued out as the best known way to build software that can grow and last. Steve calls it the Universal Design Pattern, and leans on Hofstadter to earn the name.
-
2008
Business Requirements are Bullshit
The companion piece to Good Agile, Bad Agile two years earlier. The argument is blunt: the entire industry of formal requirements gathering is built on a category error, and almost every artefact it produces is somewhere between useless and actively harmful. The cover is the joke compressed: the scroll ends in the wastebasket.
-
2008
Done, and Gets Things Smart
An essay built on the Dunning-Kruger Effect, with Steve as the example: incompetent people overestimate themselves and fail to recognise skill in others — which makes "hire smart people" circular. Classic interviews, Steve argues, mostly clone the interviewer. What you want instead is rarer: the seed engineer who finishes absurdly fast and improves the systems and the culture while they are in there.
-
2008
Dynamic Languages Strike Back
Steve's case that the two standard knocks on dynamic languages — they are slow, and their tooling is bad — are not intrinsic, and are about to be demolished by runtime techniques: polymorphic inline caches, type feedback, and the then-new trace trees. Along the way he argues static type systems have hit their theoretical ceiling, and that language popularity has frozen for a decade. He transcribed the hour-long talk himself.
-
2008
Get That Job at Google
Steve finally gave in and wrote the interview-prep post he had been dodging for years, knowing it would draw flames. Underneath the comedy is genuine service journalism: study a data-structures book, run mock whiteboard interviews, know Big-O cold, know your graphs — and understand that a rejection from a high-false-negative-rate company is often just noise, not a verdict on you.
-
2008
Portrait of a N00b
Comments are metadata, Steve argues — and so are static types, which are just a specialised kind of comment. Over-relying on either is "metadata addiction," the hallmark of a n00b; veterans instead develop "compression-tolerance," preferring dense code they can hold whole in their head. Even readers who hate the conclusion tend to concede the framing is sharp.
-
2007
Code's Worst Enemy
A self-described minority opinion, made with the rare authority of someone indicting his own work. Steve's game, Wyvern, grew to a 500,000-line Java code base — lovely outside, horrific inside, and finally too big for one person to manage. He argues that refactoring and design patterns only ever add code, that IDEs cannot save you, and that if you genuinely need to shrink a code base, you cannot keep using Java.
-
2007
Rich Programmer Food
A 2007 essay with a deliberately rude thesis — if you don't know how compilers work, you don't know how computers work — and a generous one underneath it: compiler construction is the course that ties a whole CS education together, and writing your own is how you stop waiting for someone else to fix your tools.
-
2007
The Pinocchio Problem
A long, restless essay that opens with Steve admitting he doesn't understand his own thesis well enough to write it down — and then writes it down anyway. The result is one of his most-quoted "all software is crap" arguments, and the closest he came in the 2000s to a unified theory of software design.
-
2006
Good Agile, Bad Agile
Borrowing the good-cholesterol / bad-cholesterol trick, Steve argues there is a Good Agile and a Bad Agile. Bad Agile is a consultant-born ritual of sprints and estimation that survives precisely because software productivity cannot be measured. Good Agile is what Google did instead: incentives over whips, a priority work queue, real engineering discipline, and launches treated as an emergent property of a healthy system.
-
2006
Execution in the Kingdom of Nouns
A 2006 essay disguised as a fairy tale. In the Kingdom of Javaland, verbs are slaves — every action must be wrapped in a noun, a Manager or a Factory or an Executor, before it is allowed out in public. Steve tells the story deadpan, and the joke became the standard critique of object-oriented overreach.
-
2005
Practicing Programming
Drawing on his years as a self-taught guitarist — and on the hard lesson that "practice makes perfect" is wrong, because perfect practice makes perfect — Steve argues that merely doing your job is not practice. Great engineers are great because they train deliberately and constantly. The essay closes with twelve concrete practice drills you can start today.
-
2004
Tour de Babel
A 2004 languages tour from the Drunken Blog Rants era: C, C++, Lisp, Java, Perl, Ruby and Python, each handed a verdict in Steve's voice. Profane and quietly serious — and underneath the scorecard, an argument about what a language does to the people who think in it.
