Tag: software-engineering

Will we stop asking our colleagues because AI never looks busy?

I don’t miss waiting days for an answer. I miss the conversations that sometimes started with a simple “Got a minute?”

I Stopped Checking If You Know the Answer

A candidate repeated my question back to me, waited a beat, and recited a definition word for word. The questions did not change. What they measure did.

The Plan Didn't Say Push

Instructions Aren't Contracts

The rules are written down. The agent reads them on every prompt. They still do not make it into the code, and nothing is broken enough for the bot to report. A rules file is not configuration. It is a suggestion with an unknown success rate.

Agent Config Isn't Configuration. It's an Execution Surface

I audit dependencies. Lockfiles, advisories, the whole ritual. Then the keyv worm shipped its payload as committed agent config, and I realised I have no equivalent reflex for the files that configure the thing writing my code.

Today, I Wouldn't Hire My Younger Self

I put my younger self through a 2026 hiring process. I failed the screening. What started as a joke exercise revealed something worse: the filter that used to let the right people into this profession is gone, and nothing has replaced it.

Review Is Infrastructure. Your AI Rollout Forgot to Build It.

AI adoption doubled our code output. Nobody budgeted for the reading. Telemetry from 22,000 developers shows the bill: 5x longer review, tripled incidents, and 31% more PRs merged with no review at all. This is what treating review as infrastructure actually means.

You Can't Take the Career Path I Took. It No Longer Exists.

AI code isn't dangerous at random

I almost shipped a vulnerability to production because the model handed it to me with full confidence. Then I checked the numbers — the 2026 ones, not the two-year-old ones. Models hallucinate less, but picking a model stopped helping, and the attack surface is now shared by all of them.

The Most Important AI Work Doesn't Show Up in Any Metric

Every company brags about their model's stats. Nobody tells you what you have to set up, and what you have to prompt, to actually hit those numbers. The hardest part of working with AI is the part no benchmark measures.

AI Made Me Fluent in Backend. It Didn't Make Me a Backend Engineer.

AI equalized the part of the job that was never the hard part — writing the code. The hard part was always knowing what to ask. That part it can't hand you, and worse, it hides the fact that you don't have it.

We Don't Trust AI's Code. We Also Don't Read It.

I committed code I hadn't read, felt the guilt, and almost filed it as a discipline problem. It wasn't. The gap between not trusting AI's code and not reading it is the clearest signal we have that the valuable part of the job moved downstream — from writing to reading.

Claude Fable 5 Dropped. My Pipeline Isn't Getting It (Yet)

Claude Fable 5 launched this week — record benchmarks, twice the price of Opus. My production pipeline's config hasn't changed, and that's not neglect. Benchmarks measure raw models. Pipelines run calibrated ones.

The Line Doesn't Defend Itself

I wrote about the decisions I won't hand to AI. That post quietly assumed I'm the one standing at the door. On a team of ten, I'm not — and the line that lives only in my head is the first thing AI erases.

AI Isn't Burning Us Out. The Narrative Around AI Is

Simon Willison says coding agents leave him wiped out by 11am. Engineers at every major shop are publishing essays about AI fatigue. I run a frontend team and use AI across my entire workflow — and I think we are blaming the wrong thing. It is not the model that exhausts us. It is the story we keep telling ourselves about the model.

What I Won't Let AI Touch

I run a nine-agent code review system in production. I still won't hand AI five decisions — not despite knowing these tools, but because I do. The ones you can't cheaply reverse stay mine: the decisions that compound.

AI-assisted coding has a slot machine problem

AI can make developers faster. But speed without reading is just a faster way to create technical debt.