#ai

One year later, I write for myself first

The thinking matters more than the finished article.

Stop treating shipped specs as documentation

Specs guide the build, then become history.

My first public skill - and the part it can't hand you

The one command installs the prompt. Not the work.

I don't read most of the code my agents write

I read the critical code, skim the rest, and never hand off the architecture.

I gave my blog a Markdown twin for agents

Same URL, 80% fewer tokens. No server.

I was regenerating Claude's answers just to copy them

There was a slash command for that.

LLMs are shortsighted and that's your fault

They optimize for now. You're supposed to know what comes next.

I built a statusline because /usage is a distraction

Stop interrupting yourself to check how much you have left.

The leaked code isn't the story. The response is.

Blame the process, not the person.

I wiped my Mac and forgot to install Cursor

From must-have to afterthought in under a year.

The code your AI writes is untrusted. You run it anyway.

AI-generated code is untrusted by definition. So why are we running it with full access to our credentials, filesystems, and production databases?

Your AI setup is premature optimization

You already know the rule. You're just not following it.

Not knowing how to code is not a superpower

It's a ceiling.

Code review is becoming a fiction

LGTM is not a review strategy.

Most of your AGENTS.md is wasted tokens

4% better. Or worse. Depends what you write.

Chasing models won't make you better at using them

Skill the model, not the benchmark

Why I'm just a 'decent' developer (and that's perfect for AI)

Ego kills AI adoption faster than bad code

AI hype cycle: Wright brothers edition

Flying jets, building runways

AI will write 90% of code. Here's why that doesn't scare me

It's not replacing us

We're all getting hooked on cheap AI

$20 monthly, $140 worth of compute