In November 2025 I lost my job. I was a frontend developer, and the reason underneath is the one you already suspect: most of what I did — turning a Figma into JSX, building the screen, wiring the state — AI now does. Cheaper, and almost as fast. I could tell you this as a tragedy. It isn't.
What died was the human compiler role. The person who takes a design and turns it into code without adding judgment. That job was always fragile — it depended on nobody automating the translation — and now it's automated. What doesn't die is knowing what to build, why in that order, how to structure it so it holds. The model doesn't do that. The person governing it does.
So instead of competing with AI at what it does well — generating repetitive code at scale — I decided to govern it. I learned to build the system around the agent: the contracts that tell it what it can touch, the gates that prevent it from merging, the tests that detect when its behavior degrades. Not prompts. Engineering.
"Directing AI isn't writing good prompts. It's building guardrails."
Today I have a product in production built this way: 200+ jobs shipped by agents I govern. Each one with a safety gate that classifies whether the job is agent-safe or requires human review. Each one with a PR I review before it reaches main. None reaches production without that review. The most autonomous mode is switched off on purpose until the agent's behavior regression test covers every change.
I don't sell this as magic. It's an old instinct applied to a new kind of team: tolerances, load paths, designing for the bad day. The difference is that the team is agents, and the bad day is the model getting dumber without anyone noticing.
If AI is rewriting your job too, follow along
…or if you need someone to govern it in your product, work with me.