The theory that we'll vibe code everything is 'flat wrong'.
The theory that we'll vibe code everything is 'flat wrong'.
--Anish Acharya
If you think AI is going to replace your engineers, you are about to create the most expensive mess in your org.
The failure mode is predictable: a team ships a working demo in a week, leadership assumes the rest is just "more of that", and the codebase turns into a pile of unowned artifacts.
AI is great at generating code. It is terrible at carrying accountability.
It will not sit through the ugly tradeoffs: What gets simplified so the warehouse stays consistent. What breaks when a pricing rule changes. Who gets paged at 2:00 a.m. How audit logs actually need to look.
So the org starts treating software like content. Copy, paste, merge, ship. GitLab lights up with noise. JetBrains inspections are ignored. HashiCorp configs drift. Datadog alerts get tuned out. Snyk becomes a weekly fire drill.
AI should also earn its keep by unlocking work you would not have executed at all: adding property-based tests, writing migration rehearsals, instrumenting the dark corners, generating runbooks, building internal tools that were always "next quarter."
Use AI where it compounds judgment: exploring options, accelerating refactors, tightening tests, documenting intent, shrinking the time from decision to validation.
The coding was never the bottleneck.
It was deciding what you are willing to own.