Feb 28, 2026

Another AI coding thought, sorry. I liked the answer Carson Gross gives to the question: “Given AI, should I still consider becoming a computer programmer?” in his essay “Yes, and…”. He points the the AGENTS.md he recommends to his students, which limit LLM to helping without providing direct solutions.

This clicked with Martin Fowler’s observations on pair-programming with LLM.

From my personal experience, I can see parallels between outsourcing and LLMs. Outsourcers can be good or not technically, but they will never build enough domain-knowledge of an in-house team.

Company can gain short-term boost in productivity by using an external team. Losing a chance to have people who deeply understand the system few years down the road.

Or they can acquire some technical debt by rushing developers in a crunch. Bad code is nothing new.

But now the choice is made not by management but by individual contributors. They can go slow, using LLM to validate and challenge ideas and approaches, and building the system (along with their domain knowledge) themselves. Or they can go fast by letting LLMs go wild. Of course, there’s a middle ground and we’re yet to discover the golden spot.