We're asking the wrong question about AI agents in production

We’re asking the wrong question about AI agents in production.

The debate right now is if they’re only good for prototypes or if they can actually ship to prod.

Both camps are loud. Both are right… in their own environment.

The better question is what separates them.

I’ve seen two big factors.

1/ Tech stack. LLMs know Python, Javascript, Go, Java far better than Scala or Elixir. If your stack’s niche, you’re fighting uphill.

2/ Engineering practices. This is the biggest unlock IMO, and it’s one we can all fix.

Ask yourself: How long does it take a new hire to ship a meaningful change to prod?

If you’re thinking “less than a day”, you’re probably in Camp 2. If you’re sighing and thinking “weeks…”, enough said.

Think about it. A new engineer joins, can they figure out what to change without asking fifteen questions? Do they know if their change will break something else? Can they even tell if they’re in the right part of the codebase?

The same friction that slows engineers slows down agents. Inconsistent observability tags. Services scattered across random namespaces. Abstractions that only make sense if you were there when they were built. Tribal knowledge buried in Slack threads.

We tolerate it because the team’s adapted to the pain.

If your environment isn’t agent-ready, it’s barely human-ready.