Migrations with AI — the pulse we need to talk about

Gergely Orosz at the Pragmatic Engineer is laying out a real problem: the way teams are using LLMs for code migrations is moving faster than the tools can actually help. The post walks through the gap between what people expect from AI and what's still genuinely hard — understanding context, spotting edge cases, and catching the subtle bugs that only show up after you've shipped.

The specific moves that are working are narrow and well-defined: renaming a library, updating imports, converting a config format. The things that fail are the ones that require reading between the lines — why does this function exist, what does this config value actually control, what breaks if I change this one thing. Those require a human in the loop who knows the codebase, not just a model that has seen it.

The honest takeaway is that AI is a force multiplier for the work you already know how to do. It's not replacing the engineer who understands the system. It's the cousin who helps you move boxes — useful, but you still need to know which boxes are fragile.

Why this matters for us: the same pattern is happening with every tool we use — it speeds up the work we already do, and it fails on the work that actually requires judgment.