I have a feeling that, for coding, orchestration tooling for AI agents will become more important than the actual AI models.

Just like in human teams it’s the team that makes the most difference, with its devex and processes (10x teams >> 10x developers), I believe we will rapidly get to the point where good orchestration will make it viable to run everything through open source models.

In general, directionally, if orchestration matters at all, we need to believe that the gap between running open source models and frontier ones is bound to get smaller, not bigger.

The only counter-scenario is that orchestration proves to be immensely better if vertically integrated — e.g. directly managed by Claude Code — but I am kinda skeptical this is going to be the case.