Go for agentic code is the right call—and the big labs are missing it

Microsoft released an open-source agent framework in Go last week. Google’s been building agent tooling in Go for a while now too. Meanwhile, OpenAI and Anthropic still default to Python for everything—TypeScript if you’re lucky. I think Go is the right call here.

I’ve spent most of this year building next to an agentic coding loop, and the thing I keep coming back to is that the code still needs reading. The model writes a lot, but you’re the one who has to understand it well enough to catch when it’s painted itself into a corner.

Go is simple. The syntax doesn’t hide much, and the fast compile times mean you can iterate on an agent’s output without a slow feedback loop. Compare that to Rust or C++: the code is dense, the borrow checker or template errors can obscure what the agent was actually trying to do. Python is fine for prototyping, but when you’re shipping production agents that need to be maintained, having a language that’s both performant and easy to audit is a real advantage.

The LLM companies might argue that in an agentic world, human code review matters less. The model will fix its own bugs. I don’t buy that—not yet. Taking a look over generated code yourself is never a bad idea. And Go makes that look quick enough that you’ll actually do it.

So this bet from Microsoft and Google feels like one that will age well.