Eight more months of agents

David Crawshaw’s follow-up to Programming with Agents, written eight months later. The change he measures is in what he spends his time on rather than in the tooling: at a big company his split was 80-20 reading code to writing it, at a startup nearer 50-50, and now it is 95-5. The model does the writing. A year ago Claude Code could write a quarter of his code; in February the latest Opus model writes nine tenths of it.

His flat claim about where the progress came from:

Agent harnesses have not improved much since then. Right now, it is all about the model.

That leads to the practical advice. Use the best model available, because you will not know what models will be capable of unless you use the best, and cheaper ones teach you the wrong lessons about what is possible. On sandboxing, once you turn off the built-in sandbox you have to supply your own, and having tried the alternatives he recommends a fresh VM.

The other idea worth keeping is about what to build now that customers arrive with agents in tow: the best software for an agent is whatever is best for a programmer. Also worth noting he opens by saying he no longer uses an IDE, and closes on enjoying the work more than he ever has.