Building with AI agents: where they help (and where they don't)
Agents are extraordinary at some jobs and quietly terrible at others. Knowing which is the whole game.
AI agents have gone from demo to default in an astonishingly short time. Used well, they let a small team operate with the reach of a much larger one. Used carelessly, they add a layer of expensive unpredictability to things that were fine as plain code.
Match the tool to the task
Agents shine on open-ended, judgement-heavy work — research, drafting, triage, orchestration across messy systems. They struggle where you need guarantees, determinism, or tight latency. A wire transfer should not be a vibe.
So we draw the line deliberately: agents where flexibility pays off, conventional engineering where correctness is non-negotiable, and clear seams between the two. The result is systems that feel intelligent without becoming impossible to trust or debug.