Ask an agent to build a product, and it will. At speed. With all the confidence of someone who has never had to maintain it.

The problem isn’t that agents write bad code. Give them a small, well-specified feature and they are brilliant. The problem is that vague requirements force them to make hundreds of decisions, and those decisions compound into expensive mistakes.

So I have changed how I use them. First, agents help challenge the design, explore edge cases, and spike the uncertainties. Then they break the work into small, dependency-ordered nibbles with explicit acceptance criteria. Fresh agents implement and review one nibble at a time.

It sounds slower than throwing a prompt at an agent and watching thousands of lines appear. It is. But writing code is now cheap. Bad decisions are not.

My latest on using agents to spend more time deciding what to build, and less time fixing what got built.

Read more at Agentic Nibbles