AI Agent Prototypes Can Replace Upfront Plans
When AI agents can build working prototypes in minutes, an early implementation can be more useful than a detailed specification.
Developed from a conversation between Pete Winn and Andy David
Andy describes a build-first approach that starts with only a nugget of an idea. Instead of refining it into a detailed plan, he gives it to an AI agent and asks for a working version. The result isn’t treated as the finished product. It gives him something concrete to inspect, react to and use to develop the idea further.
That first implementation might be discarded once it has served its purpose. Andy describes the first iteration as the plan, taking useful parts from the prototype and using them to shape the next version. Pete extends the idea into a visual canvas where several experiments could run at once, each with its own chat thread and the prototype being built alongside it.
Instead of planning everything before building, someone can try several approaches, compare the results and combine the most promising elements. The initial prototype helps reveal what works, what doesn’t and what is worth pursuing. When an agent can produce a working version in minutes, building something can be the quickest way to work out what should be built.