team working around a table

One Agent Is a Tool. Several Are a Team to Manage.

This year I learned Cursor, and I began experimenting with how to manage several AI development agents at once, including forking a project built around that idea. Very quickly, the challenge stopped looking like programming.

The problems sound familiar

Agents can duplicate each other’s work. They can lose the context of a decision made an hour ago. They can finish a task confidently while misunderstanding the goal. Anyone who has led a technical team has seen all of this before, only more slowly.

Structure beats cleverness

What helps agents is what helps people: a clear scope for each task, written documentation that survives from one session to the next, and a defined point at which someone checks the work. A longer prompt rarely solves what a better process would.

My role is shifting

I notice that I spend less time writing each line and more time deciding what should be done, in what order, and how we will know it is finished. That is closer to the work of an architect or a project lead than to that of a typist.

There is a real risk in this change. It is easy to accept output you did not read just because it arrived quickly. The more agents I run, the more deliberate I have to be about what I actually verify.

Leave a Reply

Your email address will not be published. Required fields are marked *