Generating more code wasn’t my problem.
I can now ask agents to explore options, implement changes and check work across the development lifecycle.
The temptation is to keep them busy. There’s something satisfying about watching several tasks progress while I’m still drinking my coffee.
Then the results arrive, and someone has to understand how they fit together.
I use agents where I can define a useful task, provide enough context and inspect the outcome. Sometimes that means implementation. Sometimes it means investigating a failure, questioning a design or finding gaps in a test strategy.
Thirty years ago, I was connecting machines. Later, I coordinated teams. Today, I also coordinate agents... and still spend a surprising amount of time checking whether everyone understood the same request.