Everyone is now a manager of agents — the four-step workflow loop

The mental model for working with AI coding agents at scale is that you are no longer a coder; you are a manager of agents. The workflow runs in a loop with four stages. First, you delegate tasks — typically 8 to 10 at once — each running in its own fresh cloud VM. Second, the agent works unattended and ships a diff, creating a pull request for you to review. Third, you review and decide: you're processing 20 to 40 PRs a day and for each one you either accept it, redirect the agent, or kill the task entirely. Fourth, your technical depth grows because the sheer volume of code review you're doing compounds your understanding over time. This is the core reframe: your job is no longer writing code, it's making a high volume of high-stakes decisions about code all day long.