Man-Months

I’ve been using coding agents every day for a while now. Claude Code, Codex, whatever I can get my hands on. There’s this thing they all do. I give them a task and they go “this will take approximately 3-4 weeks, shall I implement phase 1?” Different agents, different tasks, same habit. Then it finishes in 20 minutes.

I read The Mythical Man-Month a long time ago and this keeps reminding me of it. Even the AI thinks in man-months now. All the Jira tickets and sprint plans and PRDs in the training data, the agents absorbed all of it. People still use that framing too. “Built by one person in three months.” I keep hearing it about OpenClaw, about a lot of projects lately. But one person with a coding agent isn’t really one person in any traditional sense. The way we talk about work just hasn’t caught up.

I don’t read the code my agents write. Run it, works, keep going. Most people I know work the same way. Code review, test coverage, careful releases, all that stuff gradually became optional while everyone was busy building. Not overnight. It just kind of happened.

And it’s kind of a carnival right now. OpenClaw has hundreds of thousands of stars on GitHub, millions of people using it, security researchers keep finding new things every week, and it just keeps growing. Apple’s selling more Mac Minis than they probably expected because everyone’s running agents on them. Model providers are seeing record token usage. Everybody showed up and everybody’s doing well.

I notice the same in my own work. I used to want to get things right before putting them out. Now I just ship and deal with whatever comes up when it comes up. I find out what the real problems are much faster that way. Every time I hesitate on something I realize I’m just falling behind for no reason.

A lot of stuff I was used to, things I held onto as basics for years, just quietly dissolved. And more is getting built right now than probably ever before.