Run Claude Like a Manager, Not a Prompt Writer

Most developers re-explain the ticket in every message. After 15 workdays of running Claude Code like a small team at work, “continue” is a complete instruction for me.

Last weekend, I reviewed my Claude Code usage history. Over the course of 15 working days at my primary job, I had typed a large number of messages into it.

The median was about 90 and more characters, and 73 were under 40: “continue”, “done login”, “go”. That’s just my estimate. I’m not entirely sure.

Most days I had between three and twelve sessions open, each with a name and a job. A developer session wrote the code.

At my company, we handle a large number of PRDs, bug fixes, and other tasks simultaneously. This has been the case since we were provided with Claude Team.

A QA session built the app, tested it in the simulator and wrote up what it found. On features where the spec kept moving, a third session did nothing but reread the PRD.

I read their reports(in arifact), answered the product and backend questions, logged in wherever an OTP was needed, and kept git to myself.

The long instructions were already written down, in places the sessions read on their own. That’s why two words were enough.