Most people talk to Claude like they're Googling something. They type a vague question, get a vague answer, and blame the model.
A growing number of practitioners are pushing a different approach: treat Claude like a junior developer on your team, and yourself like the product manager. That means writing specs, not prompts.
The difference is practical. Instead of "write me a marketing email," a PM-style prompt looks more like a brief: here's the audience, here's the goal, here's the tone, here's what success looks like, here's what to avoid. You define acceptance criteria before the work starts, not after you're disappointed with the output.
This works because large language models respond well to structure and constraints. A 50-word prompt with clear requirements will outperform a 200-word rambling request every time. The same principle applies in software teams: a developer with a clear ticket ships faster than one deciphering a Slack thread.
What PM Thinking Actually Looks Like in Practice
The core habits translate directly:
- User stories over instructions. "As a newsletter editor, I need a subject line that drives opens from SaaS founders" gives Claude more to work with than "write a good subject line."
- Acceptance criteria up front. Specify word count, format, tone, and what the output should NOT include. Claude follows constraints better than open-ended requests.
- Iteration as process, not failure. PMs expect multiple drafts. Treat Claude's first output as a starting point and refine with specific feedback, not wholesale rewrites.
- Scope control. Break complex tasks into smaller, well-defined steps instead of asking for everything at once. Claude handles a sequence of focused requests better than one massive ask.
None of this is new wisdom for anyone who's managed a product team. But applying that muscle memory to AI interactions is a mental shift most people haven't made. They're still in "search engine mode" - type question, expect answer - when the real leverage comes from "manager mode" - define outcome, review work, iterate.
The people getting the most out of Claude right now aren't prompt engineers. They're people who already know how to write a clear brief.