Developing a Product Mindset

Matt Pocock's /grill-me and /grill-with-docs skills exist because this keeps happening. They interrogate you until you and the agent agree on what's in scope and why. I wrote about the skills themselves in Matt's Claude skills. Grill-me is a proxy for product direction, not a substitute for it.

Most engineers I know who use Cursor or Claude Code already write decent code. Where they stall is where I stall: fuzzy product intent. Run grill-me on a half-formed idea and you still walk out with a smaller scope. That's the win. A small, understood slice beats a sprawling prompt every time.

I learned this the hard way on a side project last year. I asked an agent to "add onboarding." It invented a five-step wizard, email verification I didn't need, and copy that sounded like a SaaS landing page from 2019. I'd never decided what onboarding meant for my app. The agent guessed. Loudly.

Nothing fixes that except knowing the product. The people who ship fast with AI usually built something they already understand. Their own itch, their own workflow.

When the product isn't yours, I keep a short list of ways to get clear before I touch Agent mode:

1. Use the competitors. Not the homepage. Sign up, hit the walls, note what annoys you.
2. Read one sharp critique of the category. You're looking for what users hate, not a feature checklist.
3. Talk it through with an AI, but ask for questions, not answers.
4. Sketch a UI in Google Stitch or Claude Design so you're arguing with something concrete.
5. Explain the product out loud. If you can't say who it's for in one sentence, stop coding.

I don't run through a nine-step research ritual every time. I do enough that when the agent hits an edge case (refunds, empty states, team plans) I'm not discovering the requirement in the diff.

When the product is fuzzy, the AI fills gaps with generic software. That's a specification problem, not a model problem. Write down what you decided, even if the agent helped draft the doc. Same idea as persisting project context: stop re-explaining the project every session.

Product mindset is easy when you care about the thing you're building. The skill worth building is getting that clarity on a codebase you inherited or a ticket you'd rather not touch. That's when grill-me, docs, and narrow scope stop being ceremony and start being how you ship without babysitting every file.

Matt walks through the workflow in his video.