Developing a Product Mindset

The /grill-me and /grill-with-docs skills created by Matt Pocock solve a major alignment issue in AI development. The goal of these tools is to ensure both the user and the agent are confident that the other party actually knows what needs to be built. They establish what features are expected and why the work needs to be done. Adding documentation to the mix creates much better handoff and persistence of concepts. This leads to more consistent results across sessions. You can see how this fits into a broader workflow in my previous post about Matt's Claude skills.

These skills are reasonable proxies for product direction. Most engineers using AI coding agents already have strong engineering skills. Their primary area for improvement is product direction. The grill-me skill forces you to understand what you want out of your product. This often ends up with a narrow scope still.

Nothing replaces detailed product understanding. Good developers understand their products. Many wunderkind developers had their first big break when they built a product they understood incredibly well. They leveraged their ability to solve engineering problems to deliver a great product experience.

We have many ways to understand a product.

1. Use every competitor product available.
2. Read articles about the product category. Finding a Wired article that complains about the shortcomings of a similar product in the same problem space is great.
3. Chat with an AI about the product.
4. Create a VC presentation deck for the product.
5. Try to explain the product to a rubber duck.
6. Use Google Stitch or Claude Design to generate a UI.
7. Do market research to find what people are paying money for to solve the problem.
8. Use voice mode with an AI and ask the AI to ask you one question at a time until you are too frustrated to keep going.
9. Search for public GitHub repos in this problem space and have an AI look at the commit history.

All of these processes are additional ways to make sure you understand the product and problem space. Using an interview skill is just one way.

When an engineer runs into a complex problem, they often run into more problems later. They easily start blaming the AI. The root cause is usually a lack of product understanding. This results in unexpected problems coming up during development and the AI simply makes guesses. When all of these product edge cases and needs are very well understood, these problems are resolved trivially.

Doing the research ahead of time is the hard part. The only other tricky part is making sure this gets communicated clearly to the AI. This is where documentation comes into play. Well structured documentation is vital. AI is often much better at translating product understanding into documentation. AI can also make documentation that looks prettier with diagrams. This makes reading through the documentation much easier and enjoyable. This aligns well with how persisting project context helps your agent work faster.

The fundamental skill developers using AI agents need to develop is adopting a product mindset. This is easy when the developer cares about the product. Developing the skill to get a product mindset even when you do not care about the product is the real power move. When that happens, the developer becomes unstoppable.

Matt Pocock breaks down this approach to working with AI agents in his video.