I build AI products. I also use AI to make product decisions. The meta approach works well.
Here's my process.
I define a feature. I ask myself if it's something an agent should handle. If yes, I build it into the agent. If no, it stays manual.
This thinking pattern shapes product direction. I'm not building features for humans. I'm building capabilities for agents.
For example, when designing OpenClaw Setup, I asked: what should be automated? What should require human judgment?
The answer determined the feature set. Core workflows are automated. Edge cases require humans.
This approach keeps the product focused. We don't over-feature. We automate what matters.
I also use AI for user research. I analyze feedback. I spot patterns. I prioritize based on actual usage, not assumptions.
AI-driven product decisions feel different from traditional approaches. You're not just building for today. You're building for how work will happen tomorrow.
If you're building AI products, think about your own workflow. Where can automation replace manual work? That's your feature list.
See my product thinking: harshith.vc
Building AI products requires thinking like an agent, not like a human.