Building is cheap now. An agent can turn an idea into a working app in an afternoon. That makes the expensive mistake building the wrong thing well, not building it badly.
So before I build anything, I make the idea answer five questions.
- How often does it hurt? Weekly beats yearly. A problem people hit once a year is one they forget about.
- What does it cost them now? In money, hours or risk. If you can’t put a number on it, nobody will pay to make it go away.
- What do they do about it today? Spreadsheets, a part-time hire, a duct-taped workflow. If they do nothing, it isn’t painful.
- Who pays, and do they have a budget? The person with the problem is often not the person with the card.
- How will you reach the first 100? “Post it on Twitter” is not an answer.
Making an agent do the homework
I turned this into a Claude Code skill called validate. Give it an idea and it starts one background agent per question, all in parallel. Each one digs through real evidence: forums, reviews, pricing pages, job posts. It cites every claim, so I can check the work instead of trusting a vibe.
Then it writes a verdict:
- Build: the pain is frequent, expensive and already being paid for.
- Narrow: there’s something here, but only for a smaller group.
- Kill: move on.
Each question gets one line, plus the cheapest no-code test of the weakest answer. A landing page, ten cold emails, a post in the forum where these people already complain.
Why it works
Most ideas die on question 3. When I look, people almost always already have a workaround they’re fine with. Finding that out takes the agents ten minutes. Finding it out the other way takes months.
A good product is a painful, frequent, already-paid-for problem, and a way to reach the people who have it. Everything else is execution, and execution is the cheap part now.