This is the same process I used for 8 years as a BA and PO. AI did not change the process. It removed the wait.
Not "we need a feature." A real pain point. Can you explain it in one sentence? If not, you do not understand it yet. I spend an hour here. Sometimes two. Everything else depends on it.
Example: "Sprint planning takes 4 hours because we argue about story points instead of discussing scope." That is a problem. "We need better sprint planning" is not.
What does "solved" look like? Not vague aspirations. Specific, testable outcomes you would recognise immediately. If you cannot measure it, you cannot ship it.
Example: "Sprint planning completes in under 90 minutes with zero arguments about story points." That is measurable. "Better planning" is not.
Describe what you want the same way you would write a user story — inputs, outputs, processing. Average: 10–20 iterations per tool, 5–20 minutes each. No code written by me. All decisions made by me.
The prompt is not "build me an app." It is: "Given a list of features with reach, impact, confidence, and effort scores, calculate RICE scores, sort descending, and render a bar chart. Export to CSV." Specific. Testable.
Show it to 3–5 colleagues. Do not explain it. Watch. User confusion = design problem, not user problem. Fix the biggest confusion point first. One fix only. Not ten.
The Stakeholder Map Generator failed its first test because users did not understand what "Power" meant on Mendelow's Grid. I added a one-sentence tooltip. Test passed. Ship.
No attachment to ideas. Attachment to outcomes. If it works, share it. If it does not, try a different problem next week. Either outcome is a win because both teach you something.
The Backlog Health Analyzer was killed in 1 day. That was a good day. I learned that "health" is context-dependent, and I saved weeks of polishing a fundamentally flawed concept.
Perplexity Pro + Gemini Pro + Claude Code. Less than a parking spot in Toronto.
From problem identification to ship or kill decision. Some take 1 day. Some take 2 weeks.
Mission Control PWA alone would cost $80,000+ via agency. Built for ~$180 in AI costs.
I wrote a 12-page guide on exactly how I use AI to build tools in 5 days. No fluff. No theory. Just the exact prompts, iterations, and kill criteria I use.
Get the AI Execution Guide