WHAT THIS PHASE PRODUCES
You leave with
- A one-liner that survives customer pressure
- An Ideal Customer Profile grounded in Jobs To Be Done
- A hypothesis statement with an explicit test sentence
THE MENTAL MODEL
How to think about this phase
Most early ideas live as a vibe. They sound good in your head, they sound good when you pitch them at dinner, and they fall apart the first time someone asks who exactly will pay. The Idea phase exists to drag the vibe out into the open and check whether there is a real bet underneath.
A bet has three parts: a specific named customer, a specific problem they are paying some cost to solve today, and a specific change you believe will follow if you remove that cost. If any one of those is fuzzy, you do not have a bet. You have a hope.
The discipline is to write the bet down in one sentence so concrete that a stranger can disagree with it. If a friend cannot tell you why you might be wrong, the sentence is not falsifiable and you are still in vibe territory.
The Idea agent is a thoughtful collaborator, not a cheerleader. Its job is to push back when reasoning is thin and to surface assumptions you have not stated. Resistance here is a feature: every weak claim caught in Idea is a month not wasted in Plan, Create, or Test.
FREE AGENTS
Practice the Idea phase in ChatGPT
Idea phase · lock the hypothesis
Founding Hypothesis
Generate a testable one-sentence bet: named customer, specific problem, and the claim about why they'd choose you. Falsifiable from the first draft.
Idea phase · sharpen the one-liner
Idea Refiner (Pitch & One-Liners)
Pressure-tests your one-liner and rewrites it with three customer-specific variants. Each variant names a different segment and comes with the exact evidence that would falsify it.
COMMON TRAPS
What goes wrong in this phase
- Asking customers “would you use this?”. Answers are nearly worthless. Probe past behavior instead.
- Naming a category instead of a customer (“startups”, “PMs”, “designers”). The named-customer test fails immediately.
- Conflating excitement with conviction. Founder excitement is necessary but unrelated to whether the bet is real.
- Skipping the falsification sentence. Without it, every result later in the loop will be rationalised post-hoc.
WHAT GOOD LOOKS LIKE
A worked example
SETUP
A solo technical founder shows up with “I want to build an AI that helps PMs write better PRDs.”
OUTPUT
After Idea: “Senior PMs at Series-B SaaS companies (50–200 engineers) currently spend 3+ hours per PRD on internal alignment back-and-forth. If I can cut that to under 30 minutes by surfacing missing context before the first draft, three of five PMs we interview will say they’d trade their current process for it within two weeks of trying.” Specific customer, observable cost, falsifiable threshold.