To validate a startup idea with AI, use the assistant to structure your assumptions and experiments. Start with a rough direction; use your experience and customer access to frame the problem if the idea is still unclear. Then collect evidence from real customer behavior. A convincing AI-generated report is a planning artifact, not proof of demand.

Find the assumption that could stop the product

For a technical solo founder, the easiest part may be building. The riskiest assumption may be that a specific buyer cares enough to change a working habit.

Write four hypotheses before researching: a problem occurs often enough to matter; a reachable group feels it; an improvement is worth a meaningful commitment; and you can deliver it within your resources. Prioritize the one with the least evidence and the largest consequence if false.

For the hypothetical AgentCRM idea, the first question is not “Can I build reminders?” It is “Do independent agents experience costly missed follow-ups, and will they try a different process?” Keep the idea and founder constraints in a startup context file.

Keep an evidence ledger, not an enthusiasm score

Ask AI to organize evidence, identify missing sources, and suggest disconfirming explanations. It should not turn a few positive comments into a market verdict.

ClaimEvidence to collectWhat it cannot establish
Follow-up is painfulA recent incident, its cost, and the current workaroundThat every agent has the same problem
A new process is attractiveAgreement to try a defined pilot with real workThat people will use a full SaaS product
The product is worth paying forA clear offer and an observable purchase decisionLong-term retention or a scalable channel
The audience is reachableA repeatable way to contact eligible buyersThat outreach alone will create adoption

Record the observation, source or interview reference, date, segment, and limitations. Public reviews can reveal a problem or alternative, but an old complaint from another market may not apply to your buyer today. Check current sources when researching competitors.

Ask about the last time, not a hypothetical future

Use the conversation to understand the person's existing behavior before describing your solution.

  • “Walk me through the last inquiry that needed a follow-up.”
  • “How did you decide what to do next?”
  • “What happened when it was delayed or missed?”
  • “What tools or reminders did you use?”
  • “What have you already tried to improve this?”

Ask permission before recording. A concise, consented summary is usually enough for your evidence ledger. Have the assistant compare summaries and flag contradictory patterns. Keep actual statements separate from its interpretation.

A small set of interviews can refine the hypothesis, but it is not a representative market study. Look for enough detail to choose the next experiment, not a claim about the entire market.

Design one small behavioral experiment

For AgentCRM, a manual pilot could test a follow-up workflow before any software is built. Here is an example design, with invented planning thresholds that you should adapt:

Cohort: five independent agents who recently described a missed follow-up.
Offer: a one-week pilot that helps organize their next actions.
Founder effort: no more than six hours.
Signal: at least two agents agree to try it with real inquiries and return for a second check-in.
Next question: would they buy a clearly priced continuation?

This tests participation and repeat use, not willingness to pay. If you later make a paid offer, be explicit about what exists, what will be delivered, and the terms. Do not imply a finished product to collect a stronger-looking signal.

Using my current product and founder context:
1. List the assumptions by consequence and evidence gap.
2. Select one assumption to test this week.
3. Propose a cohort, action, and observable decision criterion.
4. Include a result that would make us change direction.
5. Keep sourced evidence separate from proposed experiments.

Make the next decision and save what changed

Choose go, pivot or no-go for a specific investment. Go may support another interview or a manual pilot before it supports software. Pivot changes the customer, problem or offer to test. No-go stops the investment and records what would make you reconsider. If the result is inconclusive, choose a sharper test with a time and spending limit.

Save the evidence, opposing observations, constraints, unknowns, and next action. In 1PC, Validation has explicit persistence rules for market research, alternatives, ICP, and validation findings. It uses your saved founder and product context; it does not make the evidence true.

If the evidence supports a bounded software experiment, use the solo-founder MVP plan. If reaching the buyer is still the constraint, use the first-customer channel guide. Return to the founder workflow to keep the stages connected.

Decide what deserves your next investment.

Use your founder constraints and a rough direction to review evidence. Free includes clarification and validation; choose a small test, a changed hypothesis or a no-go.

Start free ↗