Trial projects beat coffee chats: how to run a two-week co-founder trial
The co-founder interview does not work. Everyone performs well across a table; nobody's actual working behavior — responsiveness, scoping honesty, how they handle being wrong — shows up in conversation. The instrument that does work, recommended in nearly every serious founder retrospective, is the scoped trial: two weeks, real work, then a decision. Here's the runbook.
Choosing the trial project
Pick something real (from your actual roadmap — synthetic exercises test compliance, not judgment), scoped (completable by a competent person in the stated hours; a trial that overruns tells you about your scoping, not their ability), and collaborative (requires at least three working sessions together — the interaction is the point; a spec they execute alone tests contracting, not partnership).
Good examples from the AI-app world: rebuild the flaky extraction pipeline; design and ship one customer-requested feature end to end; take the AI-generated codebase through a structured review and produce the hardening plan.
The one-page agreement
Before hour one: work product belongs to the venture; confidentiality on materials shared; payment terms (pay if you possibly can — it keeps the trial honest and the power balanced); an explicit statement that this is a mutual evaluation with a decision date. That last clause is what makes the ending clean.
What to actually evaluate
Not "was the code good" — you may not be able to judge that anyway (bring your retainer engineer in for review if not). Watch instead:
- Did reality match their estimates, and when it didn't, when did they tell you?
- What did they do at the first ambiguity — guess silently, ask well, or stall?
- How did disagreement go? Manufacture one small disagreement on purpose if none arises.
- Do you have more energy after working sessions with them, or less? Ten years is a long time to feel drained.
The decision
On the named date, both of you answer in writing before talking: in, out, or one specific reservation. "Extend the trial" is allowed once. Twice is a no that neither of you is saying.
If the trial is a no, you've spent two weeks and some money to dodge the most expensive mistake in startups. Cheap.
What's the trial project you'd assign from your current roadmap? Post it — sharpening the scope in public is half the work.
Replies (4)
Follow-up from maker intake: "Payment feels strange for a co-founder trial — doesn't paying make it an employment vibe rather than a partnership?"
The vibe concern is real but inverted: unpaid trials create the worse dynamic — two weeks of free work for a stranger's company reads as tribute, and it filters out exactly the experienced people you want (they have paid alternatives). Payment makes the trial symmetric: you're both spending something. If cash is truly impossible, the honest substitute is a shorter trial plus deep mutual reference calls — not a longer free one. And say the quiet part in the agreement: payment implies no employment relationship; this is a scoped mutual evaluation.
The manufactured-disagreement advice sounds manipulative until you've skipped it. Twice in my own history I reached the end of a warm evaluation period having never once disagreed with the person — because we were both selling. The first real disagreement then arrived after commitment, with stakes. Engineering one small, honest conflict during the trial ('I don't think we should build it that way — walk me through your reasoning') costs a moment of awkwardness and shows you the single most predictive behavior there is. Every experienced founder I've compared notes with does some version of this.
Follow-up from maker intake: "What if their trial work is excellent but something feels off that I can't name?"
Treat the un-nameable feeling as data pending analysis, not noise. Concretely: extend once (allowed), and use the extension to test hypotheses — is it communication lag? Deference where you wanted pushback? Energy? Write down what would confirm or dissolve the concern before the extension starts. If it's still un-nameable at the end, the answer is no: a ten-year partnership entered over a persistent misgiving fails at a much higher rate, and 'excellent work, wrong partner' is a legitimate, complete conclusion.
Follow-up from maker intake: "How do I run this when the candidate is non-technical and I'm the engineer?"
Same structure, different artifact. Real project from the business roadmap: land three pilot conversations from a cold start; produce the pricing analysis with actual customer interviews behind it; write and ship the launch plan for a feature. Evaluate identically — estimate honesty, behavior at ambiguity, disagreement, energy. One addition for this direction: watch what they do with your time, since a business partner who treats engineering hours as free is a known failure pattern. The trial shows it in miniature when they ask you for 'quick' builds to support their pilot pitches.
Threads are permanent — locked, not deleted, once resolved. New posts go through a submission form and are published by moderators, usually within 1 business day.