AI Venture Build · Capstone brief
VEN-07 · Trust as a feature
Your product will be wrong sometimes. Design the disclosure, the human review step, and the failure experience that make a buyer say yes anyway — then test whether the design actually works on someone who has not seen it before.
The question
Your product will be wrong sometimes. Design the disclosure, the human review step, and the failure experience that make a buyer say yes anyway — then test whether the design actually works on someone who has not seen it before.
System / materials
The venture's core AI-assisted flow, and an honest account of how it fails: wrong output, low confidence, no answer, slow answer, offensive or unsafe output. Three to five test readers or users from the intended audience. Comparable disclosure patterns from products the student can actually examine, described from observation rather than memory.
Expected failure modes
A disclaimer in the footer treated as disclosure. Human review promised in the pitch and absent from the cost model, which VEN-04 will contradict. Designing only the happy path plus an error message. Confidence scores shown to users who have no way to interpret them. Testing on people who already know how it works. Making the disclosure so prominent the product is unusable, which is the opposite failure and just as real.
Done looks like
A trust design: the failure inventory with rough frequency where known; the disclosure design in context, showing what the user sees and when; the human-review step with who does it, when it triggers, and what it costs — cross-checked against VEN-04's cost stack; the failure experience — what the user sees, what recourse they have, how a mistake gets corrected; and the test results, recording what fresh users actually understood versus what the design intended.
Five C's
CT: distinguishing disclosure that informs from disclosure that indemnifies. CR: designing recourse, not just messaging. CO: test users are collaborators and their misreadings are the data. CM: this is the brief's core. CZ: who is harmed by a wrong output, and whether the recourse actually reaches them.
Mentor role
A founder, designer, product manager, or customer-support lead reviews the failure inventory and the review-cost line. Standing instruction: reject any human-review promise that does not appear in the cost model. School-supervised.
Rubric calibration
R1: one flow, one failure inventory. R2: test procedure and results recorded. R3: comparator is the product with no disclosure and no review. R4: frequency and cost of failures addressed. R5: fresh users understood it. R6: names the failure the venture refuses to ship with.
Two ways this goes wrong
(a) A terms-of-service paragraph presented as trust design. (b) A human-review step that exists in the pitch and nowhere in the unit economics.
Credit lane fit
Lane A immediately (entrepreneurship or a design cross-listing). Requires VEN-04 for the cost cross-check. No verified credit claim.