# Jobs-to-be-Done for Probabilistic Software

**Track:** Foundations of the AI-Native Venture — AI for Entrepreneurship — complete (29)
**Framework / surface:** venture strategy
**Level:** Beginner
**Prerequisites:** What AI-Native Means

**In one line:** Failure probability × cost × detectability × reversibility — the calculus that finds the jobs AI can hold today.

## Theory, aesthetics & inspiration

Clayton Christensen's jobs-to-be-done lens — buyers hire products to do a job, the insight behind the milkshake study he made famous and "Competing Against Luck" (2016) — needs one amendment for probabilistic software: the job must tolerate uncertainty because its outputs are reviewable, recoverable, or measurable. Failure needs an acceptable cost structure, and the calculus has four factors: probability of failure, cost of failure, detectability, and reversibility. Draft-shaped work — writing, code, research, triage — scores well on all four and is AI-native territory today; wire transfers and dosage fail on cost and reversibility and demand a human in the loop. The evidence says this discipline is not optional: Stanford's 2026 AI Index reports that agents, for all their advances, still fail roughly one in three attempts on structured benchmarks. The frontier of "reviewable" expands with every release, but founders who map their market by failure calculus rather than by industry find the openings incumbents miss — and know exactly when the curve unlocks the next one.

**Founder question:** What happens when your product is wrong — who notices, what does it cost, and can it be undone?