Pragmatic AI: The AI Product Sprint

The overall arc: vague idea → validated spec → working tool → reliable tool → grounded tool → trustworthy tool → proven tool → efficient tool → safe tool → shipped tool → tested tool → demoed tool → career leverage.

Week 0 — Setup & baseline. You arrive with a rough idea. You pick your strongest capstone, run a first baseline test in ChatGPT or Claude (before building anything), and lock your build lane. End state: a Capstone Brief v0 and a baseline — nothing built yet.

Week 1 — Product judgment. The shift from "an idea" to "a spec worth building." You decide whether the idea deserves AI at all, then narrow it to one user, one workflow, one painful task, one metric, and five real sample inputs. End state: a validated, specific Brief v1.

Week 2 — First working workflow. The first time it actually runs. You build a version that completes one useful task end-to-end. Difference from W1: spec → a working demo you can show.

Week 3 — Reliability (Health Gate 1). You go from a single prompt that sometimes works to a multi-step workflow that recovers when it doesn't — adding a check, retry, fallback, or human-review step. Difference: "works if you're lucky" → dependable.

Week 4 — Grounding. You connect the tool to a real source of truth (your docs, tables, policies). Difference: generic model answers → answers grounded in your actual data.

Week 5 — Trust layer. You stop it from sounding confident while being wrong — adding citations, evidence checks, or refusal behavior. Difference: grounded → grounded and honest about what it doesn't know.

Week 6 — Scorecard (Health Gate 2). You move from "it seems good" to proof, scoring your tool against raw ChatGPT/Claude on a real test set. Difference: opinion → measured evidence of where it wins and fails.

Week 7 — Model & cost. You stop over-paying for the strongest model everywhere and choose deliberately per task via a small bakeoff. Difference: whatever-model → the right model at the right cost.

Week 8 — Safety & governance. You add guardrails, spend caps, and access rules. Difference: safe for you to run → safe enough to put in front of someone else.

Week 9 — Hosting (Health Gate 3). You take it off your laptop and make it accessible via a live link or demo environment. Difference: private → shareable.

Week 10 — Real-user testing. You watch an actual user (or peer) use it and capture proof. Difference: you-tested → real-world-tested, with evidence.

Week 11 — Demo Day. You learn to present it credibly, without overclaiming. Difference: built → demoed and defended.

Week 12 — Portfolio & work story. You turn the whole thing into resume language, a LinkedIn post, a manager memo, or a founder pitch.

Difference: a project → professional leverage you can use.