Week 19 · Season 5 · Shield
Real people
You can interview real users, turn what they say into tests, weigh a feature's cost against its value, add plans and usage limits, and ship through CI/CD with a rollback.
This week
- Ask about their week: talking to users
- The test set is the spec
- Is it worth it? Cost vs value
- Users, plans and usage limits
- From commit to production: CI/CD and rollbacks
How you learn
Write your interview questions before you meet anyone (only about the past), and turn every complaint into a failing test case before you change any code.
What you build
Project 5 and the boss: five real friends use FRIDAY for a week. Interview them, turn their problems into tests, give each a plan with limits, ship the fixes through staging and then production, and roll back once on purpose.
Stretch: Put your riskiest change behind a feature flag, turn it on for one friend first, and add a per-user meter with an allowance.
What you ship
Project 5: Five real friends. Five people use FRIDAY for a week; their problems become tests, and the fixes ship through CI/CD.
- Notes from five user interviews, and the test cases they became
- Sign-in, scoped data and per-user limits (429 with Retry-After)
- A CI/CD pipeline with staging, production and a rollback you have tried
Lessons
- Ask about their week
After this deck you can run a short user interview about someone's real, recent week, tell a polite yes from evidence, and pick a problem that is narrow, frequent, painful and checkable enough to build FRIDAY's next power for.
- The test set is the spec
After this deck you can turn a vague goal into a spec made of test cases (a real input, the expected outcome and a pass rule a script can check), set a ship bar before you build, and write a one-page plan with success measures and non-goals.
- Is it worth it?
After this deck you can work out whether a FRIDAY feature is worth building: the cost per successful task (all the cost divided by the results that worked), the time saved after checking and failures, and the break-even success rate.
- Who are you, and how much?
After this deck you can put a "who are you?" check on every route, scope every piece of data to the signed-in user, and give each user a plan: a per-user rate limit that answers 429 with Retry-After, plus a daily usage limit.
- From commit to production
After this deck you can describe a CI/CD road from commit to production (tests, one build, a practice copy, then live), roll back to the last good version first and debug second, and turn the cause of a bad deploy into a check on the road.
- Go deeper: meters, flags and listening
After this deck you can meter every model call per user (counting a repeated event once), compare flat, per-use and allowance pricing, roll a change out to one friend with a feature flag that doubles as an off switch, and pick the right way to learn from users: interviews, watching, surveys or numbers.
Labs
Python in the browser, next to the lessons.
Weekend build
- Project 5: Project 5: five real friends
Five real people use FRIDAY for a week. Their problems become tests, and the fixes ship through CI/CD.
Skills you prove
- Interviews
- Interview users and find a workflow worth automating
- PRD + ROI
- Write a PRD with eval-as-spec and an ROI model
Free links for this week
Hand-picked official docs, papers and talks.