Week 3 · Your Build · by Shivali & Girish
Day 21 of 21
Sketch, sequence, and submit your Build Plan
This is the day you finish the course.
You arrived at Day 1 not knowing what serverless meant, or what authentication was, or whether your idea was worth your weekends. Today you can answer all three. You also know how to actually work with the AI tools that will help you build it.
The remaining job is to take everything you have learned, package it into a one-page Build Plan, and walk out with a sequenced version of your real build that you could start on tomorrow.
Image slot
Suggested meme: a photo or illustration of someone crossing a marathon finish line with no crowd in the background, just the line and the runner. Caption: 'shipping the smallest version is winning'. Save as public/lessons/day-21-meme.png and add src='/lessons/day-21-meme.png'.
The six fields of a Build Plan
Six fields. One page. No code.
Field 1, what I am building. One sentence. This is your Day 15 answer.
Field 2, who it is for. One sentence. The specific person who has the problem your build solves.
Field 3, the user's journey. Three to five steps. What the user does from arrival to value. Each step is one short sentence. Resist adding features that are not on the critical path.
Field 4, the stack. Your choice for each of: frontend (Day 8), backend (Day 9), data (Day 11), hosting (Day 12), auth (Day 13), monitoring (Day 14). For most builds this looks roughly like: TypeScript with Next.js (frontend and backend in one), Postgres on Supabase (data and auth), Vercel (hosting), Sentry (monitoring). Defensible defaults. You should be able to defend each choice in one sentence.
Field 5, where AI fits. From Day 17, which of the four roles, and from Day 18, how you will work with the AI to build it. ("Role 2 on the summary screen, Role 1 throughout the building, Claude Code as the default surface, CLAUDE.md committed.")
Field 6, the smallest thing I can build first. One sentence. The version of your idea you could ship in a week that would teach you the most about whether the idea is real.
Draft it in twelve minutes
Open a notes app. For each field, write the first answer that comes to mind. Spend no more than two minutes per field. Total time, twelve minutes. Do not polish.
If you cannot answer a field, that is information, not a failure. Go back to the lesson that covers it and figure out what you are missing.
Now close the notes app, do something else for ten minutes, and come back.
The second draft, three patterns that usually need fixing
When you reopen the draft, three things will probably want changing.
The journey field is too long. First drafts almost always pile on. Cut to the absolute minimum number of steps the user must take to get the value. If the journey is longer than three steps, ask whether each step really has to be there.
The stack field has a tool you have never used, picked because it sounded sophisticated. If you have no specific reason to pick a non-default tool, pick the default. "I picked Postgres because it is the default" is a stronger answer than "I picked CockroachDB because we might need horizontal sharding someday." (You won't.)
The smallest-thing field still describes the whole product. "The smallest thing I can build is the entire app" is not a smallest thing. Cut features until you have something that takes a week, not a quarter.
Sequencing the build
Now we sequence. The shape below works for almost every build we have shipped or coached.
Week 1, the smallest useful version. Just the smallest-thing field, end to end. Not pretty. Not complete. Useful, deployed, and used by you. At the end of Week 1 you have a thing you can point at and say "this works."
Week 2, the one extra feature the smallest version exposed as missing. Almost always there is one. The smallest version reveals it, because you have used the smallest version for a week and felt the missing thing. Build that. Resist building anything else.
Week 3, make it not embarrassing. Visual polish, edge cases, the things that would make you uncomfortable showing this to a stranger. By the end of Week 3 you have something you can post a link to.
Week 4, the five guardrails (Day 19) and the legal sanity checklist (Day 20). Before any real human other than you gets sent to it. Do not skip this week. It is the difference between "I shipped a thing" and "I shipped a thing and lived to do it again."
That is your four weeks. Most builds will fit. The ones that do not are usually too ambitious for v1.
The pre-launch checklist
Before you make your link public, run through this.
The five guardrails (Day 19):
- Secrets in environment variables.
- Personal data minimized, not logged, encrypted.
- Money on a circuit breaker.
- Rate limits on meaningful endpoints.
- Documented kill switch.
The legal sanity (Day 20):
- Terms, Privacy, Disclaimer, Contact.
- AI disclosure if user-facing AI.
- Data deletion path.
- Privacy jurisdiction known.
- Breach plan.
The basics:
- Monitoring (Day 14) wired up.
- At least one alert that goes to a real human.
- A
CLAUDE.md(or equivalent) committed for your future self. - You have used the kill switch at least once in a non-emergency.
If any item is "no," shipping is premature.
Submitting your Build Plan
Head to the Build Plan page from the course nav. The first version is a stub (the interactive builder is the next thing this site ships), so for today, fill the six fields out wherever you keep your notes, take a screenshot, and send it to yourself.
When the interactive Build Plan builder lands, you will be able to submit it here, get a finisher's card with a unique number, and post it wherever feels right. The card is for you, but the moment of "I finished this" is more powerful when it is also visible to someone else.
What happens after Day 21:
- You start the four-week sequence above on the build you sketched.
- You come back to specific lessons as you hit specific questions. ("Wait, what was the auth-service recommendation again?" Day 13.)
- You revisit the Playbook when you hit AI friction.
- You ship.
A note from us
When we built this course we kept asking what we wished someone had handed us when we started, that no one had. The answer was always some version of: a clear map of the foundation, plainly explained, with the working assumption that we were smart enough to make our own decisions once we had the vocabulary.
That is the course you just finished.
We hope you build the thing. We especially hope you ship it. The ones that get shipped are the ones that change anything, including the version of you who shipped them.
Day 21 wrap
The thing you can now say plainly. You can take any idea, run it through the Day 15 filter, pick a shape (Day 16), name where AI fits (Day 17), commit to how you will work with AI (Day 18), sketch the stack, apply the guardrails (Day 19) and the legal sanity (Day 20), and sequence the build into something you could start tomorrow. The whole foundation, end to end.
The thing you can now do. Submit your Build Plan, and start the four-week sequence on your real idea.
The guardrail to remember. The course was the easy part. The first weekend of your build is when this all becomes real. Schedule it before you close this tab.
That is the end. Go build the thing. We will see you on the other side.