Build yours
One idea.
One useful result.
Start with a small service for people you can reach. Write the brief, build a first version, then use their feedback to improve it.
Need the concepts first? Learn how delivery and improvement loops work →
Step 1 · Write your brief
Turn your idea into six clear lines.
Choose one idea, then fill in the brief below. A first draft is enough: write what you know and mark what you still need to test. Choose a class idea or use your own.
Save your six-line loop brief.
Your draft stays in this browser. Copy it into your own document before switching devices or clearing browser data.
Use real user feedback to challenge your brief. An AI score is advice, not a grade or proof of demand.
Need help writing the brief? Use this prompt.
Help me design one useful loop for real users. My idea: [who it helps and what useful result it gives them]. What I still need to test: [the biggest assumption or weakness]. Use simple words. Ask if anything is unclear. Give six short lines: user and value; trigger, steps and delivery; pass/fail check and safe action on failure; human approvals, limits and stop; user feedback and a measured improvement; first test this week. Keep the first version small enough for a student team to build in four weeks. Mark guesses. Distinguish what exists from what we still need to build. Do not invent customers, revenue or working integrations.
Optional: inspect a workshop example first
Inspect a real result.
Open the page you saved in class. Read it as its intended user: is the offer clear, useful and believable? Note one weakness. Write down one change you would test.
No saved page? Inspect this delivered example and do the same review.
Inspect an existing example →Seeing a polished page is not evidence that customers want the offer.
Step 2 · Build a first version
Build it in five testable steps.
Use your six lines as the specification. Build one small path all the way to a real result before adding more features.
- 1 · Input
- Pick one real user and one request you can handle. Save a concrete example input and write the useful output they should receive. Start with a form, message or scheduled data source you have permission to use.
- 2 · First result
- Ask your coding agent to make the smallest working path from that input to that output. Run it yourself once. Show the result to five people outside your team; find out whether it is useful.
- 3 · Check
- Save good, bad and edge-case inputs. Define a pass/fail rule and test that it catches the bad results. Choose a safe failure action: a bounded retry, holding delivery for review, or stopping. Test the failure path too.
- 4 · Run again
- Add the trigger or daily schedule, a running server, saved run status and delivery. Record what has already been sent so retries do not send duplicates. Test a restart and a failed delivery. Set limits, alerts and a stop control before leaving it unattended.
- 5 · Measure
- Record real users, returning users, successful and failed deliveries, cost per useful result, and minutes of manual work. Review the evidence daily. Test one change at a time against saved cases and real user feedback.
A laptop script stops when its runtime stops. A hosted service still needs working credentials, scheduling, monitoring and someone responsible for exceptions. Keep the templates below with your project.
Take it into your project
Download the working templates.
Write the brief before building. Save test cases as you find them. Keep evidence of what runs, what users get and what still needs you.
- Challenge
- The challenge in one page · PDF
Understand the model
The theory has its own home.
Learn the delivery and improvement loops, then apply them to your brief.
Go deeper when you need it
Help for the next obstacle.
Choose a tool stack and follow the running-loop guide →
Inspect Danielius’s real builds and how he made them →
Four more prompts
Plan the first version
Use my six lines. Give me small steps to build the first working loop: example input, first result, checks and failure path, trigger and saved state, delivery, then measurement. For each step, say what to do, what should happen, and how to test it. Explain new words. Mark anything that needs my account, permission or approval.
Find a problem
Expected: [what should happen]. Actual: [what happened]. Error: [paste it without passwords or keys]. Give one likely cause and one test. Say what would show that the fix worked.
Write the check
Here is my loop, one good result and one bad result: [paste]. Propose pass/fail checks before delivery. Separate exact code checks from checks that need AI judgment. Give good, bad and edge-case inputs with the expected result. Define a safe action for each failure. Say what these checks cannot prove.
Find hidden human work
Read my loop: [paste]. Where must I still copy, click, fix or deliver? Which approvals should stay? Suggest one change and a test. Keep plans separate from things that already work.
Workshop recap
The 18-slide presentation from 13 October 2026 is available as an archive. Example figures are labelled snapshots; the class simulation is fictional.