Build Sprint · Open now
AI Agent Workflow Challenge.
Plan, build and ship one feature with AI agents, and show a repeatable workflow that a human clearly leads.
The brief
What you are asked to do
The goal in one line. Show a clear, repeatable path from a human-written epic to a shipped feature, with agents doing the building and you doing the steering and review.
AI agents are fast, but they are only as good as the instructions you give them. This challenge is about the instructions: how you turn a whole-product idea into small, clear issues that an agent can finish in one change, and how you check the result.
Pick an existing product, your own or an open-source one you have permission to work on. Choose one real feature. Write the epic and the milestone yourself. Break the work into issues. Let agents draft the changes. Review every change like a careful teammate would, and ship.
The result is not only a feature. It is a written workflow someone else could follow: what you gave the agents, what came back, where it went wrong, and how you fixed it forward.
We suggest one to three weeks of part-time work. This is a suggestion, not a deadline: the brief is open until we announce otherwise on this page.
Who it is for
Is this challenge for you?
Good fit
- Developers who already write code and want a disciplined way to work with agents.
- Learners who have finished week 3 or 4 of the Developer path and want to practise epics and hand-off.
- Small teams of up to three who want to compare their agent workflows.
Not a good fit
- Asking an agent to invent the epic or the roadmap for you. That breaks Rule 1.
- Accepting agent changes without reading them.
- Large rewrites. Choose one feature that fits in a few days.
You do not need to be an expert. You need to be curious and honest. Plain English is fine. You can take the free Developer path alongside.
What to hand in
Your deliverables
- A human-written epic. Written in your own words using the epic standard: goal, measure, scope in and out, acceptance, risks, owner.
- A milestone and clear issues. Issues small enough for an agent to finish in one change, each with acceptance checks. Keep them in your repository.
- The shipped feature. The feature merged and working in a public repository, with tests or a clear way to check it.
- A workflow log. For each issue: the instruction you gave, what the agent returned, what you changed, and how long it took. Be honest about the failures.
- A time comparison. A simple, honest note of how long the work took with agents compared with your own estimate of doing it by hand. Measured or labelled as an estimate. No invented numbers.
- An AI-use note. Which tools and models, which tasks, and what you verified yourself.
- A short Product Package. At least the current state, the roadmap, the demo and the evidence of epic management, linked from your README.
Rules
The rules of this sprint
These come from The Way. Two of them decide most disagreements: people write epics, and you disclose AI use.
- People write the epics and the roadmap. You write every epic and every roadmap sentence yourself, in your own words. AI never does. This is Rule 1 of The Way. Plain, simple English is welcome; clear thinking matters more than perfect grammar.
- Disclose all AI use. Keep an AI-use note in your repository: which tools, for which tasks, and what you checked yourself. Honest use is welcome. Hidden use is the only real way to lose points.
- Everything lives in a public git repository. Your work is a public repository on GitLab.com or GitHub. History matters: judges read your commits, issues and decision log, not only the final result.
- Open source first, and honest licences. Prefer open-source tools. Say which licence your own work uses, and give credit for anything you reuse.
- Fix forward. When something breaks, write down the cause and ship the fix. Do not hide it or quietly undo it. Your decision log should show this.
- No private data, no secrets. Use fake data in demos. Never commit passwords, keys, or real customers’ details. Do not submit anyone else’s work as your own.
- Team size: one person, or up to three. You may work alone or in a small team of up to three people. AI agents are tools, not team members. Name every human who took part.
- Follow the code of conduct. Be kind and specific when you give feedback. The Developer Dojo code of conduct applies to every submission and every conversation about it.
- This brief. Every merged change must have been read by a human before merging. Say so in the log.
- This brief. Include at least one thing that went wrong and how you fixed it forward.
Read more: People write epics, Disclose AI use and The Way: five rules.
How it is judged
Judged on the Product Package
Reviewers do not count lines of code. They read your Product Package and score six skills from 1 (beginning) to 4 (strong). The target is level 3 on all six.
A. Whole-product view
Does your map show every part of the product and how the parts connect?B. Customers and personas
Are your personas based on real evidence?C. Journeys and touch points
Did you map what users and customers actually go through?D. Writing epics
Are your epics clear, measurable and in your own words?E. Roadmap and priorities
Can you defend the order, and say what you chose not to do?F. Hand-off to agents
Do agents get instructions clear enough to do the right work?
- Hand-off to agents (skill F) counts most here: do the agents produce the right work first time because your issues are clear?
- Is the log honest? A candid log with failures scores better than a perfect-looking one.
A missing part of the package is a gap to fix, not a score. Full levels and examples are in the Product Package rubric; the eight parts are in the Product Package standard. Rewards are being finalised, and no prize is promised.
How to submit
Six steps. No form on this site.
You hand in a public repository. This website collects nothing from you.
Make a public repository
Create a free account on GitLab.com or GitHub if you do not have one. Create a new public repository. Use a plain name such as
ai-agent-workflow-challenge.Put your work in it
Copy the templates you need. Add the Product Package folder, your epics, your decision log and your AI-use note. Commit as you go: your history is part of the result.
Add a SUBMISSION.md file
In the root of the repository, add the short file shown below. It tells the reviewers where everything is.
Check it as a stranger
Open the repository in a private browser window. If you can see everything without signing in, so can the reviewers.
Send us the link
Use the Curiosity Research Corporation contact form (it opens on the CRC website). Use the subject Build Sprint: AI Agent Workflow Challenge. Include your repository link and the names or handles of everyone on the team.
Wait for your review
Reviewers score the six rubric skills and write short feedback. We cannot promise a date yet. You can keep improving your repository while you wait, and tell us when you have a new version.
SUBMISSION.md
# Submission: AI Agent Workflow Challenge
- Team: <names or handles of every human who took part>
- Product: <name and one sentence>
- Product Package: ./product-package/README.md
- Demo: <link or ./product-package/06-demo/>
- AI-use note: ./product-package/08-evidence/ai-use.md
- Licence: <the licence of your work>
- I confirm that I wrote the epics and the roadmap myself.