Course Challenge
Bring the four competencies together: a scenario-based knowledge check, then a capstone in which you run a small fleet of agents in Antigravity and take their work through Delegation, Description, Discernment, and Diligence.
How the Challenge Works
The challenge has two parts:
- Knowledge check: ten scenario questions covering every module. Aim for at least 8 out of 10. If you miss one, read the explanation and revisit that lesson.
- Capstone task: run two parallel agents on real work, plus one read-only automation, and take everything through the 4Ds. Keep a short log as you go.
This course is free and self-paced, so there’s nothing to submit.
Part 1: Knowledge Check
1. (Module 1) Your manager asks why you aren't running ten agents at once if Antigravity allows it. What's the fluent answer?
In agent-first development, the human's review capacity limits real throughput.
2. (Module 1) The agent hands back a polished, confident implementation plan for your first mission. What should you do before approving?
Judge content, not formatting.
3. (Module 2) Which pair of tasks should NOT run as parallel agents?
The second task depends on decisions the first hasn't made yet.
4. (Module 2) A teammate leaves Turbo on for every project because prompts are annoying. What do you suggest?
Fix prompt fatigue with better rules, not by removing every checkpoint.
5. (Module 3) An unattended agent hit a failing test it couldn't fix and marked it as skipped. Which brief line would most likely have prevented this?
Tell agents what to do when stuck. Otherwise they improvise.
6. (Module 3) Every agent should know your test command and that money is stored as integer cents. Where should that go?
Persistent context briefs every agent automatically.
7. (Module 4) The walkthrough says 'Fixed the bug; all tests pass,' and the diff shows a test file with an assertion loosened. What happened?
Walkthroughs are claims. The diff shows what really changed.
8. (Module 4) A browser recording shows a password reset completing smoothly. What does it NOT show?
Recordings show one happy path's UI. Edge cases and server-side behavior need their own checks.
9. (Module 5) An agent that has your cloud credentials and can run network commands is asked to summarize issues filed by outside users. What's the main risk?
Break the chain: run untrusted reading without secrets or outbound tools.
10. (Module 5) Nobody remembers who set up the hourly automation that runs on your most capable model. What should happen?
Every automation needs an owner, a purpose, and a review date.
Part 2: Capstone, A Small Fleet Through the 4Ds
Pick a project you work on. You’ll run two parallel agents on independent tasks and set up one read-only automation.
- Choose two real tasks that touch different files and share no undecided design choices. Write why they’re independent.
- Choose one read-only automation, such as a daily test-run summary or a weekly dependency audit.
- For each, answer: verifiable? sized for review? reversible?
- Set a permission preset for the project, plus at least three allow, two ask, and two deny rules.
- Confirm a clean git state and that non-workspace file access is off.
- Check that your AGENTS.md has exact commands, conventions, and a “when blocked” rule.
- Write a four-part brief for each agent, including what evidence the walkthrough must show and, for UI work, which browser checks to record.
- Choose a model and thinking level for each agent and for the automation, with one sentence on why.
- Start both agents in Planning mode. Leave at least two inline comments on each plan before approving.
- For each agent, read the diff first, then the walkthrough. Mark each walkthrough claim as verified, unverified, or wrong.
- Check each diff against the failure modes: scope creep, weakened tests, new dependencies, swallowed errors.
- For any browser evidence, write what it proves and what it doesn’t. Do one manual click-through yourself.
- Have a fresh agent on a different model review the riskier of the two diffs. Log what it found.
- Run your merge checklist from the last lesson on each change.
- Merge only what passes. Note anything you held back and why.
- Give the automation an owner (you), a purpose, and a review date, and check your overage setting.
- Disclose AI assistance according to your team’s or project’s norms.
- Finish your log: how many agents you could review well, one line to add to AGENTS.md, and one thing you’d change about your permission rules.
About the Framework
The 4D AI Fluency Framework was created by Rick Dakan and Joseph Feller with Anthropic, and is released under CC BY-NC-SA 4.0. This course’s lessons, examples, and exercises are original, and this course is not affiliated with Google.
Where to Go Next
- AI Fluency for Claude Code applies the same 4Ds to Anthropic’s agentic coding tool. Comparing two agents’ design choices sharpens your judgment about both.
- Discernment: Judging AI Output goes deeper on evaluating AI work.
- AI Fluency for Leaders & Teams covers turning individual habits into team practice.