Real-work exercise
Pick actual work. Run Claude. Ship the result.
At Car-Mart Technology Group: 30 min writing + 5 min grading. Pick one real artifact from this week's queue (PR, spec, dashboard, runbook, ticket, KB article) — not a demo, not a tutorial. Real examples per lane: api-refactor PR #1247 credit-bureau retry-with-jitter, q3-dashboard TX approval-rate 4-point drop, runbook-update escalation-path change, dealer-portal feature spec, security audit finding, help-desk KB article.
"Ship" means a thing exists that did not exist before and you can show it. Not perfect. Not reviewed. A saved prompt + saved output in your Project, side by side, is the receipt. The 7-criterion rubric (real work, 4-part structure, verifiable output, first-run quality, iteration evidence, saved receipt, used the Project) is the DoD check. Pass = 6 of 7. Fail = we know what to fix.
9 directors + CTO Josh are here to see real work from corporate technology teams, in real time. That's the credibility play for AI adoption. The full long-form walkthrough (5-step, 30 min, 7-criterion rubric, 5 failure modes, 3 worked examples) lives at hool.dev/aitraining/real-work (the companion doc, ~5-7 min).
I'm a senior [backend engineer / data analyst / SRE / UX designer / BA / help-desk lead] on the Car-Mart Technology Group [API team / data platform / ops / UX / BA / help-desk] lane (8 years, Python 3.12 + black + ruff, the [api-refactor / q3-dashboard / runbook-update / dealer-portal-feature / team-spec / KB-article] Project in scope). I'm at T2 with Preferences + a Project + CLAUDE.md set. I'm starting the 30-min real-work exercise on [one real artifact from your queue — paste or describe it]. Walk me through this end to end: first, what's the highest-leverage task I should do in the next 30 minutes, and why. Then: do it with me, step by step — show me the prompt you would write (role + context + task + constraints), show me what you'd check on first-run, and show me the one iteration move that closes the gap if the first run is wrong. End with the 7-criterion rubric pass/fail for the prompt you just produced.In your lane
Dev lane: T2 pastes the diff for api-refactor PR #1247 (credit-bureau retry-with-jitter change) into the Project, asks for a risk summary in the PR template. T3 sets Preferences for code-style + error-handling + 'flag any new dependency not in requirements.txt.' T4 uses a CLAUDE.md that has the 3 most-referenced helper modules + the PR template + the on-call rotation. Concrete artifact: open Desktop → Project: api-refactor → new chat → paste the diff → 'compare to our PR template, flag the missing Testing section, list 3 risks' → iterate one constraint at a time → save the final prompt to the Project's prompts/ folder as 'pr-summary-template' → re-run on the next 3 PRs.
Data lane: T2 pastes the SQL for the q3-dashboard TX approval-rate drop (week of 2026-06-01) into the Project, asks for a 5-cause hypothesis list. T3 sets Preferences for tables-first + SQL with comments + 'list 3 edge cases before recommending.' T4 uses a Project with the 2 data dictionaries + the 1 sample stakeholder summary. Concrete artifact: open Desktop → Project: q3-dashboard → new chat → paste the SQL → 'list 5 most likely causes, ranked, with the data needed to confirm or rule out each' → iterate one constraint at a time → save the final prompt to the Project's prompts/ folder as 'sql-hypothesis-template' → re-run on the next 3 investigations.
Ops/Security lane: T2 pastes the new runbook draft into the runbook-update Project, asks for a diff against the team template. T3 sets Preferences for SRE tone + 'lead with severity + hypothesis + next action' + 'flag any escalation-path or on-call-rotation change.' T4 uses a Project with the runbook library + the on-call rotation + the postmortem template. Concrete artifact: open Desktop → Project: runbook-update → new chat → drop in the new runbook → 'compare to the runbook template in CLAUDE.md, flag deltas in escalation path or on-call rotation' → iterate one constraint at a time → save the final prompt to the Project's prompts/ folder as 'runbook-diff-template' → re-run on the next 3 runbook updates.
Walk out of this room with one shipped artifact and one saved prompt that produced it. Save the final prompt + final output side by side in your Project's prompts/ folder (the receipt), post the receipt to the team's Slack #ai-training, and add a one-line note ('the work, the verb, the constraint that mattered most') to your personal prompt library or the 90-day plan export. Then share with the person next to you, or with John, or in the channel. The directors and CTO Josh are in the room to see what your team can ship today.