For leadership ยท How our AI engineering team works

One Captain. Specialist bots.
The owner approves every gate.

The Captain takes every request and hands it to specialist bots, who work in parallel where safe. Work stops at owner approval gates, no bot ever commits code, and all bots share one cache instead of re-reading documents. Click any bot in the map to see what it does.

Feature flowBug flowHuman gateFix loopShared cache

The flowHandoff chain

  1. Captain reads the request and picks the path (a question gets an answer only)
  2. Intake bots read the tracker ticket, the requirements doc, the design and the code once
  3. The owner approves the brief
  4. Gap Finder lists gaps and questions
  5. The owner answers, and the answers go to the planner
  6. Dev Planner writes the plan, test cases and must-haves
  7. The owner approves the plan and confirms the repo
  8. Devs build and write tests, QA writes test cases
  9. Analyzer gate: fail goes back to devs
  10. Review bots check in parallel, comments go back to devs
  11. QA tests on devices: fail means rework, pass means the report
  12. The owner accepts the QA report, then the tracker is updated and Done

Key rulesWhat every bot follows

โœ… Quality firstMust-haves before coding, self-check before handoff.
๐Ÿงช Light TDDTest cases planned, tests written for every new function.
๐Ÿ“Š Code gateGrade B or above, 80%+ coverage on changed code.
๐ŸŽฏ Pixel QAChecked against the design on several device sizes.
๐Ÿ’ฐ Token savingShort summary files, related files only, failures only.
๐Ÿง  Right modelHeavy model only for gaps and planning.
๐Ÿ”„ One rulebookCache Keeper syncs rules across every bot and AI tool.
๐Ÿงฉ Project firstThe project's own rules win, except safety rules.

TrustHow it stays safe and reliable

๐Ÿ“ˆ Metrics (to be measured)
  • Hours saved per ticket TBD
  • Rework loops TBD
  • Token cost per feature TBD
๐Ÿ—‚๏ธ ConcurrencyEach task has its own tasks/<TASK-KEY>/ folder. Shared files are edited by one dev at a time.
โ›” Stuck gatesGates never auto-approve. Work waits for the owner. Captain B only prepares work.
๐Ÿ” Loop metricsAnalyzer fails, PR comments and QA fails are logged in tasks/<KEY>/loops.md.
โšก Small fix routeA typo or one-line fix skips Gap and Plan: dev, analyze, quick review, 1 to 2 gates.
๐Ÿงพ Audit trailtasks/<KEY>/approvals.md records stage, decision, by whom and when.
๐Ÿ’พ CheckpointingEvery stage writes its .md, so any bot can resume from the last finished stage.

FAQQuick answers

What if bots disagree?Captain decides, and the owner has the final say.
Faster, but better?Yes: the code gate, pixel QA and loop metrics keep quality up.
What if a bot is wrong?Every stage has an owner gate plus review.
Is data safe?Test data stays local, no secrets are read, and only the owner does git.

A general explainer of how the bot team works. Metrics will be added once measured.