Skip to main content

ADW Track

The Agentic Developer Workflow (ADW) turns a task or spec into an implemented, PR'd feature — run by Claude Code agents, always gated by a human. This track takes you from zero to running that loop yourself, then onboarding your own project onto it.

What you'll learn​

By the end of this track you will be able to:

  • Run Claude Code locally on your own subscription, unmetered.
  • Ground an agent in a project's real architecture using lat.md instead of letting it guess from stale comments.
  • Write a spec good enough to survive an AI-architect critique before any code gets written.
  • Run the real local ADW loop — classify, plan, review, implement, maker-checker, commit, draft PR — and understand exactly where the human gate sits.
  • Read the evidence (RESULTS.md, validation artifacts) that gates promotion.
  • Choose between the two cloud execution models when a task outgrows your laptop.
  • Onboard a project onto ADW end to end and ship a human-merged PR that an agent built.

The arc: local-first, then cloud​

Every team should start local. ops digest runner --local dispatches work to the Claude Code CLI running on your own Claude Code subscription — flat-rate, unmetered, and fast to iterate on. Only once a task genuinely needs to run unattended, cross-repo, or at scale does it make sense to reach for one of the two cloud execution models (a central ADO pipeline, or repo-native GitHub Actions). This track mirrors that order exactly: you learn and practice the loop locally for the first several modules, and only scale out to the cloud once you understand the mechanics cold.

This document series covers the first half — Modules 0 through 4, the local foundation. Modules 5 through 8 and the capstone continue the same arc into evidence, cloud scaling, onboarding, and governance.

Prerequisites​

  • A working knowledge of git and pull requests — you don't need to be an expert, but you should know what a branch, commit, and PR are.
  • Comfort with a terminal. Every lab in this track is done from the command line.
  • A Claude Code subscription (Module 1 walks through installing it) — the labs in Modules 1–4 do not require any metered API access.
  • Access to a project repository you can experiment in. If you don't have one yet, Module 1's lab works against any small git repo, including a throwaway one.

How the track is gated​

The ADW Track is authored as a linear, gated sequence: each module builds directly on the mechanics of the one before it, so work through them in order the first time. Every module ends with a Hands-on lab and a Knowledge check quiz — your lesson completions and quiz attempts are recorded so you (and, for cohort rollouts, your manager) can see progress. If the LMS backend isn't reachable, progress falls back to your browser so the content still works standalone.

Already using Claude Code day to day?

You can skim Modules 0–1 quickly, but don't skip Module 2 (lat.md) or Module 4 (the real loop with the maker-checker gate and draft-PR merge) — those are the mechanics that make ADW trustworthy, not just fast.

Where to go for a quick reference​

This track is the full course. If you just need a fast refresher once you've completed it, the Company Docs has a condensed quick-start:

Ready? Start with Module 0 — Orientation.