Skip to content

AgentHerder Course

The fleet-orchestration curriculum — free and self-paced.

Six weeks. The Run → Fly arc, written down. No drip wall. No email-gate. Read at your own pace, work the assignments on your real fleet, install the practice on your own schedule.

Built from running this practice since September 2025 on the 4-engineer team I lead, which merged 844 pull requests out of 892 opened in June 2026, plus my own daily fleet of 8–20 parallel Claude Code sessions. Not theory. From actually running it.

Get notified when the course updates → Free · self-paced · no drip wall · email opt-in only for course-update notifications

How this course works

One week per chapter, six chapters in order. Each week describes a concrete shift in how you operate a fleet — the principles behind it, the assignments to ship on your own real work, and a self-check at the end to gauge whether the week landed.

Read at your own pace. A motivated engineer already at running stage can do a week across a few focused evenings. Most operators want the full calendar week to actually install each shift in their working day. Either is fine.

No drip wall. All six weeks are public. Email signup is opt-in for course-update notifications — it's not a gate.

Prerequisite honesty. This course assumes you're already at running stage — shipping real work with Claude Code regularly, operating 1–3 parallel sessions productively, and you have a real codebase + real work to run the assignments against. If you're not there yet, start with the Augmented Mind Course — the walking-to-running on-ramp — first.

Where the stages actually land. If you want a number to place yourself against, these are the merged pull requests per engineer per month that go with each stage — written up in full here, with the per-PR triage behind them:

  • Crawl — 5 to 10. A traditional competent team with ChatGPT open in a tab and some copy-pasting back and forth. This is where the published industry benchmarks live, elite end around 11.
  • Walk — 10 to 20. Everyone has had their Claude Code moment: an agent editing files in your repo, one session and one task at a time. The published studies of AI-assisted coding measure this stage.
  • Run — 30 to 80. The processes go AI-native. Specs and decisions in markdown in the repo, meetings run for the transcript, the agent owning the git workflow. This course assumes you are here.
  • Fly — 100 to 200 or more. Parallel orchestration as the default way of working. This course is about getting there, and June 2026 says a single engineer can go well past the top of that band.

The six weeks

One week per chapter, in order. Each week describes a concrete shift in how you operate a fleet — the principles behind it, the assignments to ship on your own real work, and a self-check at the end. By Week 6, sustained 12-stream practice + the Cert exam path.

  1. Week 1

    The flying-stage mental shift →

    The 2–3 → 10–15 leap, the "almost never say no" reflex, browser automation as the coordination layer, and comms automation for the routine surface. Practical artifact: a 1-page personal manifesto on what changes about how you work.

  2. Week 2

    Tooling stack →

    The mechanical layer. cctabs, worktrees, foreground-vs-background streams, Anthropic quota management, local-model fallback. Practical artifact: operate 6+ parallel sessions for one full working day, screen-recorded for self-review.

  3. Week 3

    Tighter integrations + the autonomous-agents question →

    The integration thesis. Browser automation as a deep discipline (auth, MFA, broken JS, "Claude says it can't, push back"). Comms automation as a deep discipline. The "fully autonomous agents" question — every cohort asks it — answered explicitly.

  4. Week 4

    Coordination patterns →

    The Team Ops sweep skill. Cross-stream context sharing. PR-nudging automation. Friday demo + weekly planning at flying-stage. Practical artifact: Team Ops sweep skill installed and running on your own work.

  5. Week 5

    The emotional spine →

    Letting go. Brain-fry mitigation. Trusting the structure. The political layer — selling parallelism to a skeptical manager, defending throughput in code review, explaining your work to colleagues who think you're cheating.

  6. Week 6

    Sustained 12-stream practice + Certification path →

    Run ≥12 parallel sessions for one full working day, three days in a row, while shipping real work. Then: what the Certification exam tests, how to know if you're ready to sit it, what passing means and what "not yet" means.

New lessons land roughly weekly until all six are out. Email below to be notified.

Get notified when the course updates

All six weeks are live. I'll keep updating them as the practice evolves and as I learn from engineers installing it — reworked lessons, new bonus material, new patterns surfacing from my own daily fleet, anonymised case studies. Drop your email and I'll send a note when there's a meaningful update.

No spam. One email per meaningful update.

AgentHerder — the professional practice. cctabs — the tool (MIT).