Skip to content

AgentHerder · Course · Week 6

Sustained 12-stream practice + Certification path

The graduation week.

No new techniques. The week is for landing what you've installed across the previous five — into a real, sustained, three-consecutive-days demonstration of flying-stage practice on real work. Then a clear-eyed look at the Cert exam: what it tests, how to know if you're ready, what passing means, what "not yet" means, and what to do either way.

Why Week 6 is just one big assignment

The previous five weeks were each one shift to install. Week 6 is the integration test: can you actually do this, day after day, in real production conditions, without the practice falling apart?

A flying-stage operator who can sustain 12+ parallel Claude Code sessions for one heroic day is interesting. A flying-stage operator who can sustain 12+ sessions every working day, week after week, without burning out, while shipping real work the team trusts — that's the practice.

Week 6 is one assignment, executed across three consecutive working days. The size is deliberate. A single-day push doesn't reveal whether the substrate (Weeks 1–5) is solid. Three days does.

The assignment — three consecutive days at 12+

Do this — Assignment 1 of 1

Run ≥12 parallel Claude Code sessions for three consecutive full working days, while shipping real work.

  • Plan the three days in advance. Identify enough real work-streams that you'll have 12+ legitimate parallel streams to run, every day, across the three days. Mix of feature work, bug fixes, refactors, exploration. The streams should pass the "would you have run this stream anyway" test.
  • Use cctabs as the management layer. Name every tab. Use worktrees for branch-divergent streams. Run the Team Ops sweep skill (Week 4) at least once per hour to surface what needs your attention.
  • Use the brain-fry response patterns (Week 5) when fatigue hits. Don't push through. The discipline this week is sustainability, not heroics.
  • Ship real work. By end of the three days, you should have a count of: PRs opened, PRs merged, streams that consolidated or got closed, integrations or skills that landed. Track the numbers.
  • Screen-record at least one of the three days for self-review. Watch it at 4–8× the next morning. Look for the brain-fry moments, the unnecessary context-switches, the streams that surfaced but you didn't notice for too long. Compare against your Week 2 recording.
  • By end of Day 3: write a short retro. What landed, what didn't, what your current sustained ceiling is.

Time estimate: three full working days, integrated into your normal week.

What "real work" means

Not made-up streams to hit a count. Not the same task split into 12 tabs to look like a fleet. Twelve genuine independent work-streams whose existence is justified by the work that needed doing this week. If you can't construct 12 legitimate streams from your real backlog, that itself is information — your work surface may not be fleet-shaped right now, and that's a real constraint, not a failure of the course.

What "shipping real work" means

By end of Day 3: actual PRs against actual repos. Actual changes merged or near-merged. Actual integration with the team's normal review and ship process. If the three days produced nothing committed to main, the fleet was theatre.

What to do if you can't hit 12

Several options, all legitimate:

  • Run the three days at your current sustained ceiling (8, 9, 10, whatever it is) and note the gap. The Cert exam target is 10+, so 10 is a meaningful threshold.
  • Run the assignment as written but accept that some days will be lower than 12 if the work doesn't support it. Honest data is better than padded counts.
  • Defer the assignment by 2–4 weeks while you consolidate Weeks 1–5 work, then come back. The course doesn't time-expire. The practice is the goal; the calendar isn't.

What the AgentHerder Certification exam tests

The exam is separate from this course. This section is so you can decide whether to pursue it.

Format

~3 hours. Live observation by a Principal AgentHerder (currently Fred personally for the v1 cohorts). You operate the exam fleet against a provided test repo, delivering specified outcomes within the window. The exam is taken in your normal working environment with your normal tooling.

What's evaluated

  • Stream count sustained. Can you actually hold 10+ productive parallel sessions across three hours? Not 10+ peak — 10+ sustained, with the work moving forward in all of them.
  • Output quality across streams. Are the agents shipping work that holds up to review, or noise? The reviewer looks at the PRs that landed during the exam window.
  • Coordination hygiene. How do you decide when to bring a stream to the foreground for QA vs let it cook? How do you handle cross-stream dependencies? How clean is your fleet-state at the end?
  • Decision-making. When agents surface choices, what's your decision speed and quality? When you have to triage, what's your prioritisation rationale?
  • Recovery. How do you handle the inevitable incidents: a quota window cap, an agent going off-course, a stream getting stuck, brain-fry creeping in mid-exam. The exam includes one or two of these by design — they're part of what's tested.

Outcomes

  • Pass = Certified AgentHerder. Listable on LinkedIn, on agentherder.com's practitioner directory, recognized across the AgentHerder community. The credential is durable; it doesn't expire on a schedule (though if the practice changes substantially in future, the cert version is noted).
  • Not-yet = Apprentice. You get specific written feedback on which evaluation dimensions weren't yet at cert-level. Includes one free re-take within 6 months. No stigma — Apprentice is a stage, not a verdict.
  • Practice-tested, not knowledge-tested. The exam doesn't ask trivia about Claude features. It observes you doing the work.

Pricing + how to take it

The Cert exam is bundled into the paid AgentHerder Bootcamp at the bundled price. It's also sold separately for engineers who think they're already cert-ready without the prep program — separate pricing applies.

This free course teaches the prep. Whether you sit the exam is a separate decision.

How to know if you're cert-ready

Honest self-assessment based on the past six weeks.

Strong indicators you're ready

  • You completed the three-day 12+ assignment, sustained, shipping real work.
  • The Team Ops sweep skill (or your equivalent) is reflexive — you don't think about running it; you just do.
  • Brain-fry happens but you handle it in real-time (consolidate streams, take breaks, switch to single-thread work) without losing fleet coherence.
  • The "Claude says it can't, push back" routine (Week 3) is reflexive on browser-automation flows.
  • You let streams cook for 30–60 minutes without checking, and they're usually fine when you come back.
  • Your written personal-experience post from Week 5 named a hard moment you got through, not one you're still in.

Indicators you're not yet ready

  • You can hit 10+ for a heroic day but it leaves you wiped for two days after.
  • You're still trying to supervise streams the running-stage way.
  • Brain-fry hits and your default response is pushing through rather than responding.
  • A meaningful fraction of your fleet ships work the team doesn't trust on review.
  • You're avoiding the "you're cheating" / skeptical-manager conversations rather than having stances ready.

What to do if you're not ready

Don't sit the exam yet. Specifically:

  1. Identify which of the indicators above you're weakest on.
  2. Run the corresponding course week again, especially the assignment.
  3. Try the three-day 12+ assignment again in 4–8 weeks.
  4. Consider whether the paid Bootcamp's cohort + coaching is the path — many engineers' last 10% to cert-ready needs an outside eye, not more solo work.

The honest framing: the cert is for engineers who have made flying-stage practice stick. If it hasn't stuck yet, the cert exam isn't going to magically make it stick. The exam is downstream of the practice, not the cause of it.

What comes after the course (and the exam, if you take it)

Three months from now, without explicit reinforcement, the practice drifts. New work patterns creep in. Old running-stage instincts re-assert. The fleet shrinks. This is what happens to most engineers who installed a discipline and then assumed it would maintain itself.

What sustains flying-stage practice past month three:

  • The Friday demo + weekly planning ritual. Even (especially) when you're solo. It's the heartbeat.
  • An ongoing reading-and-practice channel. Positively Fred for new patterns Fred is exploring. The AgentHerder Club (open via the cert waitlist) for the curated drops of new skills, mission file patterns, case studies — bundled into the paid Bootcamp for the first 90 days, then optional continuation.
  • A peer or two who also practise. Flying-stage in isolation drifts faster than flying-stage in a small group. If your team doesn't have other flying-stage practitioners, the cohort path (Bootcamp or Club) substitutes.
  • A repo of your evolving CLAUDE.md, mission files, and skills. Treat it as a working artefact, not a one-time set-up. The practice compounds in the skill library.

Graduation

That's the course.

Six weeks ago you were somewhere on the Running-stage spectrum, operating 1–3 sessions productively, hitting a ceiling. Now you're somewhere on the Flying-stage spectrum — operating 8–15 sessions sustained, with a meta-agent watching the rest, with the political conversations handled, with the brain-fry response patterns rehearsed.

That's a real shift. It compounds if you defend it. It evaporates if you don't.

If you want the Cert exam: the path is the paid AgentHerder Bootcamp (cohort + coaching + exam bundled) or the standalone exam for engineers who don't need the prep program.

If the practice is enough on its own: carry on. The course doesn't time-expire. Re-read what didn't stick. Bring it to next quarter's hires when they onboard.

Welcome to flying-stage.

Stay in the loop

Three months from now, the practice will have drifted unless you reinforce it. The waitlist below covers three things: notification when the next live Bootcamp cohort opens (for if you decide to pursue the Cert exam with cohort + coaching), notification when the Cert exam is generally available, and the occasional "here's a new pattern I'm exploring" note from Fred.

Same waitlist as Bootcamp + Certification. No spam.

AgentHerder — the professional practice. cctabs — the tool. A line of products by Augmented Mind.

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