Run an agent-driven delivery line that earns its autonomy one gated lane at a time.
Twenty-seven lessons and three projects on running coding agents as a delivery line instead of a pile of sessions: queues and gates, plan-the-release slicing, autonomy set per lane, an independent reviewer that never holds the merge key, holdouts with exact figures, protected paths, risk lanes, deterministic nodes around agent nodes, readiness scoring, staging and rollback, a registry and identity layer, outcome metrics, and the public failure record. The last module studies ten published factories side by side.
Take these courses first
- Agent Fleets. A factory is a fleet with a merge policy. This course assumes you can already contain, isolate and review several agents, and adds the delivery line, the gates and the wrapper around them.
- Loop Engineering. The gates in this course lean on a locked scorer and a ratchet. Loop Engineering teaches how to keep the measuring stick honest before a line runs unattended.
You finish holding something.
You leave with a factory charter for one real repository: which lanes the line may run unattended, which gates earn that, who owns each approval and rollback, and the scoreboard that tells you when to pause a lane.
The artifact
A factory/ directory for your own repo: line map, release slice, autonomy map, review gate, holdout set, protected paths, risk lanes, blueprint, readiness score, release path, registry, identity ledger, context map, scoreboard, incident lessons, and a charter that ties them together.
The loop you learn
Slice → Gate → Run → Wrap → Prove
Every lesson has the same shape.
You know what you are buying before you buy it.
01The failure
A specific factory that broke or stalled, with the symptom and what it cost, before any term is introduced.
02The rule
The smallest rule with its condition, and what it costs to follow.
03The build
The file in your factory/ directory, in full, inline, with a validator that fails on empty cells.
04The limit
What the evidence does not show, labelled self-reported, secondhand, vendor-hosted or contested.
Why it is built this way.
Three decisions that shaped every lesson.
A queue without a gate is a pile
Throughput only counts where a named approver and a written check sit between stages. Add the gate before you add the agent.
Autonomy is earned per lane, not granted per company
A lane runs unattended when a check that can say no and a revert that can say undo both exist. Every lane without them stays human-led.
Published numbers are self-reported until someone else measures them
No company publishes merge rate, revert rate or cost per merged PR with a denominator. Copy the mechanic, measure your own numbers.
6 modules, 30 entries.
The whole path is public. Nothing is hidden until you pay — only the bodies are gated.
MOD 01Slice the release
Draw the delivery line as queues and gates, plan the release as the unit of work, and set how much autonomy each lane earns, including when not to go dark.
- 01A Factory Is Queues and Gates, Not a Smarter AgentRead · 22m · concise path · 4 quiz · free · updated 2026-10-28
You end with: factory/line-map.md — your delivery line as stages, queues, gates and the named approver for each gate
- 02Plan the Release, Not the TicketRead · 24m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/release-slice.md — one release split into issues with a dependency map, a size cap per PR and a plan sign-off line
- 03Set Autonomy Per Lane, and Know When Not to Go DarkRead · 22m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/autonomy-map.md — each lane of your repo with its autonomy level, the check that earns it and the revert path
MOD 02Gate the merge
Decide when an unattended change may merge: an independent reviewer that never holds the merge key, holdouts with exact figures, protected paths, and risk lanes that say when a human must read it.
- 04The Author Never Holds the Merge KeyRead · 28m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/review-gate.md — reviewer config read from the base branch, round cap, finding states, and who approves per tier
- 05A Holdout Scenario Must Assert Exact FiguresRead · 26m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/holdout-set.md — 3 hidden scenarios with hand-computed expected figures, where they are stored, and the red control for each
- 06Protect the Files That Govern the AgentRead · 24m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/protected-paths.md — the protected path list with patterns, who may change each, and the server-side rule that enforces it
- 07Risk Lanes Decide When a Human Must Read the ChangeRead · 25m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/risk-lanes.md — three lanes with exact entry tests, the human action for each, the audit sample rate and the revert-rate tripwire
- 08The Merge Gate CharterProject · 19m · updated 2026-10-28
You end with: factory/merge-gate-charter.md — review gate, holdout set, protected paths and risk lanes merged into one charter for one repo, with a red control for each gate
MOD 03Run the line
Wrap agent steps in deterministic nodes with capped retries, score a repo before autonomy goes on, and land work through staging with a rollback owner.
- 09Wrap Agent Steps in Deterministic Nodes and Cap the RetriesRead · 24m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/blueprint.md — your line as a node list marking each node deterministic or agent, the retry cap and the exit for each
- 10Score the Repo Before You Switch Autonomy OnRead · 26m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/readiness-score.md — a tiered checklist scored for one repo, with the tier it unlocks and the proposed-issue rule for a tier drop
- 11Merge to Staging, and the Shipper Owns the RollbackRead · 24m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/release-path.md — merge target, staging validation, canary checks with false-positive tracking, rollback owner and the human release step
- 12The Readiness ScorecardProject · 18m · updated 2026-10-28
You end with: factory/readiness-scorecard.md — a repo scored against the readiness list, with the blueprint nodes and release path attached, and the tier it earns
MOD 04Wrap the fleet
The platform layer around the line: one owner and version per skill, every agent commit traced to a named human, and the context a repo cannot give. Taught as mechanisms, because no source measures their outcomes.
- 13One Owner and One Version per Skill and AgentRead · 25m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/registry.md — every skill and agent role with owner, version, status, lifecycle and the distinct-version count
- 14Every Agent Commit Traces Back to a Named HumanRead · 23m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/identity-ledger.md — each agent identity with its sponsor, scope, branch rights and the rule that stops self-approval
- 15Give the Line the Context the Repo CannotRead · 22m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/context-map.md — each context fact the line needs, where it lives, who keeps it fresh and which gate reads it
MOD 05Prove it
Measure outcomes, not the share of AI-written code, turn the public failure record into your gate tests, and assemble one factory charter for a real repository.
- 16Measure Outcomes, Not the Share of AI-Written CodeRead · 25m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/scoreboard.md — six metrics with definition, denominator, guardrail pairing and the tripwire that pauses a lane
- 17Turn the Failure Record Into Your Gate TestsRead · 27m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/incident-lessons.md — five incidents each turned into a gate test, the control that blocks it and its red control
- 18The Factory Charter and ScoreboardProject · 14m · updated 2026-10-28
You end with: factory/charter.md — one repo's complete factory charter: line map, lanes, gates, readiness tier, identities, context map, scoreboard and the incident tests, with the single lane you will switch on first
MOD 06Inside the factories
How ten companies run their factories, side by side: what each publishes, what each leaves out, what you can adopt for a solo, small-team, enterprise or regulated setting, and what every one of them shares.
- 19Anthropic: Review and Risk Routing Are the FactoryRead · 27m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/anthropic.md — the mechanics you adopt from the Anthropic flow, mapped to your lanes, with the one you refuse and why
- 20OpenAI: The Ticket Board Is the Control PlaneRead · 22m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/openai.md — ticket states, workpad and land-step rules adapted to your tracker
- 21Stripe: Deterministic Nodes Around Agent NodesRead · 24m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/stripe.md — your line's blueprint with deterministic and agent nodes marked, copied from the Stripe shape
- 22Spotify: One Prompt, Many Repos, and a Judge That Was RetiredRead · 25m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/spotify.md — a fleet-change plan with verifier, stop hook and your own veto rule
- 23Ramp: The Requester's Token Opens the PRRead · 24m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/adopt/ramp.md — an identity and approval rule so nobody approves their own agent's change
- 24Shopify: An Agent That Works in Public ThreadsRead · 23m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/shopify.md — a visibility and proof-of-work rule for your agent threads
- 25Uber: Cost Per Session Down, Cost Per Merged PR Still UnpublishedRead · 23m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/uber.md — an evidence-table for reviewers and a cost metric plan
- 26Cloudflare: Review as a Tiered Coordinator With SpecialistsRead · 27m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/adopt/cloudflare.md — risk tiers that decide how many reviewer agents run, with a cost ceiling
- 27Coinbase: A Plan Gate Turns One Giant PR Into Many Small OnesRead · 21m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/adopt/coinbase.md — a plan sign-off and PR size rule
- 28Intercom: The Only Published Case of Merging Without a Human ReviewerRead · 28m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/intercom.md — an eligibility rule for no-human approval with size limits, tripwires and a sample audit
- 29Seven Smaller Published Factories and What Each AddsRead · 27m · concise path · 5 quiz · updated 2026-10-28
You end with: factory/adopt/small-flows.md — one borrowed mechanic per company, each with a test
- 30What Every Published Factory Shares, and What None PublishRead · 26m · concise path · 4 quiz · updated 2026-10-28
You end with: factory/adopt/comparison-board.md — a seven-dimension comparison of the factories you studied, plus the pattern you copy and the gap you measure yourself.
Start with the free lesson.
Engineers and tech leads who already run coding agents on real repositories and need to decide how much of the delivery line may run without a human reading the change.
- One-time price, lifetime updates
- Every course included — this one and the next
- 30-day money-back guarantee
- No card required to start