All chapters
  1. Start hereOne change, from diff to production
  2. Before stage 1The team setup: what changes when you are not alone
  3. Stage 1Code change: catch it while the agent is still editing
  4. Stage 2Commit: what must be true before a commit exists
  5. Stage 3Pull request: arrive already reviewed, with evidence
  6. Stage 4Review: who reviews what, at hundreds a day
  7. Stage 5CI/CD: spend scarce capacity like a budget
  8. Deep diveCI/CD agents: where they live and how they run
  9. Deep diveBuild the agents: pinpoint the change, then fix it
  10. Stage 6When the change is an agent
  11. OperateVolume and cost: what breaks at 300 a day
  12. AdoptAdopt it in steps
  13. ReferenceHook events: which one, with a script for each
  14. ReferenceGlossary
  15. AboutAbout this site
About

About this site

What this site is, what it is not, and how the facts were checked.

This is a knowledge site about running AI coding agents in a real delivery pipeline. It follows one change through the stages from its first diff to production, weighs the approaches at each stage, and ends with what the whole pipeline costs to operate at volume and the order in which to adopt it. It is written for engineering managers and platform engineers, for the hard case: high pull-request volume and tests that need a complete running copy of the system.

What it is and is not#

It isIt is not
A design: patterns assembled from documented product features and standard CI practiceA record of any real organization
Approaches with drawbacks, so a team can chooseA best-practices list
One fictional example, followed through every stageA benchmark; the numbers are illustrative unless a source is linked
Product behavior checked against vendor documentation on 15 November 2026A promise that product behavior will not change; verify before you build

What is original here and what is not#

IdeaStatus
Hooks, permission rules, headless runs, the GitHub Action, managed code reviewDocumented Claude Code features; some other coding agents have equivalents
Merge queues, batching, test selection, lanes by path, code ownersStandard practice
The lane check and the policy app that approves green-lane changesDesigns proposed on this site, built on documented GitHub features
The review receipt, local and attestedA pattern proposed on this site; the attested form follows build-provenance practice
Triage, then controlled re-runs by a verifier with no model, then revert or fixA pattern proposed on this site
The snapshot-backed stand-in tool server, diff-selected smoke sets, shadow casesPatterns proposed on this site, built from known evaluation practice

How the facts were checked#

Kind of claimHow it was checked
Product behavior (Claude Code, git, GitHub)Checked against the vendor documentation on 15 November 2026
PatternsDesigns, not product features, and labelled as such on each page
Scripts and snippetsSyntax-checked. Several were run on sample data. None was run against a production pipeline.
NumbersIllustrative unless a source is linked, including every time, count and duration in the running example

How it was built#

StepWhat happened
Static sitePlain HTML generated from markdown and JSON; no accounts, no sign-in
Spec firstThe stages, the example and the terms were fixed before the pages were written
Multi-model reviewThe pages were reviewed by several AI models for accuracy and gaps

Who wrote it#

Written by Sam Muthu, engineer and architect.

Sibling sites#

SiteWhat it covers
ask-to-release.sammuthu.comThe stages before the code change: from the business ask to the stories
enable.sammuthu.comAdopting AI in an organization, role by role
claude.sammuthu.comAn architect's study guide to building with Claude