01 / START WITH THE STORY

17 specialists.
One accountable
delivery system.

Start with the story: Levi Strauss & Co.

At GitHub Universe 2026, Jenny Patchen described a development lifecycle spanning seventeen specialized agents, connected enterprise context and human approval gates.

The published session description follows the back pocket app's lifecycle and a wider enterprise architecture: requirements and specifications, automated testing, security, accessibility and deployment. Five to eleven reviewers examine each PR; proposed repairs do not eliminate human oversight.

Watch the original GitHub story Read the story & workshop brief

Verified against the linked video's publisher description. No internal code, exact role roster or productivity metrics are claimed. Everything we build below is an original reference implementation.

02 / WORKSHOP OVERVIEW

From the story to a system you can inspect

Understand the patternArchitecture, scope and trust boundaries.
Build the referenceActual code, contracts and agent profiles.
Prove the behaviorAnimated flow, live calculations and failure injection.
Deploy and operatePages, separate API, rollback and assessment.

Open the animated workshop →See the actual code Download PPTX

17Opt-in custom agents

5 / 8 / 11Risk-routed reviewers

12Implementation chapters

THE TRUST MODEL

Agents propose. Evidence informs. Humans authorize.

01 / WORK

Bound the change

Acceptance criteria, scoped implementation and independent specialist review at an exact commit.

02 / EVIDENCE

Fail closed

Reject stale, missing, duplicate and contradictory reports. A repair commit starts a fresh evidence cycle.

03 / AUTHORITY

Keep real gates

Protected branches, trusted identities and accountable human approvals belong outside model output.

What runs — and what does not

The repository includes a local API, deterministic evidence lab, seventeen selectable profiles, CI and a working Pages publisher. It does not include an autonomous seventeen-agent scheduler, configured enterprise MCP services or production evidence authentication.

Run the failure-injection labs →

BUILD YOUR UNDERSTANDING

The complete L400 path

01Start with the story, then build 02Reference architecture and state machine 03Seventeen specialists and evidence contracts 04Repository implementation 05MCP integration without accidental authority 06Hands-on labs and failure injection 07Deploy the guide, install agents, run the API 08Trust, approvals and threat modeling 09Operations, cost and incident response 10Troubleshooting and diagnostics 11Facilitation, assessment and adoption 12Sources, attribution and change notes

DEPLOY WITH INTENT

Publish the guide.
Run the service separately.

GitHub Pages hosts static learning content and the deck. The sample API runs on Node.js or in a container, never inside Pages.

Open deployment guide

LOCAL QUICK START

npm ci
npm test
npm run check
npm run build
npm start

Node.js 22+ · synthetic data only

Source-first, not story-shaped claims

The primary video's published description confirms the high-level pattern. This guide makes no unsupported productivity, ROI or defect-reduction claims.

Read the source register and claim ledger →