Framework Version 1.0 · September 2026

The AI SDLC.

An operating model, control framework, and maturity progression for engineering organizations building with agents.

Start reading → Download PDF ↓

Parts

Sections

References

243

Status

Released

The thesis

Generation has become cheap and elastic. Verification, review, comprehension, and operation have not. Every question of how far to push autonomy resolves to whether the organization has built enough verification and absorption capacity to consume what its agents produce.

That is not a caution. It is a construction specification, and it is unusually well supported.

The organizations that are furthest along did not get there by making agents better. They got there by building the verification substrate, the context substrate, and the review economics that let agent output be absorbed.

Read Section 1 →

The three scales

Everything in the framework is positioned on one of three ordered scales. They are separate axes and they are not interchangeable.

Three progression stages

ASSISTED

DELEGATED

OWNED

An organization cannot safely occupy a stage its verification capacity does not support.

Five autonomy tiers

Tier attaches to a deployment, not to a product.

Five control maturity levels

Overall level is the minimum across dimensions, never the average.

The six parts

44 sections

Part I §1–4

The Operating Model

Why the framework exists, the vocabulary it runs on, what the published record establishes, and the three-stage progression.

4 sections

Part II §5–9

The Mechanics

The unit of agent work, what to delegate, how output is verified, what the platform must provide, and what people do.

5 sections

Part III §10–16

The Lifecycle

The seven stages as they change, each with a Level 2 minimum bar, a Level 3 enforced state, and evidence an auditor can request.

7 sections

Part IV §17–23

The Constraint Layer

The controls, gates, maturity model, metrics, and standards mapping that determine how far the operating model can be pushed.

7 sections

Part V §24–33

Practitioner Playbooks

Ten procedures, each producing a specific artifact — an oracle spec, a work class, a qualification record, a ninety-day program.

10 sections

Part VI §34–42

Case Studies

Eight cases reconstructed entirely from published primary sources, and what they say together.

9 sections

Contents · The AI SDLC

v1.0

Part I — The Operating Model §1–4

01 Purpose and Thesis 8 min 02 Definitions and the Delivery Model 4 min 03 The Evidence Base 10 min 04 The Progression: Assisted, Delegated, Owned 5 min

Part II — The Mechanics §5–9

05 The Agentic Engineering Loop 17 min 06 Autonomous Work Classes 14 min 07 Validation Architecture 13 min 08 The AI Engineering Platform 12 min 09 The Human Operating Model 12 min

Part III — The Lifecycle §10–16

10 Stage One — Planning 6 min 11 Stage Two — Requirements Analysis 5 min 12 Stage Three — Design 7 min 13 Stage Four — Development and Implementation 11 min 14 Stage Five — Testing and Verification 8 min 15 Stage Six — Deployment and Release 7 min 16 Stage Seven — Operations and Maintenance 8 min

Part IV — The Constraint Layer §17–23

17 Cross-Cutting Domains 7 min 18 The Control Maturity Model 5 min 19 The Four Gates 3 min 20 Metrics and Executive Reporting 3 min 21 Tooling Evaluation Criteria 12 min 22 Standards Crosswalk 4 min 23 Open Problems 4 min

Part V — Practitioner Playbooks §24–33

24 Designing an Oracle 7 min 25 Loop Instrumentation 4 min 26 Specifications Agents Can Build From 6 min 27 How to Stand Up a Work Class 6 min 28 The Context Substrate in Practice 5 min 29 The Review Operating Model 5 min 30 Qualifying an Agent for Write Access 5 min 31 Fleet Operations 4 min 32 The First Ninety Days, Worked 4 min 33 Beyond Ninety Days 5 min

Part VI — Case Studies §34–42

34 Backlog Work at Repository Scale 6 min 35 Flaky-Test Repair as a Deployed Work Class 5 min 36 Two Migrations, Two Oracles 8 min 37 Assured Test Generation 6 min 38 Fleet Maintenance at Scale 5 min 39 Review at Volume 4 min 40 Agent Tooling as Attack Surface 6 min 41 An Agent Outside Its Authorized Scope 5 min 42 What the Cases Say Together 10 min

Appendices and References §43–44

43 Appendices 6 min 44 References 32 min

The complete framework, typeset for print.

All 44 sections, the standards crosswalk, the appendices, and 243 references in a single PDF.

Download PDF ↓ Book advisory →