Driving the AI SDLC with Shipwright: the whitepaper

Read the whitepaper (free PDF)

Why I wrote it

Harness engineering, loop engineering, agentic engineering, the dark factory: the concepts around AI-built software arrive faster than most of us can place them. I wanted a single place where you can understand them on the basis of a real example, and read them up when you need to.

The real example is Shipwright, the open source harness I build, which runs on Claude Code. Each chapter shows the idea first, then how Shipwright applies it, including where it fell short. So you see how the concepts work in a working setup, not on slides. The paper starts with one small change: I typed what I wanted into the Shipwright Command Center, answered two questions and went on with my day. A spec, tests, updated docs and a merged pull request were waiting when I came back, because the setup around the AI enforced each step.

It is written for people who need to know that what ships is what was decided: product owners, business analysts and requirements engineers; delivery and project leads, testers and QA, IT decision makers, security and assurance roles; and founders and builders who started with vibe coding and now need the result to hold up when they change it next month.

What you can read up

The paper explains the fundamentals and that setup in four layers:

• Baseline: what the product is supposed to do

• Harness: rules the system enforces instead of rules you hope for

• Loop: every change runs through the same checks

• Evidence: results you can read without asking the AI

Along the way it covers the maturity ladder from simple autocomplete up to the dark factory, the Shipwright Command Center with real screenshots, and it closes with ten questions you can ask of any AI-assisted way of building software.

Read it

The whitepaper is free, with no form: Driving the AI SDLC with the Shipwright AI Code Factory