Guide · v1.0 · 8 parts
AI Delivery Guide
A practical path from a useful idea to safe, observable AI-assisted software delivery. Agents can investigate, implement, test, and observe. People still own the outcome, the constraints, and the consequences.
Start with part 1The Delivery Loop
From intent to learning, one part at a time.
01
ReadPart 1 of 8
The Shift from Vibe Coding to Spec-Driven Development
Why AI-assisted delivery needs explicit intent, constraints, and proof instead of a chain of good guesses.
02
ReadPart 2 of 8
Spec-Driven Development in Practice: A Walkthrough with Spec Kit
A practical look at turning a specification into a plan, tasks, and implementation work.
03
ReadPart 3 of 8
From a Vague Request to an AI-Executable Requirement
Turn an ambiguous request into a work packet with examples, constraints, non-goals, and acceptance checks.
04
ReadPart 4 of 8
Write the Design Before You Split the Work
Give agents shared contracts, system boundaries, and decisions before parallel work creates conflicting interpretations.
05
ReadPart 5 of 8
Turn the Design into an Implementation Plan
Create bounded work packages with ownership, dependency order, verification commands, and explicit exclusions.
06
ReadPart 6 of 8
Run the Plan One Work Package at a Time
Run each work package in a clean session with a hard boundary, check the result yourself, and keep the status tracker honest.
07
ReadPart 7 of 8
AI Wrote the Code. Now Prove It.
Map each acceptance criterion to automated checks, browser evidence, review, and a human release decision.
08
ReadPart 8 of 8
The Pull Request Is Not the Finish Line
Define rollout signals, observation windows, rollback paths, and how production learning enters the next change.
What you will learn
Useful artifacts, not agent theatre.
- Turn a vague request into a requirement with examples and non-goals.
- Write the design first, so agents share one set of contracts.
- Split the work into bounded work packages with clear ownership.
- Build one work package at a time, and check each result yourself.
- Prove every change with tests and evidence before release.
- Watch production, and feed what you learn into the next change.
The worked example
One feature, six real artifacts.
The series follows one worked example: a restaurant waitlist app that auto-promotes the next party when a table frees. Each part produces one delivery artifact. Together they form a downloadable pack: requirements, solution design, implementation plan, status tracker, test package, and release package.
Download all six documents (.zip)Coming next
The AI Delivery Playbook.
The guide and its six documents stay free. I am building a fuller package for teams that want to put this into practice. It will include:
- A sample app, with the code a coding agent generated from these documents.
- Claude Code skills, one for each stage of the loop.
- Cursor rules and AGENTS.md templates for the same stages.
- The prompts from the guide as ready-to-use files.
I plan to email it to subscribers by the end of 2026. A follow-up post will run these documents through a coding agent so you can watch the feature get built.
Get the Playbook first
This subscribes you to my newsletter.