Legacy v1.x Worked Pilots Archive
Appendix
"Old worked examples are not embarrassments. They are the calibration record — the artifact that lets readers see what the framework looked like before, and what changed."
What this is
Three worked pilots from framework v1.x, kept here as a v1.x → v2.0 comparison artifact. Each pilot is a complete walk through one agent system, structured as: scenario → archetype selection → spec → agent instructions → validation suite → (where applicable) post-mortem.
The v2.0.0 release reorganized the book around the five activities (Frame → Specify → Delegate → Validate → Evolve) and introduced three running scenarios that walk one project through every activity in sequence — the in practice chapters at the end of each Part. Those scenarios supersede the v1.x pilots as the primary reading path: they are denser, more concrete, and they thread the same teams and systems through the full lifecycle.
The pilots remain useful for two readers:
- Comparing framings. Two of the v2.0.0 scenarios — Customer-support agent and Coding-agent pipeline — explicitly reference these pilots as their v1.x predecessors. Reading the v1.x pilot and the v2.0.0 scenario side by side shows what the activity-spine reorganization changed.
- Reaching for a canonical example of a single artifact. The pilots' individual chapters (Selecting the Archetypes, Writing the Spec, Validating Outcomes, Post-mortem Through Intent) remain referenced from the pattern index and the references appendix when those artifacts are useful as standalone reading.
The three pilots
Designing an AI Customer Support System
A mid-size retailer deploys a four-agent system to automate Tier 1 customer inquiries. Multi-agent Orchestrator + Executor + Guardian + Advisor composition; full SDD spec for the most complex agent; 14-test acceptance suite; post-mortem on a $0.00-refund incident traced to a specific spec gap.
- Overview
- Selecting the Archetypes
- Writing the Spec
- Agent Instructions
- Validating Outcomes
- Post-mortem Through Intent
Superseded by: Customer-support agent (running scenario) — Frame through Evolve and Operations across 90 days.
A Code Generation Pipeline
A platform engineering team builds a three-agent pipeline that takes a feature intent document and a data schema and produces a complete service scaffold. Synthesizer-Executor-Guardian composition with no live human in the loop; non-conversational instructions for all three agents; 9-test pipeline acceptance suite.
Superseded by: Coding-agent pipeline (running scenario) — Frame through Evolve and Operations across 90 days.
Designing an AI Coding Agent
An in-loop coding agent for an internal repository. Executor with Synthesizer composition, with the explicit decision against Devin-style autonomy recorded; capability-minimalist tool manifest (no general shell, no web fetch, no merge/close); four-level eval stack instantiated against a 75-issue golden set; post-mortem on a deleted-tests incident producing spec v1.1 → v1.2 with a constraint-library entry.
- Overview
- Selecting the Archetypes
- Writing the Spec
- Agent Instructions
- Evals and Acceptance
- Post-mortem Through Intent
No direct v2.0.0 successor; the Coding-agent pipeline scenario covers similar territory in the activity-spine form, and Coding Agents covers the agent-class concept.
Reading guidance
If you are new to the book, do not start here. Start with the Introduction, the Miniature Pilot, or one of the v2.0.0 in practice scenarios. The legacy pilots are a reference resource, not a learning path.
If you are evaluating how the framework matured between v1.x and v2.0, read one v1.x pilot (recommended: Designing an AI Coding Agent) and the matching v2.0.0 scenario back-to-back. The differences you will notice — the explicit five-activity arc, the running-team continuity across phases, the closed-loop emphasis in the Evolve in practice chapters — are what v2.0 added structurally.
The legacy pilots' original front-matter chapter, How to Use These Examples, is preserved for reference.