— · Changelog
Changelog.
| Date | What changed | Summary |
|---|---|---|
| Introduction — When the Role Stops Fitting | Added 2026 survey data on role overlap between designers, developers, and product managers, with sources. | |
| The Screen Is No Longer the Whole Product | New section on designing interfaces that render inside other AI products, sourced to the MCP Apps specification. | |
| The Product Is Beneath the Interface | Autonomy added as the ninth layer beneath the interface, with a levels-of-automation source. | |
| From Deterministic Software to Negotiated Software | The Negotiated Software Loop gains two steps: delegation and report back. | |
| Trust Is a Product Layer | EU AI Act Article 50 as the legal floor, plus action history and undo by reversible, compensable, or final effect. | |
| Workflow Authors Reshape the Work | New current example: a coding agent bounded by an existing pull request review workflow. | |
| Delegation Is a Design Material | New chapter on autonomy levels, intent preview, audit and undo, and escalation, placed before Trust. | |
| Technical Depth Without Technical Captivity | New section on judging generated code and model output, with DORA and METR sources. | |
| Prototype the Behavior | Added guidance on prototyping with real model output and a first evaluation small enough to read by hand. | |
| The Product Beneath the Interface | Autonomy added; the framework now has nine layers. | |
| Negotiated Software Loop | Delegation and report back added; the loop now has eight steps. | |
| JZ-007 · Two free stamps | In review | |
| JZ-006 · One mistake | v1.0.0 released | |
| JZ-005 · Show the work | v1.0.0 released | |
| JZ-004 · Pretty feels easier | v1.0.0 released | |
| JZ-003 · Line length | In review | |
| JZ-002 · First impressions | v1.0.0 released | |
| The Screen Is No Longer the Whole Product | Historical framing revised, with sources for GroupLens, mixed-initiative interfaces and the Guidelines for Human-AI Interaction. | |
| Designing Behavior, Not Just Interaction | Historical scope, confidence language and cross-modality guidance revised, with a cited source for each. | |
| Taste as Operational Judgment | Scenario marked as illustrative; restraint now reads as a release rule on evidence, reliability, authorization, consequence, and recovery. | |
| Builders Need Strategy Now | Strategy narrowed to the choices made through the build, with positioning, pricing, and go-to-market treated as inputs. | |
| Prototype the Behavior | Evaluation guidance revised to draw on Anthropic's agent evaluation guide and the NIST AI Risk Management Framework. | |
| Build Evidence of a New Identity | Proof now includes evaluation sets, graders, and regression records; the six-month trail is marked as illustrative. | |
| Repositioning Without Pretending | Opening scenario marked as illustrative. | |
| The Practice of Product Authorship | The practice now covers evaluation across repeated trials, regression ownership, and human calibration of subjective graders. |