What's New
Updates to the Hands-on AI Playbook, plus notable releases from Anthropic, OpenAI, and the other platforms I follow. Subscribe via RSS.
For long-form articles and deep dives, visit the Graymatter newsletter.
Plugin 8.1.4: the registry maps, Analyze judges
Plugin 8.1.4: Analyze stays strategic and walks your registry's processes; the registry interview starts from what you already have and serves four personas.
Plugin 8.1.2: one clear rule for autonomy
Plugin 8.1.2: Analyze and Design classify autonomy by what decides the next step, and involvement by whether a person takes part while a workflow runs.
Plugin 8.1.1: test inputs saved as files, and a clearer Test round
Plugin 8.1.1: Deconstruct saves your test inputs as files, Claude Code gets an exact start line for each test chat, and the Test skill reads more clearly.
Test, upgraded: run it, then say "test this" (plugin 8.1.0)
Plugin 8.1.0: Deconstruct proposes your test scenarios, every workflow ends with a What I did list, and Test grades the whole run in the chat where it ran.
AI Workflow Framework v8: clearer steps, a report card, and one plugin release
Framework v8: every skill rewritten for clarity — a yes/no report card for testing, Skill or Agent, a draft-then-approve Design Spec, a Run Card, and a glossary.
Build your knowledge graph with one sentence: the building-knowledge-graph skill
A new skill builds your business knowledge graph from an interview: types, their connections, a rulebook, and ingest/lint skills, in the Open Knowledge Format.
Git & GitHub setup, rebuilt as three prerequisites you can check off
Git and GitHub setup is now three numbered prerequisites — GitHub account, Git, GitHub CLI — each with a 'Done when' test; the checklist grows to ten steps.
Skills setup, rebuilt: one plugin for Claude and ChatGPT, native skills everywhere else
One plugin install now covers Claude, ChatGPT, Codex, and Claude Code; Gemini and Copilot load skills natively; the setup page is rebuilt for beginners.
Your AI Registry is now a knowledge bundle
Your AI Registry is now a knowledge bundle: a guided 30-minute setup interview, a visual dashboard that publishes itself, and a starter template repo.
New: Markdown Basics
A new Markdown Basics page for non-technical readers, plus the builder setup rebuild and the Markdown-first AI Registry.
Renamed: AI Workflow Framework
The Business-First AI Framework is now the AI Workflow Framework — same seven steps, same skills, same /handsonai:* commands, just a name that says what it is.
One Plugin to Install: Hands-on AI Marketplace Consolidation
Four plugins are now one: the handsonai plugin bundles the AI Workflow Framework, AI Registry, and Agentic Coding skills, and example agents move to a gallery.
Business-First AI Framework: 7-Step Restructure
The framework expands from three steps to seven — new Test and Improve steps, renamed skills, evaluation criteria, and a skills-only approach replacing prompts.
Outcome-Driven Path and Process Optimization in Deconstruct
Deconstruct skill adds outcome-driven path for agent systems and optimize-for-AI step for step-decomposed workflows
CLI Building Block and Smarter Construct
CLI joins as the 11th agentic building block, and the Construct step now discovers creation tools automatically.
Data Readiness, 11 Building Blocks, and a Redesigned Build Phase
Data readiness probing, 11 building blocks, Design Prompt, and the Build phase split into three focused skills.
Stronger product specs, restructured coding page, and organizational workflows
Upgraded Vision Brief and Feature PRD skills with deeper PM rigor, restructured Agentic Coding page, and added organizational workflow lens
Portable Deconstruct Prompt and Markdown-First Registry Skills
Portable Deconstruct prompt for any AI chat tool and AI Registry skills updated to save markdown files instead of Notion page bodies
Portable Analyze Prompt and Cross-Platform Install Clarity
Portable Analyze prompt for non-skill users and cross-platform skill install consistency across framework pages
Build Phase: Autonomy Levels and Orchestration Mechanisms
The Build phase now uses a two-step decision model — autonomy assessment plus orchestration mechanism — replacing the old execution pattern spectrum.
AI Workflow Design Matrix, Building Block Comparison, and Framework Terminology Update
The new AI Workflow Design Matrix defines six workflow archetypes, a Building Block Comparison Matrix helps you choose, and framework terminology is unified.
Platform Getting Started Pages, Memory Building Block, and More
Every platform now has a Getting Started checklist, Memory joins as the 10th building block, and Product & Engineering gets an Evaluation page.
Plugin Marketplace Moved to Dedicated Repo
The plugin marketplace moved to a dedicated lightweight repo so Cowork, Claude Code, and other tools can clone and install the Hands-on AI plugins faster.
Orchestration patterns, new building blocks, and framework refinements
Agent orchestration patterns, expanded building blocks, agent programming frameworks, and framework refinements.
Step 3 Redesigned: Skills Drive the Build
Step 3 — Build Workflows redesigned around skill-driven orchestration with a new Construct page, Launch Guide, and cross-platform skills standard.
Build Workflows v3.1, Expanded Agents Matrix, and Streamlined Architecture Decisions
Building Workflows v3.1 adds a plan-mode gate, spec approval, and reference specs; the agents matrix gains SDK and browser agents; architecture gets simpler.
Product & Engineering section, ADRs, and platform agent examples
New Product & Engineering section, Architecture Decision Records page, platform-specific agent examples, and fixed agentic-coding slash commands.
New Plugin: Agentic Coding
A new plugin for AI-assisted coding workflows — define requirements with structured PRDs before you build, then hand the spec to your coding agent.
New Example: Content Calendar Planning
See what the Business-First AI Framework actually produces — a complete worked example with all three deliverables for a real content planning workflow.
Agentic Coding Plugin: Stronger PRDs and Vision Briefs
Agentic Coding plugin update: the PRD skill gains paired acceptance criteria and scope sections, and the Vision Brief skill adds ten strategic questions.
Slash Commands for All Plugin Skills
All 10 plugin skills now have slash commands — invoke any skill directly instead of relying on auto-triggering.
Introducing the Changelog and MCP Server
A new Changelog with an RSS feed, plus a dedicated MCP server so Claude, ChatGPT, and other AI assistants can search the playbook directly from a conversation.
I Built an Open-Source AI Playbook for Builders. Here's What's Inside.
Why I built the Hands-on AI Playbook -- a free, open-source resource with a framework, building blocks, and ready-made tools for applied AI.