M2AI helps teams adopt AI and automation without lowering their standards. We begin with one bounded workflow, preserve human judgment, measure the result, and expand only after the system earns trust.
Built and taught in the open by Matthew Snow.
Last updated: August 2026
- Team AI enablement — practical workshops built around three disciplines: map and bound the automation, define and verify quality, and keep people accountable for judgment.
- Safe first workflows — helping teams start with one visible, manageable use case before committing to a larger system or an agent.
- Agent infrastructure — building and testing the orchestration, memory, evaluation, safeguards, and reusable skills required when the work genuinely needs agents.
The current operating principle is simple: start with a workshop, continue if it earns it.
- Learn the approach — practical AI education for teams, from frontier applications to bounded automations.
- Train your judgment — a plain-English series for giving AI a job with edges and proving it before trusting it.
- Explore the open-source tools — reusable skills, plugins, templates, and agent tooling.
- See the proof — working systems and live demonstrations, not slideware.
Use AI to improve work people already understand. Choose a workflow with visible value, clear boundaries, and manageable risk.
Teach the people closest to the work how to define quality, exercise judgment, and supervise the system rather than becoming dependent on it.
Implement the smallest system that can produce a measurable result. Keep human approval where consequences, ambiguity, or risk require it.
Add automation or agents one workflow at a time, only after the previous system has demonstrated reliability and adoption.
| Project | What it demonstrates | Explore |
|---|---|---|
| ST Metro | A multi-agent software-production ecosystem that can carry an idea through planning, implementation, review, and delivery. It demonstrates the advanced end of the M2AI spectrum—not the default starting point for every workflow. | Interactive ecosystem visual |
| ChartingHero | Voice-AI clinical documentation combining speech-to-text, model reasoning, and EMR tool use in a regulated environment. Early deployment reduced documentation time by approximately 70%. | Read the proof-of-work overview |
| M2AI Skills Pack | More than 100 reusable Claude Code skills and plugins, packaged so proven operating knowledge can move between projects instead of being rebuilt each time. | Browse the skills pack |
- ST Metro Visual — explore the full ecosystem with zone-level drill-down.
- Fable Ladder — an interactive walkthrough of staged progression.
- Engineering Loop — the compounding engineering loop, visualized.
A four-part, no-code path for turning everyday judgment into bounded, testable agent behavior:
- You're Already Doing the Training — your everyday decisions are examples of your judgment.
- Give It a Job With Edges — turn a fuzzy wish into a job the system can actually run.
- Prove It Before You Trust It — widen autonomy only after verification.
- Build a Team, Not a Cuttlefish — give each agent one bounded responsibility.
Twelve plain-English, screen-recorded walkthroughs from a new AWS account to calling Claude through Amazon Bedrock.
Copy-ready delegation patterns for demanding evidence, defining ownership, constraining scope, and verifying that work is actually complete.
| Project | Use it for |
|---|---|
| m2ai-skills-pack | Portable Claude Code skills and plugins. |
| claude-desktop-skills | Production-ready skills and a marketing-agents plugin for Claude Desktop. |
| m2ai-starters | Starter templates for new projects. |
| card-fanout | Turn a bounded work specification into guarded parallel agent execution. |
Recently active: view repositories by latest update
Every engagement starts with a short diagnostic and a practical workshop using your team's real workflows. You leave with a bounded use case, explicit quality criteria, clear human-accountability points, and a 30-day application plan.
Teams can continue into a capability sprint, train-the-trainer program, or advisory engagement—but only when the first step demonstrates enough value to justify the next one.
Matthew Snow is the founder of M2AI. He builds applied AI systems, teaches teams how to use them responsibly, and publishes the underlying tools and operating lessons in the open.