me² + AI. Production AI agent systems, Model Context Protocol tooling, and multi-agent orchestration, built and shipped in the open by Matthew Snow.
This is the front door. Start here instead of scrolling the repo list.
|
A hands-on AI engineer and systems architect who ships: ST Metro, a Level 5 autonomous build pipeline; CMD, a live multi-agent command center; a second-brain memory layer; and ChartingHero, voice-AI in a regulated (HIPAA) domain. Builder-first IC who tracks the frontier and turns papers into working prototypes. → Jump to what I build · live demos · memyselfplusai.com |
We become your AI team. We translate the internal discipline behind these systems (cost ledgers, watchdogs, staged release, portability) into plain-English wins for non-technical owners who want to delegate real work to agents and trust it. → Read How to Train Your Agent · grab the prompt goodies · book a call |
Proof of work, not promises.
| Built | What it is | Signal |
|---|---|---|
| ST Metro | A Level 5 autonomous software-production pipeline: a multi-agent ecosystem that carries an idea through to a shipped product. | Interactive ecosystem visual with zone drill-down. |
| CMD (Command Center) | The control plane that dispatches real agents to missions and reviews their output through a judge / reasoner loop. | Production multi-agent orchestration, not a demo. |
| Second Brain | A hub-and-spoke knowledge warehouse and shared memory layer. Agents capture signals, recall decisions, and compound what they learn across runs. | Agents that remember and improve. |
| ChartingHero | Voice-AI clinical documentation: speech-to-text + Claude API + EMR tool-calling. | ~70% less documentation time in early deployment, HIPAA-compliant. |
Things you can click right now.
| Demo | Live | What it shows |
|---|---|---|
| ST Metro Visual | m2ai-st-metro.github.io/st-metro-visual | The full ecosystem, interactive with zone drill-down. |
| On the Edge | memyselfplusai.com/edge | The story + series hub. Each card opens a chapter. |
| Fable Ladder | m2ai-portfolio.github.io/fable-ladder | Interactive walkthrough. |
| Engineering Loop | m2ai-portfolio.github.io/engineering-loop-showcase | The compounding-engineering loop, visualized. |
Curated bundles you can install in one step.
| Pack | For | What it is |
|---|---|---|
| m2ai-skills-pack | Claude Code | Portable Claude Code skills and plugins. 100+ skills in one marketplace. |
| claude-desktop-skills | Claude Desktop | 21 production-ready Claude Desktop skills plus a Marketing Agents plugin. |
| m2ai-starters | Any project | Starter templates. Pull one with npx degit m2ai-portfolio/m2ai-starters/<name> ./project. |
 |
 |
 |
 |
| agenttrace Logs the reasoning behind code changes and adds spaced repetition so the intent behind a change is not lost. |
claudeflow Unified AI coding assistant with fallback models and fast research. |
taskflow-mcp-server Durable task management MCP server for long-running agentic workflows. |
mythos-jr Non-initiating defensive cybersecurity worker agent (Claude Agent SDK + A2A). |
A 4-part, no-code series for anyone who keeps meaning to "use AI more." You do not program an agent. You train one, the way you train a sharp new hire: show it your judgment, give it a job with edges, prove it before you trust it, then build a team.
 |
 |
 |
 |
| 1. You're Already Doing the Training Your everyday "keep this, skip that" calls ARE the training data. |
2. Give It a Job With Edges Turn a fuzzy wish into a job an agent can actually run. |
3. Prove It Before You Trust It Trust is built, not given. |
4. Build a Team, Not a Cuttlefish One agent, one job; grow into a roster. |
▶ Read the series: memyselfplusai.com/edge · 🎬 Video walkthroughs: coming soon.
An escalating ladder of copy-paste delegation-prompt patterns, extracted from prompts a frontier model wrote for itself while orchestrating subagents. Each rung is a paste-ready asset, not a lecture. Rungs 1-5 work on any task you hand an AI; 6-8 are for when multiple agents share a workspace.
- Show me the receipt: demand the exact command + output, not a claim of "done."
- Stay in your lane: say what NOT to touch, and who owns it.
- Iterate until green: give the verify command; passing it is the definition of done.
- Warn them about your house: relay your environment's quirks, with the workaround.
- Name the slop: quote the bad output you don't want, instead of asking for "high quality."
- Pin the contract: paste the shared interface verbatim into every worker's prompt.
- Least privilege you can grep: the permission list must be greppable against actual use, both directions.
- The never list: define each agent by what it refuses to do, pointing at the neighbor who owns it.
▶ Get the copy-paste blocks: memyselfplusai.com/edge#prompt-goodies
Want to build your own agents? Join the Early AI-Dopters community on Skool. Every subscription includes a 1:1, agent-specific coaching call: we set up your first trained agent together, start to finish.
▶ Join Early AI-Dopters → get your coaching call
Filter the full org instead of scrolling:
- Showcase products:
topic:showcase - All active repos: m2ai-portfolio repositories
Matthew Snow · memyselfplusai.com · LinkedIn · @MatthewSnow2