I'm a lawyer who somehow ended up with a GitHub profile.
Years ago I took a free Python course on Codecademy and built hotseat-mediator, a little dispute resolution tool. That is the extent of my programming ability. I peaked in 2019.
Everything else you see here was built with AI doing the actual work while I pointed and said "no, more like this" repeatedly until something functional emerged. If you look at the code, you'll be able to tell.
I spend a lot of time working with actual engineers, which mostly means I understand a little but not enough to be useful. These repos are what happens when someone like me gets access to tools that let you skip the "learn to code" part and go straight to the "have opinions about software" part.
Some things I've had AI build for me
| Repo | What it is |
|---|---|
| agentproof | Zero-knowledge verification of AI agent policy compliance |
| Agent-Management-System | Applying corporate org charts to AI agents |
| ask-bigger | See the gap between what you ask AI to do and what's possible |
| beautiful-code | A gallery of elegant code snippets to appreciate |
| counselos | In-house legal team tool experiment |
| daylight | Turn any GitHub repo into a workspace for non-developers |
| dcp | Document Context Protocol |
| differential-codebase-privacy | Differential privacy for codebases |
| Digital-Advisory-Board | Customizable digital board of advisors |
| dm-chess | Chess via DM |
| doublecheck | Make it easier for humans to fact-check AI output |
| ecomode-for-code | Putting code in eco mode |
| ezconfig | 125+ deployment vehicles for vibe coders |
| faxmachine | GitHub-powered virtual fax machine |
| flatapi | Turn data files in your repo into a queryable static API |
| gh-dj-pad | CLI DJ sound pad |
| gh-flair | Your repos' highlight reel as a gh CLI extension |
| hockey-shot-tracker | I'm a hockey parent, don't judge me |
| human-mcp | Person-centric MCP server thought experiment |
| instreval | A/B test your AI custom instructions |
| instrucgen | Generate AI custom instructions from a GitHub profile |
| Internet-of-Models | Thought experiment about an Internet of Models |
| itty-bitty-tools | 99 itty bitty tools |
| lawgraph | Legislation as structured data |
| legal-move-tokenization | Taxonomy of strategic moves in legal briefing |
| list-ninja | Generate SharePoint List integration kits for AI coding tools |
| local-app-factory | Package any local web app for reliable relaunch |
| promptapp-runner | An entire app, compressed into a URL |
| quantum-circuit-playground | Quantum circuit playground |
| rendezvous | Have your LLM talk to my LLM |
| repo-resurrector | Breathe new life into stale projects |
| shelf-life | Drop a Goodreads export, get a portrait of your reading life |
| sift | Superfast fingerprinting and overlap detection for any text corpus |
| socrates | No answers, only questions |
| terminal-profile-studio | Interior design for your terminal |
| trace | A format for publishing AI-assisted writing with the prompts that made it |
| trace-collaborate | Multi-stakeholder AI drafting with provenance, extends Trace |
| upstream | An AI tool whose purpose is its own obsolescence |
| who-codes | Find out which people in your LinkedIn network code |
I think there's something to the idea of domain experts using AI to prototype things from their own fields, even if they can't write a for-loop from memory (I can't). A lawyer who has spent fifteen years reading legislation has useful intuitions about how to model it. A hockey parent who has watched four hundred games knows what stats actually matter. AI lets people like us get those ideas out of our heads and into something you can click on, even if the underlying code would make a senior engineer cry.