15 projects
A vendor-neutral reference architecture for building and operating AI agent systems - the products built around persistent, tool-using agents that keep a durable workspace and accumulate real work over time. Published openly as a working draft, for collaboration.
Meta-agent that scaffolds, reviews, and simulates multi-agent project workspaces
A possible route to general intelligence in which homeostatic imperatives drive behavior: many self-improving agents, each keeping one variable within bounds, with no central controller and a structure of motivation that grows instead of being designed.
A call stack for AI agents - `fork()` + `join()` semantics for long-horizon LLM workflows.
Decision-intelligence infrastructure: LLMs as local judges, deterministic Bayesian search on top.
A line of native macOS apps with practical AI built in.
AI-powered macOS window manager - Magnet/Rectangle competitor with Claude-driven smart layouts.
A short e-book coining **Harness Programming** - the discipline of composing the natural-language and configuration artifacts (prompts, skills, hooks, tools, memory) that shape how an AI agent behaves.
Collective intelligence platform for organizations - "OpenClaw for Enterprise"
Outside-in web observer for Claude Code sessions and callstack call trees.
Agent-native Figma-like design surface.
Make Python async functions durable and resumable with a minimal, pluggable runtime.
Subproject of Playbooks: published VSCode extension that debugs natural-language Playbooks programs via the Debug Adapter Protocol.
AI life-coaching product (paused) - Rails + Hotwire chat app with progressive conversation compaction and a Person/Sage domain model.
Flagship: a semantic programming system for AI agents - Markdown-as-source, compiler-to-IR, durable runtime, VSCode debugger. Three years on one thesis: natural language as a programming language.