Building PENSO — corporate governance, implemented for AI organizations. governances.ai

I design and operate AI agent organizations: multi-agent systems where Claude, Codex, and Gemini work as a governed team — not as individual tools.

My focus is AI management, not prompt engineering:

• Prompt engineering → how to talk to an AI
• AI management → how to build systems where AI judgment can be trusted, reviewed, and gradually delegated under human oversight

PENSO — an operating system for AI organizations, built on one idea: governance by structure. Oversight, separation of duties, and accountability are implemented as architectural constraints the system cannot bypass — not as policy PDFs or dashboards bolted on afterwards.

• Agent runtime governance — permission models, execution boundaries, and decision records for agentic systems
• Separation of execution and audit — the component that reviews an action is independent of the one that performs it
• Human-reviewed AI — keeping humans meaningfully in the loop without giving up the speed
• Multi-agent orchestration — routing, role separation, and quality gates across Claude / Codex / Gemini

More on the thesis — Separation of Powers as Governance Architecture — at governances.ai.

Shipping merged code into AI-infra OSS, and contributing to the semantic-convention work that defines how agent runtimes are described — kept implementation-neutral, with producer-owned context left out of scope:

• litellm (LLM gateway, ~50k★) — merged PRs »
• OpenTelemetry GenAI semconv — design input on agent telemetry: decision / outcome attributes and opaque, payload-free governance references for agent decision points; an acting-vs-target agent framing for multi-agent traces; runtime threat-signal correlation
• Agent execution-record proposals — review input on keeping the normative contract in the spec itself, rather than in any single reference implementation, so independent implementations interoperate on equal footing
• Microsoft Agent Governance Toolkit (AGT) — telemetry and observability design discussions
• otel-agent-evidence-sample — a small reference for the opaque correlation_id evidence-linking pattern (MIT)

→ All merged contributions, always current »

• 🎓 Tokyo Institute of Technology — Robotics (graduated top of class)
• 🦅 Human-Powered Aircraft Competition — 1st place, as aircraft architect
• Robotics background, not CS — running a fully AI-native org. The interesting problem isn't can-build vs. can-deploy; it's can-build vs. can-govern.

Daily drivers: Claude, Codex, Gemini, Python, GitHub Actions, Cloud Run

Interested in agent governance, human-AI decision systems, or AI org design? Open an issue in this repository to start a conversation.

I don't publish a public email address on GitHub.