At work

Senior SWE at Meta. My motivating factor across teams has been the same: building developer experiences and tooling that people didn't think were possible. Two examples I keep coming back to are autonomous vision for AI agents — built to enable agentic debugging and testing of Horizon Worlds on fully on-demand Linux servers, exposed to agents as a tool they invoke when they decide they need to see the running app, on top of an earlier project where I ported Horizon Worlds to those Linux servers and gave developers a new workflow that streams the running app back to their local machine — and C++ Hot Reload on Windows for Horizon Worlds and the Meta Horizon Studio, integrated via Live++.

Right now I'm focused on Meta's internal AI harnesses — a multi-agent orchestrator and Devmate, Meta's developer-assistant platform. The technical problems I'm digging into are adding back pressure to the agentic loop, enabling ultra-long-running autonomous work (a next-level Ralph loop, along the lines of the /goal command in OpenAI's Codex harness), and shifting error detection left so failures get caught inside the harness instead of in CI or in a human reviewer's queue.

The back-pressure piece is the one I'm most active on. A lot of harnesses already build context iteratively from past sessions — auto memory and the /dream command in Claude Code, the session-context layer in the Hermes agent harness. What I'm working on is a harness that also generates stop hooks and pre/post-tool-use hooks from past sessions, so the agent has executable guardrails preventing it from repeating known mistakes. The shape I want at the end is a deterministic harness with a short iteration loop.

I dogfood the harnesses heavily — last I checked I'm a 99th-percentile token user across Meta.

Career context at Meta: I was headhunted internally twice. First out of boot camp into Workrooms VR, Product Infra — DEVX work on what would become the unified Meta Horizon Engine. Then from Workrooms VR into Horizon Infrastructure and Tooling, to continue that DEVX work on the (then-new) Meta Horizon Engine. I founded the Product Infra group on the Horizon side and still own that surface — build systems, developer tools, and the AI tooling Horizon's engineers use day-to-day.

What I'm thinking about outside work

The same problem space, with different constraints. Outside Meta I read other people's attempts at agentic systems and take notes — autonomous "AI CEO" agents that ship companies end-to-end, review-surface UIs for fleets of agents, platforms that operate businesses 24/7 with no human employees. Most of it confirms that determinism, context-sharing, and short feedback loops are the hard parts.

What I'm reading

Mostly the Anthropic safety canon:

• Dario Amodei's essays — Machines of Loving Grace, The Adolescence of Technology, Surgical Intervention, Country of Geniuses in a Datacenter.
• The Responsible Scaling Policy and the technical reports underneath it.
• Constitutional AI, Mechanistic Interpretability, Alignment Faking, Instrumental Convergence.
• The economics around AI: Jevons Paradox, Comparative Advantage, Lump-of-Labor Fallacy, Goodhart's Law, Socialist Calculation Problem.
• On capability scaling: Sutton's Bitter Lesson, The Scaling Hypothesis, The Compressed 21st Century, Five Limiting Factors, Marginal Returns to Intelligence.