/now
What I'm working on right now
Activity feed: GitHub commits on my 6 production repos, Medium + Substack writing, and ways to book time. It updates each time the site deploys. Inspired by nownownow.com (opens in a new tab).
Shipping right now
Last 14 days of pushes to agent-airlock, agent-audit-kit, proofloop, mnemo, provael, and this site.
No recent pushes (or GitHub API rate-limited the request).
Writing this month
- SubstackNine MCP Servers Put a Model's String Into a Shell. One Was a Gemini CLI Extension (opens in a new tab)
- SubstackMy Smog Map Counted 539 Fires. It Also Says Why That Number Is Low (opens in a new tab)
- Substack296,000 stars, 45 credited clones, under 4,000 stars between them (opens in a new tab)
- SubstackMy Agent Memory Database Has Five Ways to Forget. None of Them Erase the Trail (opens in a new tab)
Talking
Open to podcasts, conference talks, and 1:1 advisory sessions on agentic AI, LLMOps, and AgentOps.
Reading
- Project Glasswing announcement, Anthropic
Claude Mythos is the upstream attacker model. Decides what agent-audit-kit needs to detect next.
- OWASP Agentic Top-10 RC
Mapping current agent-audit-kit coverage against the upcoming RC release.
- Salt Security 1H 2026 report
88% of enterprises had agent-security incidents: agent-airlock's market sizing.
- VLA red-teaming: SmolVLA, π0, OpenVLA papers + LIBERO
Current focus: building Provael, the red-team & assurance layer for physical AI. Mapping which vision-language-action attacks transfer from simulation to real policies.