Now

Updated march 2026

Living in San Francisco. Working at Writer on Cerebro. Running EvalOps on the side.

Shipping

Cerebro — operations data platform for cloud and SaaS security posture management
DiffScope — AI code review engine with confidence scoring

Reading

Towards a Science of AI Agent Reliability — formal taxonomy and metric suite for agent reliability
Policy Compiler for Secure Agentic Systems — policy compliance from 48% to 93% across frontier models
Red-Teaming LLM Multi-Agent Systems via Communication Attacks — agent-in-the-middle attacks on inter-agent messages
Agentic AI Security: Threats, Defenses, Evaluation — threat taxonomy and defense strategies for agentic systems

Thinking about

How to make agents self-correct without infinite loops
Patterns for secure tool execution that don't cripple agent agency
Whether multi-model deliberation is the best path to better reasoning

Inspired by Derek Sivers' now page movement.