· agents
Agent Architectures: A Map
A classification system for understanding agent architectures based on who initiates action and when the process terminates. Four modes: Copilot, Daemon, Subagent, Workflow.
Blog
Writing about agent architecture, Solana engineering, and building in the open.
· agents
Your Multi-Agent System Needs a Kernel, Not a Bigger Brain
LLMs deadlock 95% of the time in simultaneous coordination. Communication makes it worse. The fix is not more intelligence — it's a 50-year-old CS primitive.
· agents
Your Agent Is Not Stupid. Your Harness Is.
How scaffold engineering delivers 10x more improvement than model upgrades — and why 2026 is the year it becomes obvious.
· agents
Safe Outputs: The Most Important Pattern in GitHub's Agentic Workflows
Everyone's talking about 'Continuous AI.' Nobody's talking about the write-ahead log for agent actions that makes it safe.