Simon Willison
Creator of Datasette and LLMNotes from building with LLMs and coding agents, including their security risks, written as he tries each new model and tool.
Building with agents
The program around the model that assembles its context, runs its tool calls and enforces its permissions, as Claude Code and Codex do. Running a real task through one shows where an agent needs more context or tighter limits.
Building with LLMs Various speakers 1 video
Notes from building with LLMs and coding agents, including their security risks, written as he tries each new model and tool.
Posts on how the Grok Bot team ships quickly with agents, and on habits that make agents write better code. Previously on the React Compiler core team and at Cursor, Meta and Netflix.