Sanjay Shekhar

Building things, writing some of it down.

Posts

• Code execution: a more efficient way for AI agents to use tools (June 29, 2026) — Why sandboxed code execution against a tool API beats both raw MCP tool calling and local code interpreters for unattended agents.
• Stop putting every tool in your agent's context (June 21, 2026) — Why binding every tool to an LLM agent overloads its context, and a dynamic search-and-cache alternative that doesn't.

Projects

• Agent Tool Search — the demo project behind Stop putting every tool in your agent's context.