An Agent in 100 Lines of Lisp, or How my Prof was Right - Just 25 Years Early
Summary
An in-depth exploration of building an AI agent loop in Lisp, arguing that Lisp’s homoiconicity enables runtime capabilities by treating code as data. The post demonstrates a minimal 8-line agent-loop in Common Lisp, discusses memory persistence with JSON-like structures, and shows how a single eval-based tool can enable live code generation and web search, challenging typical fixed tool catalogs. The piece also reflects on the historical arc of Symbolic AI and what it means for modern AI agent design, including caveats about eval safety and sandboxing.