An Agent in 100 Lines of Lisp: Why My Professor Was Right

An agent in 100 lines of Lisp

I revisited Lisp after 25 years to build an AI agent, discovering that its recursive nature perfectly mirrors agent loops. Instead of pre-defined tools, I gave the model a single eval command, allowing it to write its own functions like web search on the fly. This experiment proves that Lisp's homoiconic design, where code is data, remains the ideal substrate for modern symbolic AI.

"In every agent platform I know of, the tool catalog is fixed at design time; here the catalog is open, and the model decides what it can do."