Clojure in the Age of Language Models

This argues that as generative models make low-level coding trivial, the valuable developer skill shifts to recognizing problem shapes and expressing solutions at a higher level. Clojure’s live REPL, immutability and emphasis on small pure functions let developers and agents iterate inside a running process - redefining functions without restarts, preserving application state, and testing pieces in isolation - so changes can be validated immediately instead of enduring long compile/startup and state-reconstruction cycles. Agents benefit from observability and the ability to load code into an active system, which reduces iteration count and makes working with large codebases tractable.

Clojure’s homoiconic S-expression syntax and powerful macro system make code manipulable as data, simplifying domain-specific abstractions and machine understanding. Uniform data structures and declarative function composition produce concise code, lowering token costs and keeping more of the call graph in an LLM’s context window, improving automated edits. JVM startup and perceived complexity remain barriers, and Jolt is presented as a response: a single-binary compiler with built-in tooling, FFI and standalone distribution to remove friction. Overall, Clojure’s combination of live programming, data orientation and brevity is presented as well-suited to the constraints and strengths of language models.