JetBrains open-sources a Kotlin plugin that turns LLM output into reviewable source code

JetBrains has open-sourced KotlinLLM, a research prototype that lets a large language model write parts of a running Kotlin application's behavior, then saves the result as ordinary Kotlin source code. The repository went live on July 28 under the Apache 2.0 license, with the IntelliJ IDEA plugin (JetBrains' Kotlin/Java IDE), a "Smart macros" API, and example projects, as reported by InfoWorld.

Smart macros are LLM-backed call sites: a function declared in source that the plugin flags for the model to fill in. The JetBrains Research team outlines three design choices that set KotlinLLM apart from generic LLM-at-runtime work. The call site is marked as LLM-backed, so the dependency surfaces in code review. The generated behavior is written to disk as plain Kotlin source, not hidden in prompt state, and can be committed, reviewed, tested, and shipped like any other code. Once generated, the resulting code runs as plain Kotlin, without the plugin or the model in the loop, so covered scenarios add no extra latency or cost.

Most prior work in this space, including the byLLM, nightjar, and Healer projects, targets interpreted languages like Python. KotlinLLM is positioned for a compiled, statically typed language, where type checking, code review, and existing test infrastructure already shape how code ships.

The trade-offs the prototype aims to address remain real. Direct runtime delegation to a model is still slow, non-deterministic, and costly, and KotlinLLM only sidesteps those for scenarios the developer has already covered. Adoption, performance, and independent developer reaction are not yet in the record.