July 29, 2026
Decompiler? I barely know her
Kuna: Decompiler Development in the Age of Coding Agents
A professor says AI wrote almost all of this code, and commenters are calling it the right kind of chaos
TLDR: Kuna is a new code-unpacking tool whose creator says AI wrote almost all of it, and it’s already getting close to a top commercial product on one major test. Commenters are unusually impressed, calling it a smart use of AI and immediately asking the obvious chaotic follow-up: can the bot start naming everything too?
The big gasp in this release is almost too juicy to ignore: Kuna, a new tool that turns machine code back into readable code, was built with an AI doing "nearly every line" of the coding. And somehow, this experimental project is already brushing shoulders with the big-name commercial rival, IDA Pro, at least on one key benchmark. That alone is catnip for the comments section: part applause, part disbelief, part "wait, are we really here already?"
The strongest reaction is surprisingly upbeat. One commenter flat-out called it "a very good example of how to use AI powered development", arguing that this is exactly where AI should shine: helping with painfully hard, valuable work instead of churning out junk. That praise matters, because the author clearly anticipated the eye-rolls and tried to head them off, insisting Kuna is not just slop, but an experiment built on years of real research, older open tools, and lots of expert guidance.
Then came the wish-list energy. Another commenter immediately pushed the drama forward with a very internet-brained next step: if the AI can follow the logic faster than a person, why not let it rename confusing functions and variables too? That suggestion has strong "give the robot even more power" vibes, and you can almost hear the crowd revving up for the sequel. The mood, overall, is less doom and gloom than fascinated chaos: people aren’t just reacting to the tool, they’re reacting to the idea that AI may now be learning from expert software and coming back weirdly competent.
Key Points
- •Zion Leonahenahe Basque released Kuna as an experimental decompiler developed largely with LLM-generated code while working with AFRL and Metalware.
- •The article reports that Kuna achieved perfect control-flow structuring on 44.4% of functions in recent DecBench results, compared with 45.7% for IDA Pro 9.2.
- •Basque says Kuna improved through autonomous refinement, where an LLM studies cases where it underperforms on decompilation metrics and iterates through trial and error.
- •The refinement process involved comparisons with IDA Pro, Ghidra, and angr, and the article says an LLM reimplemented more than 20 fundamental angr features in Kuna.
- •The article states that Kuna is a Rust port of the NSA’s Ghidra adapted to more closely match angr’s pipeline and that meaningful automatic refinement still depends on prior research, metrics, and benchmark design.