A new tool promises to turn plain-English descriptions into working neuro-symbolic AI code, no library expertise required.
The tool, called AgenticDomiKnowS (ADS), is built on top of DomiKnowS, an existing framework that lets developers bolt symbolic rules onto deep learning models through a declarative programming interface. Researchers built ADS to remove the biggest hurdle to using that framework: you had to already be fluent in its specific syntax. ADS instead runs an agentic workflow that builds and tests each program component on its own, then assembles them into a full DomiKnowS program from a free-form task description, with an optional human-in-the-loop step so experienced users can correct intermediate outputs. The team reports that both newcomers and DomiKnowS veterans can produce a complete program in 10-15 minutes, versus roughly an hour to hand-code even a single component the old way.
Neuro-symbolic AI is supposed to make models more robust, interpretable, and data-efficient by mixing learned and rule-based reasoning, but it has stayed a niche pursuit partly because the tooling demands real fluency in bespoke syntax. Automating that fluency away, rather than just automating the model training itself, is a more interesting move than another wrapper around a chatbot, because it targets the actual adoption bottleneck.
The 10-15 minute figure comes from the paper's own authors testing their own demo, so it is a lab measurement of a self-hosted UI, not an independent benchmark against real production coding workflows.