Reproduce “Agentics 2.0: Logical Transduction Algebra for Agentic Data Workflows” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Alfio Massimiliano Gliozzo, Junkyu Lee, Nahuel Defosse

Agentics 2.0, developed by IBM researchers, introduces a Python-native framework that formalizes Large Language Model (LLM) inference as a "logical transduction algebra" to construct reliable, type-safe, and explainable agentic data workflows. This framework achieved an average final score of 37.27 on DiscoveryBench, surpassing the best-known baseline of 33.7, and outperformed most submissions on the Archer NL-to-SQL benchmark.

Key insights

  • Re-framing LLM inference as a 'typed semantic transformation' (transduction) provides a principled foundation for building robust, verifiable, and explainable agentic workflows.
  • Formal properties of transducible functions (type safety, explainability, evidence locality, provenance) are crucial for ensuring semantic reliability and transparency in LLM-generated outputs.
  • An algebraic structure combined with Map-Reduce semantics enables modular composition, scalability through parallel execution, and reusable design patterns for complex data workflows.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.