Reproduce “ActionEngine: From Reactive to Programmatic Web Agents via State Machine Memory” 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
Hongbin Zhong, Fazle Faisal, Luis França, Tanakorn Leesatapornwongsa, Adriana Szekeres, Kexin Rong, Suman Nath
ActionEngine learns reusable models of website structure before tasks, then plans programs that execute against those models without repeated planning calls. On a 655-task WebArena evaluation across four domains, it reached 91.2% success versus 82.7% for Claude Code and averaged 27 seconds versus 87. The evaluation disabled online repair, and search-dependent tasks remained a weakness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.