FloWright Co-Evolves Multi-Agent Workflows, Boosting Small Models by up to 7.41%
Xuehang Guo · hf · 2026-10-02
The paper introduces FloWright, which uses the workflow itself as a harness for optimization. A hierarchical, structure-aware reward paradigm lets one agent role self-evolve or two or more roles co-evolve, with no extra models, labels, or executions. To fix the problem that workflow benchmarks are often solvable by a single agent, the authors propose DataWright, an adaptive data-hardening method that converts existing datasets into harder workflow-level tasks. Across document, slide, chart, code, math, and finance tasks, small open models trained with FloWright improve by up to +7.41%; co-evolving multiple roles (+5.03%) beats optimizing just one (+2.83%). Project page: xhguo7.github.io/FloWright.
More from coding & agent
- Dev recreates a DOOM-like game with GPT, nailing the original's gory feel — DeryaTR_ · 2026-10-02
- Process-mining agents found 20 steps and 7 loops in a workflow documented as 7 steps — vasuman · 2026-10-02
- Geoffrey Huntley: Forget reading code—your verification properties are all that matters — kieranklaassen · 2026-10-02
- Claude Skills explained: why prewritten PDF scripts beat pasting prompts every time — lxfater · 2026-10-02
- Eye.Art Polyphemus: a chat-first MCP for image generation and reference-based edits — axiomofaxiom · 2026-10-02
- Dev uses GitHub Copilot as a project lead: files issues, lets the agent do everything else — 0xkarasy · 2026-10-02