WorkForge begins with expert-defined workflow blueprints describing the resources, decisions and outputs involved in professional tasks.
Workspace-grounded synthesis scales verifiable training for long-horizon work agents
WorkForge generated 16,700 realistic workspaces whose tasks and evaluators were derived from facts found directly in their files.
Chinese Tech
Jiazheng Zhang · Long Ma · Yunxian Yang · Zhiheng Xi · Zhikai Lei · Yajie Yang · +20 more
Tencent · Fudan University
Research Digest··2 min read
Zhang and colleagues introduce WorkForge, a pipeline for creating training environments in which agents must navigate multiple files and produce professional deliverables over extended interactions.
Why this paper
From Tencent and Fudan University
In one line
WorkForge builds verifiable work-agent environments from real-world files by anchoring tasks and verifiers in observable workspace facts, improving agent performance across benchmarks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (3 noted)
- ✓Reports numbers on named benchmarks (2 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§