WorkBorne first builds coherent and diverse artifact corpora through iterative web retrieval.
Task synthesis via artifact transitions yields long-horizon agentic trajectories.
WorkBorne generates 15,485 trajectories from 7,882 tasks, with over half exceeding 30 assistant turns, and boosts Terminal-Bench accuracy from 43.1% to 61.8%.
Chinese Tech
Junyi Zhou · Yufei Wang · Qiyuan Zhang · Chuhan Wu · Zhengxi Li · Wentao Zhao · +4 more
City University of Hong Kong · WeChat AI, Tencent · MBZUAI
Research Digest··2 min read
The authors propose WorkBorne, a framework that models agentic tasks using artifact transitions—specifying who performs the work, which artifacts anchor the initial state and intended outcome, and the relation between them.
Why this paper
From WeChat AI, Tencent and 2 others
In one line
Modeling agentic tasks as artifact transitions between input and output artifacts yields long-horizon trajectories that improve agent performance on terminal benchmarks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓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.
§