The authors study how skill-mining evidence and representation affect downstream agent performance in enterprise settings.
Mined agent skills need domain-specific evidence and format choices
A controlled study of skill mining from production traces shows that workflow and ontology representations, and the availability of outcome labels, have varying effects on downstream task success across enterprise benchmarks.
Big Tech
Yue Ran Kang · Colton Mikolajczyk · Chhaya Methani · Hazel Mak · Sahil Bhatnagar · Susheel Suresh · +1 more
Massachusetts Institute of Technology · Microsoft Corporation
Research Digest··2 min read
The authors compare six combinations of mining evidence (successful traces only, successes and failures with labels, or the same mix without labels) and skill form (ordered workflow plans or declarative ontologies) while holding the mining pipeline fixed.
Why this paper
From Microsoft Corporation and Massachusetts Institute of Technology
In one line
Mined skill performance depends on task domain; workflows beat ontologies on one benchmark, outcome labels help on another, but no configuration wins everywhere.
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
- ✓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.
§