Divide, Diagnose, and Compose: Optimizing Agentic Pipelines and Harnesses
4.40T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2611.08723.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryDeCAF adapts pipeline optimization to a benchmark, evaluates prompts, topology, models, and tool policies separately, and records reasoning traces. Independent coding agents make constrained edits before a synthesis session localizes weak components and revises them. Reported tests show higher task performance, 4.30% lower inference cost, and 42.45% lower latency than the best baseline.
Why it mattersUseful for practitioners who need a repeatable optimization procedure rather than a single prompt recipe, with auditable comparisons across coupled pipeline decisions.
Cited by
No citations on record.