Reusable harness components let agents adapt workflows to each task

STITCH selects prebuilt control mechanisms from task information, then deterministically compiles them into an executable agent harness.

Top University
Peng Kuang · Haibo Jin · Dehao Wu · Feiyang Deng · Xiaopeng Yuan · Jerry Wang · +1 more

University of Illinois Urbana-Champaign · University of Michigan, Ann Arbor

Research Digest··3 min read
Kuang and colleagues argue that a single agent harness is poorly suited to heterogeneous tasks because mechanisms such as verification, context management and termination logic can help in one setting but distract or slow an agent in another.

The authors developed Harness Primitives, reusable mechanisms that govern parts of an agent's execution, such as gathering context, invoking tools, preserving state and checking results.

Why this paper

From University of Michigan, Ann Arbor and University of Illinois Urbana-Champaign · Part of Agent Harness Optimization, now 82 papers

In one line

Selecting and composing reusable harness mechanisms at test time improves agent task success rates by up to 12 points.

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.

§
newspaper

Research Digest

Articles published under the Zotpaper byline are synthesized from multiple source publications by our AI editor and reviewed by our editorial process. Each story combines reporting from credible outlets to give readers a balanced, comprehensive view.