AgentWeave (arXiv 2608.23078) is a deterministic pre-inference routing layer that shrinks the candidate tool set an LLM sees, filtering on eligibility, requirement, capability and routing signals before an unchanged downstream model does function calling. On 48 fresh BFCL V4 multiple-function tasks with the frozen Hammer2.1-1.5b model it reached 6/48 successes versus 0/48 for all-tools, random-top-8 and semantic-top-8 baselines (McNemar p=0.03125), while exposing 70.18% fewer tools, using 61.70% fewer input tokens and cutting mean local-model latency by 50.95%. The authors themselves frame it as a BFCL-derived routing-pressure study, not a leaderboard result, and absolute success stays low.
AgentWeave filters the tool list before reasoning to cut agent cost
2026-08-26Frontier watchWATCH
AgentWeave (arXiv 2608.23078) is a deterministic pre-inference routing layer that shrinks the candidate tool set an LLM sees, filtering on eligibility, requirement, capability and routing signals before an unchanged downstream model does function calling. On 48 fresh BFCL V4 multiple-function tasks with the frozen Hammer2.1-1.5b model it reached 6/48 successes versus 0/48 for all-tools, random-top-8 and semantic-top-8 baselines (McNemar p=0.03125), while exposing 70.18% fewer tools, using 61.70% fewer input tokens and cutting mean local-model latency by 50.95%. The authors themselves frame it as a BFCL-derived routing-pressure study, not a leaderboard result, and absolute success stays low.
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
Tool-catalog bloat is a real cost driver now that agents routinely face hundreds of MCP tools, and the token and latency cuts here are large. But treat the headline as directional: 48 tasks and a frozen 1.5b model is a stress test, not production evidence, and the 0/48 for semantic top-8 selection is the more interesting warning — similarity-based tool selection may be the wrong default.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Method | deterministic pre-inference filtering: eligibility, requirement, capability, routing signals; downstream model unchanged |
|---|---|
| Results | 6/48 vs 0/48 (all-tools, random top-8, semantic top-8) on BFCL V4, frozen Hammer2.1-1.5b; -70.18% tools exposed, -61.70% input tokens, -50.95% mean latency |
| Statistics | McNemar p=0.03125; 95% bootstrap CI +4.17 to +22.92 pp |
| Caveats | authors label it a routing-pressure study, not an official leaderboard result; absolute success low; small model, 48 tasks |
| Code | open-source implementation and reproducibility artifacts stated in comments (no direct link on paper page) |
| Arxiv | 2608.23078, submitted 2026-08-24 |
Technical impact 2
Engineering value 3
Adoption signal 1
Maturity 1
Verification cost 2
Risk 2