The authors present a head-to-head comparison of two paradigms for tool-augmented LLM agents: Retrieve-then-Plan (RtP), which embeds the user query and retrieves similar APIs, and Plan-then-Retrieve (PtR), which first generates abstract tool stubs (hypothesized function signatures) to use as retrieval queries.
Tool discovery paradigm crossover: Retrieve-then-Plan wins on simple queries, Plan-then-Retrieve on complex ones at scale
Authors systematically compare two retrieval paradigms across 47,000 APIs and show that the optimal choice depends on query complexity, with implications for adaptive routing.
Big Tech
Mingfei Zhang · Amit Surana · Ziauddin Majid · Chenxi Wang · Michael Xu
Apple
Research Digest··3 min read
The authors present a head-to-head comparison of two paradigms for tool-augmented LLM agents: Retrieve-then-Plan (RtP), which embeds the user query and retrieves similar APIs, and Plan-then-Retrieve (PtR), which first generates abstract tool stubs (hypothesized function signatures) to use as retrieval queries.
Why this paper
From Apple
In one line
Plan-then-Retrieve outperforms Retrieve-then-Plan on multi-tool queries, while the opposite holds for single-tool queries.
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
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§