Why Search When You Can Transfer? Amortized Agentic Workflow Design from Structural Priors

2026-04-27

AI Summary

This work addresses the high computational cost and limited cross-task reusability of existing automated agent workflow designs, which typically rely on task-level iterative search. The authors propose SWIFT, a novel framework that reframes workflow design as a transferable structural prior. By contrastively analyzing search trajectories from source tasks, SWIFT distills compositional heuristics and interface contracts, then leverages conditional generation with large language models to synthesize complete, executable workflows for new tasks in a single inference pass—eliminating the need for iterative optimization. Experiments demonstrate that SWIFT outperforms state-of-the-art methods across five benchmarks, reduces per-task optimization cost by three orders of magnitude, and successfully generalizes to four unseen tasks and three distinct base models.

Abstract

Automated agentic workflow design currently relies on per-task iterative search, which is computationally prohibitive and fails to reuse structural knowledge across tasks. We observe that optimized workflows converge to a small family of domain-specific topologies, suggesting that this combinatorial search is largely redundant. Building on this insight, we propose SWIFT (Synthesizing Workflows via Few-shot Transfer), a framework that amortizes workflow design into reusable structural priors. SWIFT first distills compositional heuristics and output-interface contracts from contrastive analysis of prior search trajectories across source tasks. At inference time, it conditions a single LLM generation pass on these priors together with cross-task workflow demonstrations to synthesize a complete, executable workflow for an unseen target task, bypassing iterative search entirely. On five benchmarks, SWIFT outperforms the state-of-the-art search-based method while reducing marginal per-task optimization cost by three orders of magnitude. It further generalizes to four additional unseen benchmarks and transfers successfully from GPT-4o-mini to three additional foundation models (Grok, Qwen, Gemma). Controlled ablations reveal that workflow demonstrations primarily transfer topological structure rather than surface semantics: replacing all operator names with random strings still retains over 93% of the full system's average performance.

Problem

• agentic workflow design
• iterative search
• structural priors
• knowledge reuse
• computational cost

Innovation

• amortized workflow design
• structural priors
• few-shot transfer
• agentic workflows
• topology generalization