WorkflowGen:an adaptive workflow generation mechanism driven by trajectory experience

2026-04-23 · Source: cs.LG updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Software Development & Engineering · Depth: Expert, long

Summary

WorkflowGen is an adaptive framework designed to automate workflow generation for large language model (LLM) agents, aiming to reduce token consumption and improve execution efficiency and success rates. It addresses issues like high repeated reasoning overhead, excessive token consumption, and unstable execution chains in existing LLM-based methods. WorkflowGen structurally captures full execution trajectories, extracting reusable experiences at both node and workflow levels, including error fingerprints, optimal tool mappings, and exception-avoidance strategies. The framework employs a closed-loop generation mechanism that integrates trajectory rewriting, experience updating, and template induction, performing lightweight generation only on variable nodes. It also features a three-tier adaptive routing strategy that dynamically switches between direct trajectory reuse, rewriting-based generation, and full initialization based on semantic similarity to historical queries. Qualitative comparisons indicate WorkflowGen reduces token consumption by over 40% compared to real-time planning and improves success rates by 20% on medium-similarity queries.

Key takeaway

For AI Architects and AI Engineers designing LLM-powered agents for business automation, WorkflowGen offers a robust approach to significantly cut operational costs and enhance reliability. By implementing its trajectory experience extraction and adaptive routing, your systems can move beyond costly real-time planning, reusing historical successes and failures to achieve over 40% token consumption reduction and improved execution stability. Consider integrating similar memory-driven, rewrite-based mechanisms to make your agent deployments more efficient and fault-tolerant.

Key insights

WorkflowGen optimizes LLM agent workflow generation via adaptive experience reuse and minimal rewriting, reducing token costs and improving robustness.

Principles

• Reuse historical trajectories to minimize LLM inference.
• Extract experiences at dual granularities: node and workflow.
• Adapt generation strategy based on query similarity.

Method

WorkflowGen extracts structured experiences from execution trajectories, performs lightweight rewriting on variable nodes for similar queries, and uses a three-tier adaptive routing strategy (direct reuse, rewriting, or full initialization) based on semantic similarity.

In practice

• Implement dual-granularity experience extraction for LLM agents.
• Use adaptive routing to balance cost and success rate.
• Prioritize structural reuse over full re-planning.

Topics

• Automatic Workflow Generation
• LLM Agents
• Trajectory Experience
• Token Efficiency
• Adaptive Routing

Best for: AI Architect, AI Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• FlowBank: Query-Adaptive Agentic Workflows Optimization through Precompute-and-Reuse — Optimizing LLM multi-agent systems requires a compact bank of reusable, complementary workflows selected adaptively at inference time.
• Tree-of-Experience: A Structured Experience-Management Solution for Self-Evolving Agents under Low-Repetition and Implicit-Reward Environments — Structured experience management is crucial for self-evolving LLM agents in low-repetition, implicit-reward environments.
• Another try at LLM workflow automation — Effective LLM automation requires extensive scaffolding and iterative tooling, not just smarter models.
• Remember Me, Refine Me: A Dynamic Procedural Memory Framework for Experience-Driven Agent Evolution — ReMe dynamically manages LLM agent procedural memory, enabling smaller models to surpass larger, memoryless counterparts through experience-driven evolution.
• TrajGenAgent: A Hierarchical LLM Agent for Human Mobility Trajectory Generation — TrajGenAgent's hierarchical LLM-agent framework generates realistic human mobility trajectories by decoupling semantic planning from spatiotemporal grounding without fine-tuning.
• Tools as Continuous Flow for Evolving Agentic Reasoning — FlowAgent reconceptualizes LLM tool chaining as continuous trajectory generation for robust, generalized reasoning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.LG updates on arXiv.org.