RSIFlow: Adaptive Self-Improvement Workflows with an Evolving Meta-Agent
We propose RSIFlow, a Recursive Self-Improvement framework that integrates adaptive workflow orchestration with meta-agent capability evolution.
ProofPaper ↗
Key points
- Large language model (LLM) agents have demonstrated strong capabilities in complex tasks, including code generation, tool use, and multi-hop question answering.
- Existing self-improvement methods enhance agent performance through model optimization, harness evolution, and executable artifact refinement, while recursive self-improvement (RSI) further extends optimization to the improvement mechanisms themselves.
- However, adapting self-improvement workflows to evolving performance bottlenecks while improving workflow orchestration through execution feedback remains a key challenge.
- RSIFlow diagnoses performance bottlenecks from task-agent trajectories and environmental feedback to dynamically select component-specific improvement workflows.
Sources (1)
- [1]RSIFlow: Adaptive Self-Improvement Workflows with an Evolving Meta-AgentarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 12:45 PM
We propose RSIFlow, a Recursive Self-Improvement framework that integrates adaptive workflow orchestration with meta-agent capability evolution.
Large language model (LLM) agents have demonstrated strong capabilities in complex tasks, including code generation, tool use, and multi-hop question answering.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 15, 2026Recursive Reward Improvement via Adversarial Meta-Prompt Evolution
- Oct 15, 2026Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio
- Oct 15, 2026VLA-$Δ$: Learning Physical State Transition for Vision-Language-Action Models
- Oct 14, 2026$τ$-Emotion-Bench: Exposing the Interaction Fragility of LLM Agents through Real-World User Emotions
- Oct 14, 2026Active Generative Simulation: Meta-Learning What to Simulate for Sim-and-Real Policy Co-Training
- Oct 14, 2026ReaDy: Recurrent States as Dynamic Context for Speculative Drafting in Hybrid Linear Attention Large Language Models