Drafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic Systems
We introduce Speculative Recursive Self-Improvement (SpecRSI), which drafts executable successor agents before their predecessors are fully evaluated, while requiring complete verification before adoption.
ProofPaper ↗
Key points
- Improving an agent through successive revisions requires balancing exploration and evaluation under a limited budget.
- A narrow search can miss useful sequences of edits, while fully evaluating every intermediate version can exhaust the budget before those sequences yield a verified improvement.
- Verification-Aware Tree Search (VATS) allocates the budget between further edits and outstanding checks using the predicted verified return of a continuation.
- Group Verification Optimization (GVO) learns the improver from completed searches while keeping the real evaluator fixed.
Sources (1)
- [1]Drafting Agents, Not Just Tokens: Speculative Recursive Self-Improvement for Agentic SystemsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 09:43 AM
We introduce Speculative Recursive Self-Improvement (SpecRSI), which drafts executable successor agents before their predecessors are fully evaluated, while requiring complete verification before adoption.
Improving an agent through successive revisions requires balancing exploration and evaluation under a limited budget.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026DuoDiscover: Scaling Autonomous Discovery Beyond the Evaluation Bottleneck
- Oct 11, 2026Heterogeneous Structured Pruning Under Resource Constraints for Hardware-Efficient Large Language Models
- Oct 11, 2026REPRO: Reflective Evolutionary Prompt Optimizer for Automated LLM Red-Teaming
- Oct 11, 2026SpecFit: Not All Draft Tokens Cost the Same in Offloaded MoE Inference
- Oct 11, 2026Imperfect Verifiers, Exact Targets: Inference-Time Scaling for Continuous Generation
- Oct 10, 2026How Should a Camera-First AI Agent Work?