Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-02 · source-backed
Counterintuitive finding: 36.4% of multi-turn prompts are fully self-contained. Removing prior assistant responses yields up to 10x context length reduction with minimal quality loss. Models over-condition on previous responses, introducing errors. Directly actionable for multi-turn agent builders. arXiv 2602.24287
Each link below shares sources, entities, or timing with this story.
Models multi-hop agent reasoning as a dynamic search tree, transitioning from exploration to exploitation as token budget depletes. Includes formal convergence proof. Directly applicable to any agent loop burning tokens on redundant calls. arXiv 2603.12634
arXiv 2603.15566 proposes treating commit history as machine-readable knowledge substrate for coding agents. As AI-written commits lack semantic trails, institutional knowledge loss compounds run-over-run. Directly actionable: structured commits become persistent agent memory...
Three-component infrastructure for maintaining agent coherence in a 108K-line C# codebase: hot-memory constitution, 19 specialized domain-expert agents, and cold-memory knowledge base. Quantitative metrics from 283 sessions. Directly actionable for multi-agent coding workflows...
Introduces temporal causal diagnostics to distinguish legitimate task execution from injected manipulation in multi-turn agent interactions, plus context purification to neutralize poisoned content. Directly applicable to anyone building agents that call external tools. arXiv...
Addresses the fundamental privacy dilemma: cloud models need data access but enterprises can't share sensitive information. Splits execution between enterprise-side privacy agents and cloud-side capability agents. Directly relevant to AWS AgentCore and enterprise adoption. arX...
CAM-DF reframes tool acquisition as a stopping problem, training on the offline gap between stopping now and the best continuation, using the sign to label the decision and the magnitude to weight each error by payoff at stake. Across five domains and 1,343 tasks: 37% fewer to...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.