SourcesStitchCUDA Multi-Agent RL GPU Kernel Generation Near 100 Percent SuccessarXiv·high signalMulti-agent RL system writing GPU kernels at production quality, 1.72x speedup over baselinesSourceSource pagearXiv↳ Follow the threadStack layer / Contrast'Resample or Reroute?' Reframes LLM Routing Under Real Budget ConstraintsarXivStack layer / Follow-up threadCache and reuse agent-generated tools instead of regenerating code every request (arXiv 2607.08010, July 9)arXiv 2607.08010Stack layer / Follow-up thread'Who Broke the System?' AgentLocate Pinpoints the Failing Agent and Step in Multi-Agent RunsarXivThreat pattern / Follow-up threadVEXAIoT: arXiv paper demonstrates autonomous IoT vulnerability exploitation using AI agentsarXivPolicy dependency / Stack layerReason Less, Verify More: Deterministic Gates Catch a Silent Policy-Violation Failure in Tool-Using AgentsarXivStack layer / Threat patternMitigating Taint-Style MCP Server Vulnerabilities Without Touching Server Code (SPELLSMITH)arXivFollow-up threadLDT-Coord Coordinates Heterogeneous LLM Embodied Agents With 70x Less Communication via a Digital TwinarXivFollow-up thread'Hindsight Gating' Cuts Communication in Bandwidth-Constrained Cooperative AgentsarXiv