Skip to content
AI News HubLIVE

This edition’s highlights

Agents

Organizing Context in a Multi-Agent Harness

  • Context modes give subagents two options: isolated (fresh context) or fork (inherit the supervisor’s full conversation).
  • Forked subagents reduce duplicated context-gathering and can take advantage of prompt caching.
LangChain BlogIn-site articleOrganizing Context in a Multi-Agent Harness

How HPE Zerto built an agentic troubleshooting system with Amazon Bedrock

  • Deployed as a pod inside customer environments, the agentic system is accessible from the existing Zerto UI and uses SSE streaming for real-time progress.
  • A multi-agent architecture with an Orchestrator, ZVM agent, and VRA agent handles everything from routine questions to log-heavy investigations.
AWS Machine Learning BlogIn-site articleHow HPE Zerto built an agentic troubleshooting system with Amazon Bedrock

Cohere's North Mini Code Megakernel Serving Engine

  • Megakernel runs the whole decode forward pass as a single persistent kernel, avoiding launch/sync stalls that leave HBM bandwidth unused.
  • Outperforms vLLM by 1.58x at batch size 1 (292 vs 185 tok/s) and sustains the gain across batch sizes and up to 256K context.
Cohere BlogIn-site articleCohere's North Mini Code Megakernel Serving Engine
Policy

Called to serve: Tech, research, and positive impact with Chris White

  • White's postdoc DARPA project on data analysis sent him to Afghanistan to prove the approach on the ground.
  • He later worked on dark-web search tools to help combat human trafficking, earning the 2016 Presidential Award.
Microsoft Research BlogIn-site articleCalled to serve: Tech, research, and positive impact with Chris White

You have reached the end of this edition.

Your reading list