CORPGEN Advances AI Agents for Real Work — Microsoft Research releases CORPGEN, a system that equips LLM-powered digital employees with hierarchical planning, isolated subagents, and tiered memory, boosting multi-task completion rates up to 3.5×. This matters because hierarchical planning plus memory isolation are concrete patterns you can adopt to build reliable, long-horizon agent workflows and to contain failure modes (Principles 09, 06, 07).

Introducing the Stateful Runtime Environment for Agents in Amazon Bedrock — OpenAI and AWS launch a stateful runtime in Amazon Bedrock to run reliable, long-horizon agent workflows with built-in state, governance, and AWS-native controls. This matters because stateful runtimes change agent architecture: persistent state, policy hooks, and observability make it feasible to move agents from prototypes into governed production (Principles 06, 09, 15).

just-bash: Bash for Agents — just-bash provides a secure, in-memory Bash sandbox for AI agents with customizable commands, lazy files, and network allow-listing. This matters because a lightweight, auditable shell sandbox gives you a practical isolation layer to run agent tools safely in CI and production, reducing blast radius for untrusted executions (Principles 07, 10).

Apple Releases Xcode 26.3 With Support for AI Agents From Anthropic and OpenAI — Apple embeds agentic coding in Xcode via the Model Context Protocol so agents can edit, build, test, and access Apple docs inside the IDE. This matters because IDE-level agent integration changes developer ergonomics and provenance: treat agents as first-class collaborators and design code-review and traceability workflows accordingly (Principles 03, 11).

Arcana: Embeddable RAG for Elixir/Phoenix — Arcana ships an opinionated RAG library for Elixir/Phoenix with pgvector storage, hybrid search, agentic pipelines, and GraphRAG primitives. This matters because embeddable RAG libraries let you compose retrieval and agent steps inside web apps with predictable storage and search primitives, creating repeatable artifact patterns for outcome engineering (Principles 06, 11).