Mastra ships runtime model routing and schema unification
Mastra 1.11.0 adds dynamic model fallback routing and schema unification, while AWS debuts disaggregated inference and NVIDIA launches OpenShell for safer long‑
llm
Agent Harnesses, HF Buckets, and Bedrock Fine-tuning: Making Agents Actually Shippable
Hugging Face adds Storage Buckets, AWS streamlines Oumi fine-tuning into Bedrock, and LangChain dissects the “agent harness” pattern powering real production AI
Million-Token Contexts, Serverless Nemotron, and Type-Safe Agents
Hugging Face pushes million-token training with Ulysses SP, AWS ships serverless Nemotron 3 Nano, and LangGraph 1.1 adds type-safe streaming for serious AIagent
Token Diets, Talking Agents, and Smarter Memories
Cut MCP token waste with mcp2cli, boost tool-calling with Mastra 1.10.0, and explore open-source high‑fidelity speech with MOSS‑TTS for production AI agents.
AI Agents On Free Tiers, In Your Org Chart, And Inside Cyberattacks
AI agents now run a one‑person company for $0, coordinate in 13‑agent swarms, and power cyberattacks. Plus new orchestrators, semantic merge, and OpenClaw play‑
Elixir agents, on-call bots, and AI that actually ships code
Elixir agent framework Jido 2.0, autonomous on‑call engineer Steadwing, and Vela’s multi-party scheduling agents show how production AI agents are maturing fast