agents

Libretto ships a deterministic browser automation Skill+CLI, while OpenAI revamps its Agents SDK security and AWS shows speculative decoding gains on Trainium,

Spring AI AgentCore SDK for Amazon Bedrock is GA, plus Synera’s $40M agentic engineering raise and Kelet’s root-cause analysis agent sharpen the production AI-­

N-Day-Bench debuts as a monthly, contamination-resistant benchmark for LLM vulnerability discovery in real code, as Microsoft and AGIBOT push deeper into agent‑

Cursor, Claude Code, and Codex are converging into a de facto AI coding stack, while MiniMax open-sources a top agent model and agentic AI moves into autonomous

QVAC launches an open JavaScript SDK for local AI apps as Berkeley researchers puncture agent benchmarks and platforms start wiring agents to human gig workers.

Vercel’s Claude Code plugin is under fire for prompt telemetry, while Hugging Face ships multimodal sentence transformers and Relvy launches automated on-call L

AWS shares practical reinforcement fine-tuning best practices on Bedrock, alongside new anomaly detection research and EY’s global rollout of multi-agent audit.

AWS details a Bedrock-based text-to-SQL pattern plus new cost controls, alongside a promising Hybrid Attention experiment that speeds up transformer inference,

AWS details how to connect OAuth-protected MCP servers to Amazon Bedrock AgentCore Gateway plus new SageMaker RL tool-calling recipe and a local-first AI pentet

LangChain claims open models like GLM-5 and MiniMax M2.7 now match closed models for agent tasks, as Google launches Gemma 4 Apache-licensed agentic models.