AI Briefing#agentic-ai
The latest AI and developer news about #agentic-ai, with the original source and a short summary.
The latest AI and developer news about #agentic-ai, with the original source and a short summary.
TokenSpeed outperformed TensorRT-LLM in B200-based LLM inference benchmarks for agents.
NVIDIA and ServiceNow announced autonomous AI agents for enterprise and an OpenShell collaboration.
Amazon demonstrated supply chain cost savings in a 9-week pilot with Flo Pro, which combines VCG and CPP.
DigitalOcean unveiled an AI-native cloud built for inference and agent workloads.
AWS implemented a dynamic swarm combining 7 expert agents on top of AgentCore.
QClaw-4B has been released as a 4B model for OpenClaw-compatible agents.
OpenAI has announced GPT-5.5, an agent-enhanced model.
With Qwen3.6 35B and a routing scaffold, 9 out of 10 real Go tasks passed.
Deep Insight isolated agent execution from code execution by separating AgentCore Runtime and Fargate.
TPU 8i and 8t for agents were unveiled, strengthening inference and training performance respectively.
GitHub Copilot has locked individual plans and moved Claude Opus 4.7 up to Pro+.
Qwen has open-sourced Qwen3.6-27B, a 27B dense multimodal model.
AI agents are already too human, repeatedly showing distraction and inconsistency.
Moonshot AI has released Kimi K2.6 via Chat and API, touting it as open-source state-of-the-art.
The Stanford AI Index 2026 report analyzes model performance, compute, investment, and environmental impact.
Morgan Stanley analyzed that the rise of Agentic AI will expand AI chip spending beyond GPUs to CPUs.
Exa built Canon, which compiles search paths into a DAG to control complex pipelines.
Released a workflow and paper for modifying existing code using the Github Copilot agent.
Z.AI's GLM-5.1 has been added to the NVIDIA API and is offered as a free endpoint.
In a 30k LOC agentic evaluation, Qwen 3.6 35B outperformed Gemma 4 26B.
By centering on Kiro and combining three layers—execution, generation, and accumulation—Eximbay standardized organizational AX.
Qwen3.6 adds preserve_thinking, improving reasoning context and KV cache efficiency.
The authors propose **ManyIH**, which resolves agent instruction conflicts using up to **12 tiers** of authority.