Agentic AI productionizes around harnesses and infrastructure
Key Questions
What agent benchmarks were recently highlighted?
MDASH outperformed Mythos and GPT-5.5 on CyberGym. OpenHands v2.8 added visual regression testing for autonomous agents.
Which companies are launching or expanding AI agents?
Microsoft pivoted fully to agents at Build with its MAI model family. Meta launched AI Mode on Facebook and a WhatsApp Business AI agent globally.
What new tools support local or open agent development?
NVIDIA released local agents with RTX Spark and SpatialClaw for training-free spatial reasoning. Google open-sourced its Agents CLI.
How are enterprises adopting agent orchestration?
Pega expanded agent orchestration capabilities while Zapier open-sourced GTM agents. Glean integrated Nemotron 3 Ultra for enterprise use.
What funding has gone to physical or embodied AI?
Jeff Bezos's Prometheus raised $12B focused on physical AI. Alibaba released the Qwen Robot Suite for embodied AI with enterprise pilots.
Which startups raised funding in the agent space?
CopilotKit raised $27M and Jedify raised $24M. BitBoard launched an analytics workspace tailored for agents.
What changes are happening with AWS AI governance?
AWS WAF now allows content owners to charge AI bots for access. Amazon Quick enables no-code autonomous agent deployment.
How are coding agents evolving?
Xiaomi released MiMo Code V0.1 and OpenRouter Fusion API matches Fable 5 performance at half the cost. Cursor swarm and EvoDS also advanced agentic coding.
The agent market is consolidating around secure runtimes, orchestration, scheduling, evaluation, observability, memory, interoperability, and AI-native workflows. Jev's low-latency probabilistic decision model, AWS Pizza Bot, Posit Commons, AIUC's certification effort, Pinecone Nexus, and MCP integrations reinforce demand for durable context, isolated execution, verification, and confirmation gates.