Google puts a work agent in the prompt box
At Gemini at Work 2026, Google Cloud introduced the Gemini agent — a single enterprise interface meant to plan tasks, call tools and business systems, and return finished work inside docs, inboxes, and developer environments. It auto-picks a model (with Anthropic Claude in the picker first, open and private models later), and adds spend caps and multi-model routing so costs stay controllable. Early testers include On and PayPal; Gemini already claims 1B+ monthly users and near-90% Fortune 100 enterprise adoption. The consumer rollout comes later — businesses are the beachhead.
What else matters
Anthropic launches Cyber Mission. Anthropic opened a long-term defender program: Critical Infrastructure Defense Program with partners including CrowdStrike, Palo Alto Networks, and Dragos for OT/grid security, plus free OSS Scanner scans for enrolled open-source projects (model-generated PoCs and suggested fixes, no human review, >90% true-positive target).
Claude usage policy sharpened for elections and abuse. Anthropic’s 2026 Usage Policy update renames elections rules to “Do Not Undermine Democratic Processes,” bans deceptive voter disruption, and newly forbids sustained needless cruelty toward models — aligned with Claude already ending persistently abusive chats.
Manus raises $500M+ after Meta unwind. Parent Butterfly Effect said it raised more than $500 million led by Boyu and IDG, with Tencent and HSG participating — first round since Beijing forced the Meta deal off. Hiring continues; Cue and Manus 2.0 remain the product story.
Nous Research: $90M at $1.5B. FinSMEs reports Robot Ventures led a round with NVIDIA, M12, Samsung, and USV to fund GPU clusters and Hermes-family open models for enterprise agent orchestration.
Worth your time
NVIDIA and Microsoft on local AI for Windows — RTX Spark preorders and a DGX Station for Windows preview put agents on the desktop, not just the cloud.
