AI news: Google's Gemini agent, Copilot goes local on Windows, OpenAI Decisions API
6 to 9 October 2026: Google gives AI agents their own account and email, Microsoft moves Copilot onto the PC, OpenAI sells fast decisions by API, and Anthropic updates its usage rules.
The thread running through this week: the AI agent is no longer a chat window. Vendors now treat it like a member of staff, with its own account, permissions, budget and a record of what it did.
Google: the Gemini agent gets its own account and email address
On 8 October, at its Gemini at Work 2026 event, Google Cloud introduced the Gemini agent, which it describes as a universal agent for work. From one prompt box it answers questions, writes documents, creates images and runs code.
What matters for businesses:
- Coworker agents. A team describes a role and the agent gets its own Workspace account with email, calendar and Drive, on an address of the form @agents.company.com. It only sees what is explicitly shared with it.
- Several models. The agent picks a model per task and today combines Gemini models with Claude models from Anthropic.
- Spend caps. You set a hard limit per project in the Cloud Billing Console. When it is reached the agent pauses and resumes with one click.
- Connections. Beyond Workspace it works with Microsoft 365, Slack, Jira, Salesforce, ServiceNow and SAP, and with any MCP server.
Google's post gives no price and no date for wider availability. According to Android Headlines, the agent is in private preview for selected Gemini Enterprise customers.
Microsoft: Copilot on Windows also works locally
On 7 October Microsoft presented an approach it calls hybrid intelligence: agents run locally when that makes sense and use the cloud when they need to. On Copilot+ PCs it gains three abilities:
- with your permission, it understands files and recent activity on the PC;
- it takes actions in Windows, such as organising files or running diagnostics;
- it uses on-device models and combines them with the cloud.
These features arrive "in the coming months", and availability varies by device, market and processor. Already generally available is Microsoft Execution Containers for Windows 11: the company defines which files and networks an agent may reach, and the policy is enforced while the agent runs.
OpenAI: Decisions API for fast decisions
On 6 October OpenAI opened the Decisions API in public beta to all developers. Instead of prose it returns a structured decision: the probability that a statement is true, a choice from fixed options, or a score on a scale. It accepts text and images and runs on GPT-6 Luna. The price is USD 0.10 per million input tokens, and output tokens are not billed. OpenAI says it is up to 10 times faster than the same model through the Responses API. That is the vendor's own figure, so measure it on your own data.
Anthropic: updated usage rules and workflows for agents
On 8 October Anthropic published changes to its usage policy that take effect on 12 November 2026. The rules on fake accounts and on hiding who is behind a message now sit in one section and explicitly cover commercial campaigns as well as political ones.
A day later, on 9 October, Claude Managed Agents gained dynamic workflows in beta. The agent writes a program that runs many agents in phases and combines their results, for example to review hundreds of documents. One run can start up to 1,000 agents and lasts 24 hours at most by default. There is no separate price: tokens are billed at each model's rate and count towards the session budget.
What it means for your business
- Treat an agent like a new hire. Give it its own account, the minimum permissions it needs and a monthly spend limit.
- Classification is the cheapest place to start. Is this email an enquiry or a complaint, which team gets the ticket, does the product photo meet the guidelines: decisions like these now cost a fraction of a cent.
- Review your marketing automation. If AI writes reviews, comments or messages on your behalf, the recipient must be able to tell who is behind them. At Anthropic that becomes an explicit rule on 12 November.
If you want to find out which part of your work is worth automating, see our AI automation service or get in touch.
Sources
- Google Cloud Blog: Gemini at Work 2026: Introducing Gemini agent (8. 10. 2026)
- Google Blog: Google Cloud introduces the Gemini agent (8. 10. 2026)
- Android Headlines: Google launches universal Gemini AI agent for enterprise work (8. 10. 2026)
- Windows Experience Blog: Building Windows for hybrid intelligence (7. 10. 2026)
- OpenAI Developer Community: Decisions API is now available in public beta (6. 10. 2026)
- Let's Data Science: OpenAI Opens Decisions API Public Beta (7. 10. 2026)
- Anthropic: 2026 Usage Policy update (8. 10. 2026)
- Claude Platform Docs: Release notes (9. 10. 2026)
- Claude Platform Docs: Workflow runs (10. 2026)