For two years an "AI agent" was mostly a demo. In 2026 it became a product. Models now use apps, browsers and databases reliably, remember very long tasks, and run in the background.
In October 2026 Anthropic launched Claude Duet, an agent that briefs you each morning, checks your apps during the day and keeps working while you are away — but asks before it sends, pays or signs anything. OpenAI's Dots runs a task board with sub-agents, and Google previewed an agent that works across Workspace, Microsoft 365 and Slack.
Reviews of the year say most large companies now run agents in production, often as teams of specialist agents working together. The catch: an agent with access to email, finance and documents needs clear limits, logs and human checkpoints — exactly what the vendors built into their own products.
What it means for your business
- Start with simple, high-volume tasks like HR requests, service triage or invoice matching.
- Require a human to approve payments, external messages and record changes.
- Treat each agent like an employee account: least access, full logs, regular review.
How MTH can help: Our Agentic AI for Every Department solution delivers governed agents with approval gates and audit logs — on your data, on your cloud.
This briefing is original commentary by MTH based on public reporting; figures are as reported by the sources named above and may be revised over time.
