Every Major AI Lab Just Built the Same Thing. Here's Why That Should Worry Your IT Team

Anthropic just launched Claude Duet, a proactive AI agent that does not wait for you to type a prompt. It watches your connected apps, decides what needs your attention, and keeps working after you put your phone down.

Here is the part that makes this bigger than one product launch. Anthropic is the fourth major lab to ship this exact idea in a matter of weeks. Meta has Muse. OpenAI has Dots. Google has its Gemini agent. Now Anthropic has Duet. When four competing labs independently land on the same product shape at the same time, that is not a coincidence. That is the industry telling you where it is headed next.

What Claude Duet actually does

Every morning, the agent reviews your connected email, calendar, and work apps and sends a short rundown of what matters. Through the day, it keeps checking those same accounts and reaches out only when something needs a decision. Users can also hand it bigger jobs directly, like building a dashboard, drafting a sales forecast, or preparing a research brief, and it figures out which tools it needs and does the work in the background.

It runs in its own virtual machine with a browser, so it can keep working even when you are not at your computer. And it connects to a lot. The agent plugs into thousands of connectors across email, messaging, and finance tools, so it can do things like spot an unpaid invoice or prepare payroll information for review.

It also has guardrails. Anthropic says the agent checks with the user before taking sensitive actions on their behalf, like sending something, paying something, or signing something.

Why it is not in your company yet, and why that matters

Right now, Claude Duet is a research preview. It is limited to individual Pro and Max subscribers in the US, and team and enterprise plans are not supported at launch. That sounds like a reason to ignore this story. It is actually the opposite.

This is the pattern every one of these labs follows. Ship the proactive agent to individual users first. Work out the rough edges. Then push it into business and enterprise tiers once trust is built. That second step tends to move fast. The gap between a consumer preview and a company wide rollout has been shrinking all year.

What this means for a large company

If you run a department, lead IT, or own security at a company this size, here is where your attention needs to go.

Adoption. Your people are already using tools like this on their personal accounts, often without telling anyone. A proactive agent that reads email and drafts replies is useful enough that employees will not wait for an official rollout. Shadow use is already happening or about to.

IT and security. These agents do not just answer a question. They connect to your inbox, your calendar, your Slack, and potentially your finance tools. Every one of those connections is a new surface to secure. Before any proactive agent touches company systems, IT needs a clear map of what it can see, what it can touch, and what it can do without a human checking first.

Governance. "Checks before it acts" is a good default, but someone has to define what counts as sensitive. Is sending an external email sensitive? Is updating a shared doc? Every company rolling this out needs a written policy, not a vague assumption that the AI will figure out the line.

Training. An agent that works in the background only saves time if people know how to hand it tasks, review its output, and catch a mistake before it ships. That is not intuitive. It takes real training, not a company wide email with a login link.

Measurable time saved. Done right, this is exactly the kind of tool that should show up in your monthly hours back report. A daily brief that used to take twenty minutes of inbox triage, done automatically, is a real number. But you can only measure it if someone set up the tracking before the rollout, not after.

The takeaway

Four labs building the same proactive agent in the same month is not noise. It is a signal that this becomes standard in business software soon, probably sooner than most IT roadmaps assume. The companies that win here are not the ones who wait for the enterprise version to show up. They are the ones who build the governance, the training, and the audit trail now, so that when it lands on their desks, it is a rollout instead of a scramble.