Cowork Hits GA With the Three Features Enterprise Clients Actually Need

The Wall Every Enterprise Deployment Hits

You build something useful with Claude. The client sees it work. Then it goes to IT.

IT wants to know who has access to what. Security wants logs they can audit. Compliance wants to know the system is observable. The conversation stalls, not because the AI is bad, but because the tooling is opaque.

That is the enterprise deployment wall. Most AI tooling hits it. Cowork hit it too, until April 9.

What Cowork GA Added

Anthropic released Cowork as a shared Claude workspace for small teams. The original version was genuinely useful if everyone on the team trusted each other, had full access, and you did not need to answer questions from a security team.

General availability landed on April 9, 2026. Anthropic framed the GA story directly in their announcement, "Making Claude Cowork ready for enterprise". It changes the deployment profile. Three features matter most.

Usage Analytics Dashboard

Cowork now has built-in visibility into how teams are using it. Admins can see which conversations are happening, which agents are running, and where activity is concentrated.

This answers a question that every operations lead eventually asks: "Who is actually using this, and how?"

Before GA, answering that required custom instrumentation. You had to build something just to find out what was happening. Now the answer is in the dashboard.

This is not just a reporting convenience. It makes internal accountability possible. If you deploy Cowork for a client, they can now see utilisation without relying on you to extract it for them.

OpenTelemetry-Compatible Observability

This is the one that unblocks enterprise deployments.

Cowork now emits telemetry in the OpenTelemetry format. OTel is the standard that enterprise observability stacks run on. Datadog, Grafana, New Relic, Honeycomb: they all speak OTel. Your client's IT team almost certainly already has one of these running.

What this means in practice: Cowork is no longer a black box. It becomes observable alongside everything else in the stack. Security teams can monitor it. Compliance teams can audit it. Incidents are traceable.

Without OTel support, any Claude-based tooling sits outside the client's existing monitoring infrastructure. That is a hard no for most enterprise security reviews. With it, Cowork is just another observable service.

This is the feature that gets you past security sign-off.

Role-Based Access Control

RBAC is the governance layer. Without it, Cowork was binary: someone had access or they did not.

GA introduces granular permission management. You can define who can do what at which scope. Admins have different capabilities than end users. You can configure the system to match an organisation's actual structure.

For agencies, here is the pattern that matters: you can give a client's IT admin full control of their Cowork instance while restricting what their end users can configure. That is the segregation compliance teams require.

Before RBAC, you could not deploy Cowork into an organisation with multiple teams, different access levels, or compliance requirements. Now you can.

Three Features, One Checklist

Enterprise IT and security teams run a checklist before approving AI tooling for production use. It looks roughly like this:

• Governance: who has access, and can we control it?
• Observability: can we see what is happening in real time?
• Auditability: can we trace what happened after the fact?

Cowork GA answers all three. RBAC covers governance. OpenTelemetry covers observability and auditability. The analytics dashboard makes utilisation visible to the people managing the deployment.

Before April 9, Cowork passed none of those checks reliably. Now it passes all three. That is not a minor update. It changes what you can propose.

The Broader Infrastructure Picture

Worth noting the timing. The Managed Agents API launched on April 8, one day before Cowork GA. (If you missed that release, the Managed Agents API post covers what it changes for execution infrastructure.) The New Stack's coverage puts it plainly: "What's been holding Claude Cowork back from full enterprise rollout wasn't the product, it was the governance layer."

These two releases are not coincidental. Managed Agents handles the execution layer: long-running agents, persistence, multi-agent coordination. Cowork handles the collaboration and governance layer: who uses the system, what they can do, and what you can observe.

You can read about using the Claude Code toolchain alongside these infrastructure layers for a complete picture of where Anthropic's tooling is heading.

The pattern Anthropic is building is coherent. A production-grade stack for organisations running AI at scale. The infrastructure layer is now there.

What This Means for Agency Deployments

If you build AI products for enterprise clients, your deployment conversation just changed.

Previously, proposing Cowork to a client with a real IT department meant hitting governance objections you could not answer. You either pushed a stripped-down version of the product or lost the conversation entirely.

Now you can go into that conversation with answers. Yes, you can control who has access to what. Yes, it integrates with your existing observability stack. Yes, there is an audit trail.

That is the difference between a pilot and a production deployment. Between a department trial and an enterprise rollout.

What to Do Now

If you have a client deployment where Cowork was previously off the table due to governance concerns, revisit it. The blockers are gone.

Pull up the security review questions that stopped the last conversation. Check them against RBAC (governance), OpenTelemetry (observability and auditability), and the analytics dashboard (utilisation visibility). The objections no longer hold.

If you are proposing Cowork to a new client, lead with these three features in the IT conversation. Do not wait for the objections. Name the checklist and show how Cowork answers it.

Where We Are

Cowork GA is not a major product announcement in the traditional sense. No new capability, no new pricing tier, no new use case.

What there is: a tool that was useful for small trusted teams and is now deployable into enterprise environments with real governance requirements.

That is the relevant change. Not what Cowork can do. What it can now be deployed into.

---

If you are building AI infrastructure for $500k+ clients and want to work through what Cowork and the Managed Agents API unlock for your specific context, that is the work we do.

Start with the IP Assessment

Already know what you want to build? The Program covers how we structure a 90-day engagement.

Explore the Program