Anthropic merges Claude's chat and Cowork for enterprise

For months, enterprise teams using Claude have faced an annoying micro-decision before starting any task: do I just need to chat, or do I need Cowork to handle actual work? Pick the wrong one, and you'd find yourself copy-pasting prompts across interfaces.

Anthropic has now eliminated that friction. Following its debut on Pro and Max plans back in September, the unified Claude experience is now generally available for Claude Enterprise across web, desktop, and mobile.

Cloud sandboxes keep running when you close your laptop

Basically, Claude is now a single conversation window. You can ask a quick factual question and get an instant response. But if that query evolves into crunching spreadsheets, running Python code, or building a presentation, Claude quietly shifts into cloud agentic mode without missing a beat.

Because the heavy lifting runs in an Anthropic-managed cloud environment, you don't need to leave your browser tab open. You can kick off a complex task at your desk, shut your laptop, and pick up the results on your phone via mobile notifications. Existing connectors, skills, projects, and MCP plugins carry over automatically.

Early enterprise users are already reporting major efficiency gains. Carvana noted its teams can now initiate tasks on computers and track progress out in the field. Another public-sector customer reported feeding Claude stacks of county, state, and federal budgets, letting it map complex funding streams in ten minutes—a report that previously took weeks.

On the security side, each agentic session runs inside an isolated virtual machine. Network egress is turned off by default for enterprise users, and connector credentials live outside the execution sandbox. Existing compliance setups, including HIPAA readiness, zero data retention for Claude Code, and customer-managed keys, remain untouched.

Curbing runaway agent costs

The biggest barrier to rolling out autonomous AI across a company has always been token anxiety. Agents can chew through budgets exponentially faster than standard chat.

To address this, Anthropic is leaning on model efficiencies alongside tighter administrative controls. The new Claude Opus 5.5 costs 40% less to run on typical enterprise workloads than its predecessor, while Sonnet 5.5 runs up to 30% cheaper than Sonnet 5.

Admins also get four new levers to manage spend. The standout is Efficient mode, a toggle that restricts the unified experience strictly to Sonnet and Haiku models. IT teams can also set role-based model entitlements, enforce spend caps down to individual members, and configure pooled budgets for entire departments.

Why it matters

We think this is one of Anthropic's smartest product refinements yet. The true hurdle to enterprise agent adoption hasn't been raw intelligence, but operational friction.

For enterprise employees, eliminating the mental barrier between chatting and delegating work changes the game. You no longer need to know whether a task requires a code sandbox before asking; the model orchestrates the backend automatically. The handoff between desktop and mobile also matches how modern teams actually operate.

For the AI sector, this sets a clean benchmark for how business agents should behave. Turning agentic execution into an invisible backend layer rather than a separate, intimidating developer tool will likely pressure competitors to follow suit.

That said, companies shouldn't blindly flip the switch without caution. While VMs and disabled network egress limit immediate exposure, autonomous workflows parsing sensitive files still require strict connector governance. We recommend admins lock down Efficient mode and test pooled budgets with small pilot groups before rolling it out company-wide.