A few months ago, when Anthropic launched Claude Cowork, a lot of people wondered why. There was already Claude Code. There was already Claude.ai. Why ship a stripped-down version of Claude Code for non-engineers?

The answer became clearer with each weekly release.

Claude Cowork isn't a stripped-down anything. It's the result of Anthropic noticing that people had already been using Claude Code for things it was never designed for — knowledge work, ops, sales, finance, marketing — and deciding to build a product around that pattern, on purpose, with enterprise controls baked in from day one.

Five months in, Cowork is the most interesting AI productivity product on the market. Here's why.

How we got here

Claude Code launched in mid-2025 as a coding assistant. Within six months, it hit $1B in annualized revenue. But Anthropic saw something the launch positioning missed: a lot of the usage wasn't engineering. People were using Claude Code to draft proposals, run finance models, build internal tools, scrape data, automate workflows — knowledge work, not code.

Cowork was the response. Same harness, same agentic loop, same MCP plumbing, repackaged for non-engineers.

At launch, this looked redundant. It wasn't. By separating the two, Anthropic could ship the enterprise-shaped features knowledge workers actually need — without dragging Claude Code's developer surface into the same product.

The pace since launch tells the rest of the story. Weekly releases. Skills. Plugins. Connectors. Group-level controls. Each release moves Cowork further beyond a "Claude Code for non-engineers" framing and closer to a category-defining enterprise productivity platform.

More than AI chat

If you've used ChatGPT, Copilot, Gemini, or even Claude.ai, you've used AI chat: a conversation in a tab, with whatever context you can paste in or whatever the chat product happens to fetch.

Cowork is a different shape entirely. It runs as an assistant in your environment — with access to your files, your tools, your skills, your data. The interaction isn't "I'm asking a question." It's "I have a job for you, here's the context, go do it."

That gap matters. Two examples:

• AI chat: You paste a sales call transcript and ask Claude to summarize it.

• Cowork: Cowork pulls the call from your meetings tool, drafts the follow-up email in your voice, attaches the relevant case study from Drive, and updates the deal in your CRM.

The first is a productivity upgrade - which is already great. But the second is a job done. Even better.

This is the same gap between Claude Code and a chat-based code assistant. Once you've worked with the harness, the tab-based version feels like a demo.

The killer angle: a harness, not an agent

Here's where Cowork is doing something most competitors aren't.

The dominant model in enterprise AI has been: embed an AI agent inside an existing tool. Salesforce embedded Agentforce. Microsoft embedded Copilot in every product. Google embedded Gemini in Workspace. The pitch is "AI where you already work."

It's the wrong shape.

A single-tool agent can only do what that tool can do. The agent in your CRM doesn't know about your inbox. The agent in your inbox doesn't know about your Drive. The agent in your Drive doesn't know about your CRM. Each one is fluent in one room and deaf in every other.

Cowork inverts the topology. You bring one harness that talks to all your tools through MCP. Skills, sub-agents, plugins, connectors — they all plug into the same harness, sharing context. The agent isn't trapped inside a vendor's product. It runs alongside you, with the same multi-tool view of your work that you have.

This is the winning position, and Anthropic is driving MCP harder than anyone — which is why Cowork's connector ecosystem is growing so quickly. Google Workspace, DocuSign, Outreach, Clay, Apollo, Harvey, FactSet, WordPress, and a fast-growing list of others are already integrated, all callable from the same conversation, all operating with shared context.

A harness with tools beats an agent in a tool. Every time.

Built for enterprise from day one

The other striking thing about Cowork: how much enterprise plumbing is already in place for a product this young.

A non-exhaustive list of what's shipping today:

• Skills at the org level — admins create skills once and share them across the team or company. Operators don't have to re-invent prompts and workflows individually.

• Plugin starter templates and custom plugins — bundles of skills, commands, and connectors that admins can ship to specific teams.

• Per-group plugin overrides — auto-install a plugin for finance, hide it from everyone else.

• Granular connector access controls — role-based access on which tools and data each group can reach.

• Group spend limits — budget guardrails per team.

• OpenTelemetry observability — stream tool calls, file access, and human-approval decisions to your SIEM.

• Usage analytics — admin dashboard and Analytics API to see which skills, connectors, and patterns are actually getting used.

Most enterprise AI products take years to ship this list. Cowork is doing it on a weekly release cadence, while still in its first year.

Demand is shifting fast

The numbers tell the rest of the story.

Anthropic's annualized revenue has gone from $1B at the end of 2024, to $9B at the end of 2025, to $30B in April 2026 — surpassing OpenAI for the first time — and over $44B by May. Eight of the Fortune 10 are Claude customers. Customers spending over $1M a year on Claude have grown from a dozen two years ago to over 500. Business subscriptions to Claude Code have quadrupled since the start of 2026.

That's the macro view. From where we sit at 9x, the micro view is the same: a clear acceleration over the last few months. Many of the companies we work with are switching to Claude as their primary AI stack and asking us specifically to train their teams on Cowork. The shift was visible quarter over quarter in 2025. In 2026, it's week over week.

Now is the time to get on Cowork

Three things if you've read this far.

The mindset shift is real. Adopting Cowork is not the same as picking a better AI chat tool. The leverage comes from a different way of working — one where you stop typing into a chat box and start handing jobs to an assistant in your environment. I wrote about why this is one of the three tiers of AI skills your team needs in The Three Tiers of AI Skills.

We've been running Cowork programs since February. The companies that started in Q1 are already shipping internal tools, automating workflows, and freeing up serious capacity. If you want the same for your team — or if you just want to talk through where to start — book a call with us.

Our next public cohort starts June 4. Claude Cowork for Growth is a 3-week live cohort for Growth, Marketing, Sales, and RevOps operators — designed to take you from beginner to deploying your first automated workflows on your real stack. Limited seats — €599.