What Happened When Anthropic Launched Cowork

On February 24, 2026, Anthropic released the enterprise version of Claude Cowork with connectors for Google Drive, Gmail, DocuSign, and FactSet. Wall Street’s reaction was disproportionate — but not irrational.

Weeks earlier, the mere preview of Cowork had already triggered what Jefferies traders called the “SaaSpocalypse”: roughly $285 billion in market cap evaporated in a single day. ServiceNow, Salesforce, HubSpot, Atlassian, Workday — all dropped in unison. Gartner fell 21%, S&P Global 11%.

Why so much panic over a productivity tool?

Because Claude Cowork isn’t a chatbot that answers questions. It’s an agent that does the work. It opens your files, reads them, creates new documents, sends emails, fills out forms. Delegating tasks to an AI assistant no longer requires writing code or hiring an engineering team. It works from any employee’s desktop.

That’s what spooked the market: the middleware layer — the one that connects data to actions — is exactly what Cowork threatens to replace.

What Claude Cowork Is and How It Works

In simple terms: Cowork is a desktop app that gives Claude access to your files, your email, and your work tools. You can queue multiple tasks in parallel, like delegating to a fast junior colleague who never drops a detail.

The key technical difference is MCP (Model Context Protocol), which Anthropic built as an open standard. Instead of fragile point-to-point integrations, MCP defines a standardized way for AI agents to read and write across any external system. According to TechCrunch, the “Deep Connectors” for Google Workspace and DocuSign are the centerpieces of this expansion.

What’s already available:

• Active connectors: Google Drive, Gmail, Google Calendar, DocuSign, FactSet, Apollo, Clay, Outreach, SimilarWeb, MSCI, LegalZoom, WordPress, Harvey
• Department templates: HR, design, engineering, operations, financial analysis, investment banking, equity research, private equity, wealth management
• Enterprise controls: SSO, SCIM, audit logs, compliance API, configurable retention policies

This isn’t a demo. It already has 13 active enterprise plugins and ready-to-deploy templates by department.

Real Use Cases by Department

The department templates are where Cowork becomes practical. In our experience helping B2B teams adopt AI, the question is never “can AI do this?” but rather “what specific task do I delegate first?” Here are four real scenarios.

Operations: The Invisible Work Between Tools

An operations manager we worked with spent roughly 3 hours daily copying data between three systems: CRM, ERP, and spreadsheets for weekly reports. Nobody called that “productive work,” but nobody automated it either because each report had manual exceptions.

With an agent like Cowork connected to Drive and with local file access, that flow becomes: “take this week’s sales data from the CRM export, combine it with ERP inventory, and generate the weekly report in the format we use.” The agent doesn’t eliminate the ERP or CRM — it eliminates the 3 hours of glue work between them.

Finance: From Scattered Data to Consolidated Analysis

Before	With Cowork
Manually download data from FactSet	The agent pulls data via direct connector
Cross-reference in Excel with internal data	The agent consolidates local files + FactSet into a single analysis
Format report for the CFO	The agent generates the document in the usual format
Total time: half a day	Total time: human review (~30 min)

The FactSet connector is particularly relevant for companies with financial analysis teams. Excel isn’t going away — the work of preparing the spreadsheet so someone can analyze it is what gets compressed from hours to minutes.

HR: Documentation Nobody Wants to Do

The pattern in HR is different. It’s not a complicated data flow — it’s the volume of repetitive documents that eats time:

• Updated internal policy drafts
• Offer letters with candidate data
• Onboarding documentation customized by role
• Performance review summaries

A client asked us to evaluate how much time their HR team (3 people) spent on documentation. The number was ~35% of their weekly hours. Most were documents with a fixed structure and variable data — exactly the type of task an agent with access to templates and employee data handles well.

Sales and Marketing: Proposals That Don’t Start from Scratch

• Before: A salesperson spends 2 days assembling a commercial proposal. They search Drive for previous presentations, copy relevant sections, update client data, adjust pricing, format.
• With Cowork: “Generate a proposal for [company X] based on last month’s proposal for [company Y]. Update prices from the current catalog and add the compliance section we use for financial clients.”
• Result: The draft comes out in minutes. The salesperson reviews, adjusts the tone, and sends. Close time doesn’t change — preparation time does.

The Market Reaction Isn’t Overblown

It’s easy to dismiss a $285 billion drop as irrational panic. It’s not entirely.

The value proposition of many enterprise SaaS tools is, at its core: “we connect data on one side with actions on the other, with a nice interface in the middle.” When an AI agent can read your Drive files, cross them with FactSet data, and generate a formatted report on your desktop — what part of the SaaS stack remains indispensable?

Not the interface. Not the integration layer. What remains is the system of record (the CRM, ERP, HRIS) and the business logic inside it. But the middleware — the dozens of tools that exist to move data between systems and format it — is exactly what Cowork compresses.

Dario Amodei had to do a media tour on February 24 alongside Marc Benioff of Salesforce to calm the market with the “augmentation, not replacement” narrative. And it’s true that Cowork won’t kill Salesforce tomorrow. But it does change the math on how many SaaS licenses a 200-person company needs.

In our experience, the companies benefiting first aren’t replacing tools — they’re eliminating the glue work between them. That work is currently done by people, and it consumes ~25% of an average operations team’s hours.

How to Prepare Your Team

Don’t wait for Cowork to be perfect before you start getting ready. The product will improve. Your team needs time to learn how to delegate.

Identify delegable tasks. Every department has tasks that are repetitive, follow a predictable structure, and are currently done by someone who’d rather be doing something else. Those are your candidates. In HR it’s documentation. In finance it’s report prep. In sales it’s proposal assembly.

Define the human review line. An agent can generate a proposal draft, but who approves it before it goes to the client? The sales manager or the rep? This isn’t an AI question — it’s a process question. If you don’t define it before implementation, you’ll have people approving documents they haven’t read.

Establish internal governance. If Claude Cowork has access to corporate files, you need clear policies: what documents it can read, what actions it can execute unsupervised. Someone specific must own the authority to expand those permissions — don’t leave that ambiguous. Anthropic’s enterprise controls (SSO, audit logs, compliance API) are necessary but not sufficient — you need internal rules.

Training isn’t about prompts. Forget “prompt engineering.” What your team needs to learn is how to spot delegable work and break it into clear instructions. Then they need to review results critically — questioning the output, not just accepting it. It’s work management, not technical skill.

At IQ Source, we help teams identify their first 5 delegable tasks per department. It doesn’t take weeks — it takes a focused work session with the people who know the day-to-day.

If you want to better understand how AI agents fit into enterprise operations, our practical agent playbook for decision-makers explains the full framework. And if the challenge is more about your team’s skills than the technology, we have an analysis of the AI fluency gap that applies directly.

The Gap Isn’t Technological

Claude Cowork already works. The connectors are active. The templates exist. The pricing is published. The gap isn’t whether AI can do the work — it’s whether your team knows what work to delegate.

The companies that will gain an advantage over the next 6 months won’t be the ones buying more AI licenses. They’ll be the ones that do an honest inventory of where their people’s time goes, identify the glue tasks, and start delegating the simplest ones first.

If you want to map which of your team’s tasks are candidates for an AI assistant like Cowork, let’s do a delegation diagnostic — it’s shorter than you think.