Anthropic is positioning its new Claude Cowork platform as the missing piece between experimental AI agents and dependable enterprise-grade automation. At its virtual “Briefing: Enterprise Agents” event, the company argued that 2025’s rush into AI agents mostly yielded stalled pilots and proofs of concept — not production outcomes. The company’s message to technology leaders: the problem was not effort, but approach, and Anthropic now believes it has a repeatable model, drawn from the rapid enterprise adoption of Claude Code.
From Claude Code to Claude Cowork: Anthropic’s new enterprise thesis
Anthropic’s head of Americas, Kate Jensen, framed the company’s recent trajectory as a two-stage playbook: first developers, then knowledge workers. In 2025, she said, Claude “transformed how developers work.” In 2026, Anthropic expects the same shift for knowledge work through Claude Cowork, its AI productivity environment that entered research preview in January.
The core idea is delegation of genuinely hard work, not just assistance with superficial tasks. Scott White, head of product for Claude Enterprise, described Cowork’s ambition as enabling Claude to deliver “polished, near final work” — complete projects and deliverables, rather than simple drafts and suggestions. That statement sets a high bar for outcomes and raises implied questions for CIOs about quality control, governance, and change management.
Jensen also pushed back on the narrative that early enterprise AI agents simply “failed.” In Anthropic’s view, most of 2025’s efforts ran into architectural and organizational limitations: agents were bolted onto workflows rather than embedded into systems and data, and they stayed generic instead of adapting to company-specific processes. Claude Cowork, she said, is meant to embed deeply enough that employees can offload complex challenges and refocus on higher-value work.
What’s actually new in Claude Cowork for enterprises
Beneath the positioning, Anthropic announced a dense set of capabilities designed to make AI agents feel native to each organization’s tools and standards.
Private plugin marketplaces. Enterprise administrators can now stand up internal marketplaces that expose only approved plugins. These marketplaces can connect directly to private GitHub repositories as plugin sources and give IT leaders fine-grained control over which tools are available to which employees. The intent is to avoid the “one-size-fits-all bot” and instead reflect how each company already works.
Sector-specific plugin templates. Anthropic is shipping prebuilt templates that target concrete workflows across HR, design, engineering, operations, financial analysis, investment banking, equity research, private equity, and wealth management. While details of each template were not broken down, the breadth signals an effort to reduce the initial build cost for common industry patterns and narrow the gap between generic models and domain workflows.
Expanded MCP connectors. The Model Context Protocol (MCP) now reaches deeper into the SaaS stack enterprises already rely on. New connectors span Google Drive, Google Calendar, Gmail, DocuSign, Apollo, Clay, Outreach, SimilarWeb, MSCI, LegalZoom, FactSet, WordPress, and Harvey. With MCP, Claude can pull and act on context across these systems rather than waiting for users to paste or upload information manually.
Cross-application context for office work. Claude can now pass context seamlessly between Cowork, Excel, and PowerPoint, even across multiple files, without requiring users to restart when switching applications. For knowledge workers, this is positioned as a step toward continuous multi-document workflows: for example, using Claude to synthesize source materials in Cowork, analyze data in Excel, and generate or revise presentations in PowerPoint while preserving shared context.
White summarized the design goal clearly: enterprises do not want “Claude for legal” in the abstract — they want “Cowork for legal at your company.” The platform is being pitched less as an out-of-the-box product and more as a configurable layer that conforms to each organization’s structures, terminology, and controls.
Spotify, Novo Nordisk, Salesforce: early signals from production deployments
To substantiate the claim that Claude-based systems can escape the prototype phase, Anthropic highlighted three large-scale deployments spanning software, pharmaceuticals, and enterprise collaboration.
Spotify: large-scale code migration. Spotify has historically struggled with the slow, manual work of migrating and modernizing code across thousands of services. After integrating Claude directly into engineers’ daily systems, the company says “any engineer can kick off a large-scale migration just by describing what they need in plain English.” Reported outcomes include up to a 90% reduction in engineering time, more than 650 AI-generated code changes shipped per month, and roughly half of all Spotify updates now passing through the system. For technology leaders, this suggests AI agents can become core to the software supply chain rather than peripheral helpers.
Novo Nordisk: regulatory documentation at pharmaceutical scale. Novo Nordisk built an internal platform, NovoScribe, with Claude as the intelligence layer to tackle the laborious process of generating regulatory documentation for new medicines. According to Jensen, staff writers previously produced just over two reports per year; with Claude, documentation creation dropped from 10+ weeks to about 10 minutes, with a reported 95% reduction in resources needed for verification checks. The company also used Claude Code to build NovoScribe itself, enabling contributions from non-engineers — including a digitalization strategy director with a PhD in molecular biology who now prototypes features using natural language. Jensen characterized the 11-person team behind NovoScribe as “operating like a team many times its size.”
Salesforce: AI inside Slack. Salesforce is using Claude models to help power AI features in Slack. The company reports a 96% satisfaction rate for tools such as its Slack bot and estimates that customers save about 97 minutes per week through summarization and recap features. Anthropic highlighted Salesforce, along with others, as “Claude partners and domain experts with the data and trusted relationships that make Claude work in the real world,” underscoring its reliance on ecosystem partners to bring AI into frontline workflows.
For enterprises evaluating Anthropic, these deployments are still early but show Claude being embedded into core production systems — engineering workflows, regulatory processes, and collaboration platforms — rather than sitting on the edge as an experimental chat interface.
Inside the boardroom: leaders confront accountability, risk, and change management
A panel with executives from Thomson Reuters, the New York Stock Exchange (NYSE), and healthcare technology firm Epic offered a more candid view of what it takes to operationalize AI at scale — and the organizational friction that comes with it.
NYSE: from deterministic systems to probabilistic accountability. Sridhar Masam, CTO of the NYSE, described the exchange as “rewiring our engineering process” using Claude Code and internal agents built on the Claude Agent SDK. These agents can take instructions starting from a Jira ticket and move all the way to committed code. But Masam emphasized a shift in how leaders must think about responsibility. In deterministic systems, accountability often ends once a project goes live. With probabilistic AI, he argued, it extends into continuous monitoring of behavior and outcomes. He also described a move beyond the classic “buy versus build” decision to a new concept of “assembly” — combining multiple models, vendors, platforms, data sources, and internal capabilities into composite solutions. For heavily regulated organizations, he said, the stance must shift “from risk avoidance to risk calibration,” since opting out of AI is no longer competitively viable.
Thomson Reuters: tools outrunning change management. Steve Haske, speaking about the company’s Co-Counsel product (now at one million users), bluntly noted that “the tools are in many senses ahead of the change management.” Legal, tax, accounting, and audit organizations, he said, need to “rewire the processes” to capture the benefits. He estimated it could take roughly 18 months for organizational change management to catch up to the tools’ capabilities. Haske also stressed an “ironclad guarantee” that Co-Counsel customers’ inputs will not be reused as outputs and urged leaders to be “feverish” about protecting institutional intellectual property — a point likely to resonate with CIOs concerned about data leakage.
Epic: non-developers as primary AI users. Seth Hain from Epic highlighted a notable adoption pattern: “Over half of our use of Claude Code is by non-developer roles across the company.” Support and implementation staff have embraced the tool in ways Epic did not initially anticipate. To build trust with clinicians, Epic’s first AI feature was a medical record summarization tool that surfaced links to underlying source material, enabling verification. Only once that trust was established did the company move toward more autonomous agent capabilities. For enterprise leaders, this underscores the importance of starting with transparent, auditable use cases before scaling automation.
MCP and the platform shift: why this feels different from last year’s chatbots
Anthropic’s announcements are the culmination of a year in which the company moved from research-focused lab to full enterprise platform vendor. Two developments explain why this moment feels different from 2025’s chatbot experiments: Claude Code’s evolution and the spread of the Model Context Protocol.
Jensen said Claude Code moved coding use cases “from assisting on tiny tasks to AI writing 90 or sometimes even 100% of the code, with enterprises shipping in weeks what once took many quarters.” But the deeper change is MCP, which acts as connective tissue for Claude across organizational systems. Rather than being limited to user-pasted content, MCP-connected Claude can fetch and reason over data from Slack threads, Google Drive documents, CRM records, and financial systems at the same time.
This context access underpins the new plugin architecture. Claude is not framed as an isolated chat interface but as a reasoning layer that sits across an organization’s infrastructure. Private plugin marketplaces, portable file-based plugins, and a growing list of MCP connectors echo the ecosystem strategies of platforms like Salesforce and Microsoft — but Anthropic is attempting to compress into months the ecosystem build-out that those firms executed over years.
That platform ambition has already rattled public markets. IBM shares dropped nearly 13.2% in a single day after Anthropic published a blog post showing how Claude Code can modernize COBOL, the aging language underpinning many IBM mainframe systems. Enterprise software stocks, including ServiceNow, Salesforce, Snowflake, Intuit, and Thomson Reuters, had already been under pressure since Claude Cowork’s January 30 announcement. Cybersecurity firms sold off after Anthropic introduced Claude Code Security on February 20.
Yet the latest event produced a partial rebound. Companies named as partners and integration targets — Salesforce, DocuSign, LegalZoom, Thomson Reuters, FactSet — all rallied, with Thomson Reuters jumping more than 11%. Markets appear to be drawing a new line: vendors embedded in Anthropic’s orbit may benefit, while those outside it face sharper questions about their role in an AI-native stack.
To balance product optimism, Anthropic’s head of economics, Peter McCrory, presented findings from the Anthropic Economic Index, which uses privacy-preserving methods to study how people and businesses across more than 150 countries and every U.S. state are using Claude.
One headline statistic: a year ago, about one-third of U.S. jobs had at least 25% of their tasks reflected in Claude usage data. That figure is now roughly one in two jobs. McCrory characterized AI as a “general purpose technology” in the economic sense, indicating that almost every sector of the economy will be affected.
He drew an important distinction between automation (Claude executing tasks end-to-end) and augmentation (Claude collaborating with a human on more complex work). In embedded, API-based deployments, the pattern Anthropic sees is “overwhelmingly” toward automation, mirroring how prior transformative technologies have diffused through the economy.
McCrory was careful on the topic of job loss. Roles that typically require more years of schooling are currently seeing the largest productivity gains — a pattern known as skill-biased technical change. However, he expressed concern about “jobs that are pure implementation,” pointing specifically to data entry workers and technical writers, where tasks central to those jobs are already being handled by Claude. He emphasized that Anthropic has not yet observed clear evidence of widespread displacement but said forthcoming research will focus on detecting whether highly exposed workers begin experiencing it.
For enterprise leaders, McCrory’s key message was that model capability is only part of the equation. Organizations must build the “right sort of data ecosystem” and infrastructure so that Claude can access the information needed to perform sophisticated tasks. If essential knowledge lives only in employees’ heads, he argued, “that’s not a technical problem, per se. That’s an organizational problem.”
Strategic takeaways for CIOs: the emerging “thinking divide”
Jensen introduced the idea of a “thinking divide” — the widening gap between organizations that embed AI simultaneously across employees, processes, and products, and those that adopt it as a narrow point solution. The former group, she argued, will see compounding advantages over time; the latter risks falling progressively behind.
The current landscape remains ambiguous. It is not yet clear whether Anthropic will function primarily as an enabling platform that lifts much of the enterprise software ecosystem or as a disruptive force that erodes the value of some incumbent applications. Tuesday’s event, which both strengthened the case for Anthropic’s named partners and intensified scrutiny on unaligned vendors, underscores that duality.
McCrory urged humility, noting that capabilities are “moving very, very quickly” and may represent “an innovation in the method of innovation” — not only making organizations better at existing processes but helping them discover new ways of working altogether.
Thomson Reuters’ Haske translated that urgency into a leadership imperative: executives need to become “personally involved and personally invested” in using these tools and “move fast” in an environment that is evolving rapidly. Jensen recounted a Fortune 10 CIO’s assessment that enterprises must compress a decade of innovation into the next few years — and their claim that, with Anthropic, they intend to do it in a single year.
Whether that level of confidence proves accurate or premature, the direction of travel is clear. For enterprise technology leaders and knowledge-work managers, the window to decide how — not whether — to integrate platforms like Claude Cowork into production workflows is closing faster than most boardrooms have planned for.
Hi, I’m Cary Huang — a tech enthusiast based in Canada. I’ve spent years working with complex production systems and open-source software. Through TechBuddies.io, my team and I share practical engineering insights, curate relevant tech news, and recommend useful tools and products to help developers learn and work more effectively.
