What Claude Cowork Actually Does

Announced March 19, 2026, Claude Cowork represents a meaningful step beyond the single-session AI assistant model that has defined the first generation of AI productivity tools. The core capability is persistent project context: Claude Cowork maintains awareness of ongoing work across multiple sessions, eliminating the “context reset” that currently forces users to re-explain their situation every time they open a new conversation.

To appreciate why this matters, consider the difference between having a capable colleague who forgets everything at the end of each workday versus one who remembers where you left off. The former is useful for bounded, self-contained tasks. The latter is useful for sustained, multi-week projects — the kind of work that actually defines professional output for knowledge workers in legal, financial, engineering, and research roles.

Claude Cowork’s framing as a “productivity tool for knowledge workers” is deliberate and strategically important. Anthropic isn’t positioning it as an AI assistant (a category with consumer connotations and price sensitivity) but as a professional tool in the same category as Notion, Figma, or Salesforce — software that professionals pay enterprise rates for because it’s integrated into how they work, not just how they browse.

The deeper implication is for autonomous work. Persistent context is a necessary precondition for meaningful multi-session autonomous AI assistance — the kind where an AI agent can genuinely be delegated a multi-day research or analysis project, return to it across sessions, and maintain coherent progress. Claude Cowork isn’t fully autonomous AI yet, but it’s building the infrastructure that makes that a near-term possibility rather than a distant aspiration.

For enterprise buyers evaluating AI productivity tools, Claude Cowork addresses one of the most common objections to current AI adoption: “It’s useful for individual tasks, but it can’t replace the institutional memory that a junior analyst builds up over months.” Persistent project context is a direct answer to that objection — and positions Claude as a genuinely more capable professional partner rather than a sophisticated search engine.

The Anthropic Institute’s Role

The Anthropic Institute, launched in March 2026, is a research body specifically focused on the challenges that powerful AI poses at societal scale — distinct from Anthropic’s product development and separate from the safety research that informs Claude’s training. The most useful comparison is to DeepMind’s research arm: world-class researchers working on problems that are two to five years ahead of the current product cycle, funded by commercial revenue but operating with academic independence.

Why does this structure matter strategically? Because the hardest regulatory and institutional relationships to build for an AI company are with governments, healthcare systems, financial regulators, and legal frameworks. These institutions move slowly, require extensive evidence bases, and distrust organizations that appear to be primarily motivated by commercial interests. A credible research institute that produces independent, peer-reviewed analysis of AI’s societal impacts is one of the most effective tools an AI company can have for building those relationships.

The Anthropic Institute is doing for Anthropic what Bell Labs did for AT&T: creating a research halo that validates the parent company’s claim to operate in the public interest, while also producing genuine intellectual output that shapes how policymakers and regulators think about AI governance. This is a decades-long credibility investment, not a quarter-by-quarter product feature.

The practical consequence for Anthropic’s business: organizations in regulated industries — hospitals, banks, law firms, government agencies — can justify selecting Anthropic as an AI vendor in part because Anthropic is a credible participant in the governance conversation, not just a technology vendor. That credibility is genuinely hard to replicate, even with billions in marketing spend. It requires time, research output, and relationships that are built over years rather than launched in a product keynote.

The Economic Index: Real Evidence of AI Productivity

The Anthropic Economic Index, published in March 2026, is more significant than it appears at first read. Most AI company claims about productivity impact are anecdotal — “our customers say they save X hours” — which is useful marketing but not evidence. The Economic Index is different: it uses privacy-preserving analysis of actual Claude usage patterns to produce the first systematic, sector-by-sector evidence of AI’s economic impact at scale.

The sectors showing the strongest productivity signals in the March 2026 report are consistent with Claude’s known enterprise customer base: legal (contract review, research, drafting), software engineering (the Claude Code story), and financial analysis (due diligence, modeling support). These aren’t surprising findings — they’re the sectors where knowledge work is most language-intensive and where Claude’s capabilities map most directly to existing workflows.

The privacy-preserving methodology matters as much as the findings. Anthropic designed the Economic Index to produce aggregate insights without exposing individual user data — a methodological choice that reflects constitutional AI principles applied to data analysis, not just model training. For regulated industry customers concerned about data handling, this demonstrates that Anthropic takes privacy seriously as an engineering constraint rather than a compliance checkbox.

The strategic use of the Economic Index is as a sales tool for enterprise buyers who want to justify Claude adoption to their boards. “Here is peer-reviewed evidence that organizations in your sector that use Claude see specific productivity improvements” is a fundamentally more compelling enterprise pitch than “our model scores highest on benchmarks.” The Economic Index turns Anthropic’s user data into evidence that drives adoption among precisely the high-value enterprise customers they’re targeting.

Why $2.5B From Claude Code Is the Key Signal

Claude Code generating $2.5 billion in ARR as the fastest-growing B2B AI tool in history is the most important single data point in understanding Anthropic’s competitive position. Here’s why: it proves that Anthropic can build a dominant product in a competitive, technically sophisticated market (software development) against OpenAI’s Codex/GPT-4o, GitHub Copilot, and Google’s Gemini code tools.

Software developers are among the most demanding, most skeptical, and most vocal technology consumers on the planet. When they adopt a tool at enterprise scale — embedded in their IDEs, integrated into their CI/CD pipelines, used in their daily terminal workflows — it’s because the tool genuinely makes them more productive, not because of marketing. Claude Code’s $2.5B ARR is validated by the behavior of technically sophisticated users, not marketing surveys.

The enterprise customers confirm the thesis. Deutsche Telekom, Revolut, Meta, and Salesforce using Claude in production deployment means Anthropic’s safety-focused, enterprise-grade approach is passing the scrutiny of organizations with serious compliance requirements and significant reputational stakes in getting AI governance right. These are not early adopter experiments — they’re production deployments at scale.

The $2.5B Claude Code figure also contextualizes Anthropic’s broader $14B ARR. Claude Code represents approximately 18% of total revenue — a significant and growing product line that de-risks Anthropic against any single revenue concentration risk. As Claude Cowork scales, this diversification will increase, giving Anthropic the financial resilience to sustain safety-first product development without the kind of consumer adoption pressure that forces competitors to ship faster than their safety review processes can accommodate.

Claude Opus 4.6 maintaining the #1 position on Terminal-Bench 2.0 and Humanity’s Last Exam — the most demanding capability evaluations available — while Anthropic invests in safety and governance infrastructure is the proof point for Constitutional AI as a development methodology. Safety and capability are not as deeply in tension as critics of the safety-first approach claim. Anthropic’s benchmark record is the empirical evidence.

Anthropic vs OpenAI: Two Bets That Are Both Right

The AI commentary landscape has a tendency to frame the Anthropic vs OpenAI competition as a zero-sum contest where one safety-focused enterprise AI company and one consumer-scale platform are fighting for the same customers. This framing is both commercially useful (it drives coverage) and analytically wrong.

OpenAI’s 900 million ChatGPT users represent a consumer mindshare advantage that Anthropic has explicitly not pursued. OpenAI’s developer ecosystem — millions of developers building applications on the GPT API — creates network effects that are genuinely hard to disrupt. OpenAI’s government contract expansion is real revenue and real strategic positioning. These are genuine competitive advantages in the consumer-to-prosumer segment of the AI market.

Anthropic’s $14B ARR growing 10x/year, Claude Code’s $2.5B, and customers like Deutsche Telekom, Revolut, and US government agencies represent a different but equally real competitive position in the regulated enterprise segment. The regulated industries — healthcare, finance, legal, government — have procurement processes that are long, risk-averse, and heavily dependent on vendor trust and compliance history. Anthropic’s Constitutional AI methodology, the Anthropic Institute, and the Economic Index are all specifically calibrated to build the trust that enables these sales. OpenAI’s consumer focus means they’re less optimized for this procurement process, even if their models are technically competitive.

The distillation lawsuit deserves mention as a signal of Anthropic’s strategic boundaries. By pursuing DeepSeek, Moonshot AI, and MiniMax for allegedly illicit model distillation via 16 million fake account queries — and forgoing $100M+ in China revenue to do so — Anthropic is demonstrating that IP protection and market integrity are genuine strategic priorities, not just rhetorical ones. This matters for enterprise buyers: an AI vendor that aggressively defends its model quality and IP is one that’s investing in maintaining the capability advantage that enterprise customers depend on.

The most interesting scenario over the 2026-2028 horizon isn’t Anthropic displacing OpenAI or vice versa — it’s a market that segments along the lines that enterprise software markets always have. Consumer and SMB customers will trend toward whoever offers the best price-performance-features combination in a commoditizing consumer AI market. Regulated enterprise customers will trend toward the vendor with the deepest institutional trust, compliance infrastructure, and governance track record. That is, by design, Anthropic’s lane.

Anthropic vs OpenAI: Strategic Positioning Compared

Frequently Asked Questions