Anthropic is continuing rapid growth in the enterprise market. Talk is also growing that it is threatening major players that have long dominated the enterprise software market.

With annual recurring revenue for its AI coding tool Claude Code topping $2.5 billion, Anthropic has also recently launched Claude Cowork. It has moved to target the workflow automation market beyond developers.

As it grows, what competitive edge will Anthropic have beyond model quality?

Unlike Microsoft, which operates on top of Microsoft 365, or Google, which builds around Workspace and BigQuery, Anthropic is in a position where it must create reasons from the outset for companies to keep using it. On that point, Foundation Capital partner Jaya Gupta (자야 굽타) highlights “state” as a keyword that becomes more important as AI burrows deeper into corporate work.

In a post she recently shared on social media platform X (Twitter), she wrote that state is what accumulates in AI systems over time. Examples include how a company makes decisions, what exceptions occurred and how employees work.

Gupta said, “Once enough state accumulates, switching AI vendors becomes not a simple purchasing decision but an organisation-wide issue.”

State accumulates in 4 forms. The first is system configurations tailored to a specific model’s approach. Costs arise the moment it is changed. The second is what AI remembers and learns beyond a session. Gupta said this compounds the fastest. The third is accumulated understanding of how an organisation functions. She said this can be the most sensitive part. The fourth is understanding of how a specific person thinks and works. Gupta said, “It takes the longest, but it is the hardest element to replace.”

Strategies by major companies diverge over state.

Gupta said, “Anthropic and OpenAI try to own state by placing execution environments and memory systems behind the API. Companies use AI capabilities, but the most important state accumulates inside systems that companies cannot look into or extract.”

Microsoft already has state in the emails, documents and meeting minutes that companies use. Copilot is an interface that accesses that state. Google is focusing on connecting the data companies have already accumulated inside Google with AI as one. Gupta said, “Emails and documents in Google Workspace, analytics data in BigQuery, and models in Vertex AI all run inside Google. Gemini becomes a single window that accesses all of this. What has accumulated inside Google is hard to move without Google.”

Databricks chose to place state in infrastructure that companies own directly and to use AI models only as tools that read it. Gupta said companies own the state and the model becomes a replaceable part.

The battle over state is not new. Email felt personal, but became a corporate asset as legal obligations to preserve evidence emerged. Slack messages and recorded sales calls were the same. Important state ultimately became something companies claim ownership of. AI state is the same fight, though the scale and stakes are bigger. Gupta stressed, “When a company chooses an AI architecture now, it is really deciding who will own the state that accumulates around it. Most companies think they are choosing a model, but in reality they are deciding something much bigger.”

Back to Anthropic. Gupta said Anthropic does not have something already embedded in corporate systems like Microsoft or Google. Model quality alone makes it hard to give companies a reason to keep using it.

At this point, Anthropic appears to be emphasising a safety-first philosophy to target the enterprise market. But safety-first thinking cannot be a sustainable and certain moat. Gupta said, “A safety-first philosophy is a way to buy time while Anthropic builds a moat,” adding, “If it builds the moat in time it will look prescient, but if not it will be no more than filling with words what it failed to show in a product.”