Anthropic is now the top choice for business customers when it comes to AI, having taken 32% of enterprise usage from a company that had just 12% two years ago. OpenAI, which once dominated with 50% of the corporate market, now sits at 25%. This reversal happened almost entirely because of one product category: software development tools. In coding-related work, Anthropic's Claude holds 42% of enterprise usage compared to OpenAI's 21%.
The revenue numbers confirm the shift is real, not just survey data. At the start of 2025, Anthropic was generating about $1 billion a year. By the end of 2025, that number had climbed to $9 billion. By early 2026, it had crossed $30 billion. The company now has over 1,000 clients each spending more than $1 million a year on Claude, and eight of the ten largest companies in the world are paying customers.
This trajectory is why investors are reportedly in discussions to value Anthropic at close to $900 billion, which would pull it ahead of OpenAI's $852 billion valuation from earlier this year. That is a remarkable position for a company that most consumers have never heard of.
The product logic behind this growth is straightforward. Anthropic moved in a specific direction: start with developers, prove the tool works in real production environments, then expand to all office workers. Claude Code came first, built for software teams. Then came Claude Cowork, launched in January 2026, which extends the same idea to anyone who works with files, emails, documents, and data. Cowork connects to Gmail, Google Drive, DocuSign, and FactSet, and can read your files and handle multi-step tasks on your behalf.
Microsoft noticed. It built its own version, Copilot Cowork, and rolled it out in March 2026. The fact that the world's largest enterprise software company felt the need to respond to Anthropic's product naming says something about how much ground Anthropic has taken.
The part that deserves the most attention is where this goes next. Anthropic's Cat Wu describes the next step as "proactivity." Right now, AI tools respond when you ask them something. The step after that is routine automation, where you set up a process once and the AI runs it repeatedly, like handling customer support tickets. The step after that, which Wu says is coming, is where the AI watches how you work and starts building those automations for you without being asked.
This is meaningfully different from anything currently available. Today's AI tools are reactive. What Anthropic is describing is an AI that studies your work patterns and then, effectively, starts managing your workflow calendar, your recurring reports, your standard client communications, without you instructing it to do so.
Gartner predicts that by 2029, AI agents will autonomously resolve 80% of common customer service issues without human involvement. That timeframe may be conservative given the pace Anthropic is moving.
The real question this raises for any business operator is about control and accountability. When an AI sets up an automation you did not explicitly request, who is responsible for what it does? The people who manage these systems will need to understand the work deeply enough to audit it, catch errors, and course-correct. That is not a technology skill. It is a management skill applied to a new type of team member.
What is already clear is that the companies spending nothing on understanding these tools are falling behind the ones that are. The gap between them is widening every quarter, not every year.