Anthropic Competes With Its Customers

Anthropic is no longer just selling the engine, it is moving up to sell the whole car. That changes the relationship with application companies that built their product on Claude, because those customers now face the risk that their model supplier can launch a first party product in the same workflow, use its brand to win distribution, and force them to reconsider whether Anthropic should remain their default provider at all.1 2 3 4 5

• The Spellbook example makes the tension concrete. Spellbook had been using Claude for more than 500,000 monthly contract reviews, then moved much of its model spend to OpenAI after Anthropic entered legal, a direct signal that application builders will switch suppliers when a model vendor starts competing for the same budget.1 2 3
• Anthropic has been expanding from model access into packaged workflows. By May 2026 it had launched Claude for the legal industry with integrations into tools like Relativity and legal data providers, and internal research also shows legal plugins inside Claude Cowork. That means Anthropic is not just offering a raw model, it is shipping opinionated product on top.4 5 6
• This is the same pattern already visible in coding. Claude Code turned Anthropic from supplier to competitor for customers like Cursor, while OpenAI pursued Windsurf to own more of the application layer too. The broader market is converging on vertical integration, where labs want model margin, user distribution, and workflow data in one stack.7 8 9 10

Going forward, more AI application companies will hedge by going multi model or shifting spend to whichever lab is least threatening in their category. That pushes Anthropic and OpenAI into a tradeoff between short term application revenue and long term platform trust. The winners at the application layer will be the ones that own a workflow deeply enough that swapping the underlying model does not break the product.1 3 7 8 10