Replit co-founder and chief executive Amjad Masad revealed the split during an appearance at TechCrunch Disrupt 2026, confirming that the company is no longer working with Anthropic. The announcement caught many in the AI coding space off guard, given how closely Replit had leaned on Anthropic's Claude models to power its AI-assisted software development tools over the past couple of years.
Replit built much of its recent reputation on letting users describe an app in plain language and watch it get built automatically, a workflow that depended heavily on large language models capable of writing and debugging code. Anthropic's Claude family had been a central engine behind that experience, making the companies' relationship a closely watched example of how AI labs and application-layer startups divide value in the fast-growing "vibe coding" market.
Masad did not use the Disrupt stage to lay out a detailed roadmap for what comes next, but the move raises immediate questions about which models will now power Replit's products. AI coding tools live or die by the quality, speed, and cost of the underlying models, and swapping out a foundation partner is rarely simple. Companies in this position typically pivot toward a rival lab, build more of their own infrastructure, or adopt a multi-model approach that lets them switch providers depending on price and performance.
The breakup also reflects broader turbulence in the relationships between AI model developers and the startups built on top of them. As foundation model companies increasingly release their own consumer and developer-facing coding products, tension has grown between labs and the application companies that once depended on them as customers. A visible partner like Replit walking away from Anthropic could signal that more of these alliances are becoming strained as competitive lines blur.
For Anthropic, losing a prominent customer in the coding tools space is a reputational data point, even if the company has no shortage of other enterprise and developer partnerships. For Replit, the challenge now is proving it can maintain product quality without the model that helped define its offering. Investors and developers will be watching closely for Replit's next move, including whether it names a new primary model provider or leans further into building proprietary AI capabilities in-house.