How Anything Uses Claude to Power a No-Code App Builder for 1.5M Users

Anything was founded on a simple but ambitious premise: anyone should be able to build real software, regardless of technical background. Long before their current stack took shape, the team was experimenting with AI-assisted code generation — but the models they relied on were unreliable, prone to outages, and limited in scope. To compensate, engineers built cumbersome multi-stage pipelines that called separate agents to generate code and manually stitched outputs together, while RAG pipelines handled documentation lookups. The result was a system that could produce single-page applications but couldn't come close to the complexity of a real product.

The gap between a simple tool and a production-quality application — one with databases, backend logic, third-party integrations, and mobile deployment — was enormous. Serving a non-technical audience made the challenge even steeper. The agent couldn't just generate code; it needed to manage complex, multi-step workflows autonomously, recover from its own errors during long build sessions, and maintain consistent quality throughout. The team needed a model that could handle all of this reliably at scale before they could even think about improving the user experience.

After running structured evaluations across multiple criteria — instruction compliance, complex application generation, multi-round tool calling, error correction, and hallucination rates — Anything selected Claude as the foundation for their agent. Claude outperformed alternatives across all dimensions, and Anthropic's focused investment in coding capabilities was a decisive factor. Integration via the Claude Agent SDK took just one day, and Claude Opus 4.6's out-of-the-box performance was strong enough that the team skipped the weeks of iteration they'd typically budget for a new model. Claude now powers the core product-building agent, internal subroutines, and operational workflows including support, inbound sales, product feedback analysis, and codebase improvement via Claude Code.

The results have been transformative. In five months, Anything's users built more than 800,000 apps — many of them full-stack, multi-platform products with backends, mobile apps, API integrations, AI features, and monetization layers. The agent maintains a 91–96% success rate as measured by generation completion rates, agent self-reports, and in-product user feedback. Real users are shipping real products: a non-technical founder built and is actively selling PipelinePro, a recruiting ATS with AI interview bots and a client portal; a French film director launched a streaming app with voice-driven discovery; and a user with no engineering background published a social app to the App Store.

Looking ahead, Anything is building toward agents capable of massive parallelization and long-running autonomous processes. The team envisions a future where user-built apps become self-improving — with agents sophisticated enough to handle complexity that users previously had to manage themselves. For Anything, Claude didn't just improve their product; it unlocked an entirely new category of what their platform could deliver.