Coframe Raises $25 Million to Let a Team of AI Agents Rewrite, Test and Redeploy a Website Around the Clock

Coframe, a San Francisco company that wants websites to rewrite themselves, has raised a $25 million Series A round co-led by Khosla Ventures and Hanabi Capital, with participation from executives at OpenAI, Anthropic and Google DeepMind. The round brings total disclosed funding to $35 million. No valuation was announced.

Founder and chief executive Josh Payne, who previously co-founded the celebrity-licensing startup Autograph with Tom Brady and wrote an early autonomous coding agent called GPT-Migrate, describes the product in one line: "We're building an internet that improves itself." In practice, Coframe plugs into a company's existing site or ad campaigns and keeps generating, testing and deploying variations without waiting for the next manual optimization cycle.

Under the hood sits a transformer model trained on user interactions across hundreds of millions of sessions per month, wrapped by a set of agents in fixed roles: strategist, designer, software engineer, quality assurance and data analyst. The system proposes changes, runs them as live experiments, and uses the results to steer the next round. The company claims a 39 percent experiment win rate, which is a notably honest number to publish in a category where vendors usually quote only the winners.

The headline claims are large and entirely company-reported: more than $300 million in incremental customer revenue, average conversion improvements above 20 percent, and client-specific lifts of 158 percent at Dropbox, 410 percent at Replit and 60 percent at Cash App. Coframe also says a Fortune 500 client saw conversions rise 70 percent within 30 days of signing. None of these figures has been independently audited, and conversion attribution in always-on testing is famously easy to overstate. Customers named across the announcements include Dropbox, Replit, Cash App, BCG, Block, Backcountry and Sur La Table.

Two details distinguish Coframe from the crowded A/B testing market it is trying to replace. First, it is working with OpenAI to optimize ChatGPT ads and their landing pages as a single loop, an arrangement that matters if advertising increasingly happens inside chat interfaces. Second, the company frames its buyer not as a marketing team that wants more variants, but as one that wants the coordination work between analysts, copywriters, designers and developers to disappear.

The competitive field is real. Adobe Target, Optimizely and VWO own the enterprise experimentation budget today, and several newer startups are automating pieces of the same workflow. Coframe's bet is that owning the full loop, from hypothesis to deployed variant to measured result, is the durable position, with the roadmap extending from websites into apps and email. If agentic browsing grows, the same machinery may eventually have to optimize pages for AI shoppers rather than human clicks, which is a different objective function entirely.