Coframe Raises $25M for AI Website Experimentation | Magica
Coframe Raises $25M to Automate Website Experimentation
Editorial Team
••📖6 min read
Coframe's $25 million Series A backs AI agents that build and test websites and ads. Its offering combines automation with human expertise and customer approvals, while its claimed gains lack comparable measurement detail.
Coframe raised a $25 million Series A co-led by Khosla Ventures and Hanabi Capital, bringing total funding to $35 million.
Its service combines AI experiment generation with human experts and customer approvals; automated personalization already has established alternatives.
The startup's claimed customer revenue and conversion gains remain unverified independently.
Coframe, a San Francisco startup that uses AI to generate and test websites, landing pages and ads, announced the financing on November 9, 2026. The round backs an effort to automate more of the work between a marketing idea and a deployed experiment. For buyers, the question is how much that reduces the cost of securing a measurable improvement over their existing tools and teams.
Coframe's $25 million Series A follows an approximately $9.3 million seed round in October 2024. Source: Crypto Briefing.
Funding a faster experiment cycle
Venture investors Khosla Ventures and Hanabi Capital co-led the round, Coframe said in its announcement. Participation included leaders at OpenAI, Anthropic and Google DeepMind. That description identifies individuals associated with those organizations; it does not establish that the organizations themselves invested.
The financing follows a roughly $9.3 million seed round in October 2024, according to the funding report. The new valuation was not disclosed. A deal analysis places the Series A in the 90th percentile of 5,242 comparable US enterprise software rounds at that stage, although it does not specify the comparison period.
Founder and chief executive Josh Payne started Coframe in 2023 after co-founding Autograph with Tom Brady. Autograph was a platform for non-fungible tokens, or NFTs; Coframe says it reached a valuation above $2 billion. Coframe also credits Payne with creating GPT-Migrate, an early autonomous coding agent. That experience in software generation is relevant to a service intended to perform the design and engineering work needed to put experiments into production.
ByteDance is reportedly accelerating plans to combine its Spell project with Doubao. Existing travel bookings reveal a transaction-fee business, while the phone assistant's history shows how access to other apps can constrain execution.
According to Coframe, a specialized transformer model trained on user interactions works with agents assigned strategy, design, software engineering, quality assurance and data science roles. The system can generate, test and deploy hundreds of website and advertising variations in minutes. Its proprietary user-behavior model is trained on activity spanning hundreds of millions of sessions each month, the company says. That is a claimed data footprint, not a disclosed count of paying customers.
Coframe's product illustration connects a brand kit to an example spring-sale landing page. Source: Coframe.
Personalization already exists; execution is the contest
Payne calls the ambition a “living internet,” with digital experiences adapting continuously to visitors. But automated personalization is an existing product category. Adobe's Target product description says its Auto-target feature uses machine learning to select marketer-defined experiences for individual visitors based on their profiles and behavior.
Coframe's pitch concerns the breadth of the production cycle: generating new creative material and code, running experiments and deploying changes together. These product descriptions do not establish that this combination is unique. They do establish that showing different visitors different pages is insufficient on its own to distinguish the startup from Adobe's marketing software.
The startup is also drawing on the established experimentation industry. It says it acquired companies founded by two early Optimizely executives: Bo Mohazzabi, now Coframe's vice president of go-to-market, and Steven Ebin, now general manager of revenue optimization. Those appointments put experience from an incumbent experimentation business inside the challenger.
The degree of autonomy also needs qualification. Coframe's product page describes AI agents working alongside human experts to develop roadmaps and execute programs. Customers set priorities and approve what goes live. Coframe therefore competes for work done by conversion optimization agencies and internal teams as well as software vendors.
It can also complement existing infrastructure. Coframe says customers add a JavaScript snippet to their websites and can use it alongside their existing testing tools. Its page describes automated quality assurance across devices and automatic fallbacks. Those are company descriptions of deployment safeguards, rather than published measurements of reliability or site-performance overhead. The page supplies no prices with which to compare the combined service against software and staff costs.
Company-reported customer outcomes, not independently verified. The revenue figure concerns customers, not Coframe; its measurement period and the conversion average's weighting are undisclosed. Source: PR Newswire.
Large claimed gains, limited measurement detail
Coframe says in its announcement that it has generated more than $300 million in incremental revenue for customers, with average conversion-rate improvements exceeding 20%. These are company-reported customer outcomes, not Coframe's own revenue. The funding reports flag the figures as unverified independently.
These percentages concern different windows and, for Replit, an explicitly narrower conversion category. Neither page supplies baseline conversion rates, sample sizes or sufficient experimental methods to make them comparable. They cannot establish what a new buyer would achieve. The cumulative revenue claim also lacks a stated measurement period, and the average improvement lacks a disclosed customer count or weighting method.
The operational claims require separate scrutiny. Coframe says one Dropbox experiment that would previously have required four weeks of engineering, design and quality assurance was built and deployed in 30 minutes. That is a claim about producing and launching an experiment, not completing the measurement of its effect.
It reports a 39% experiment win rate across thousands of variants, calling that more than three times the industry average. The announcement does not define the win threshold or identify the benchmark's population and period. It also claims testing velocity at least ten times higher, while its product page separately advertises eightfold higher test velocity for Backcountry. Those claims do not establish a uniform speed improvement across customers.
Coframe says one unnamed Fortune 500 customer achieved a 70% conversion lift within 30 days of signing. That supplies a time window, but leaves the customer, baseline and experimental design undisclosed. Faster execution alone does not establish the economics after service fees and human work.
The next test is durable improvement
Coframe plans to extend into apps, email and other digital experiences. It also says it is collaborating with OpenAI on ChatGPT advertising to optimize ads and landing pages together, following an earlier partnership to train a specialized model for on-brand code and creative assets. That collaboration is distinct from the participation of AI-company leaders in the financing.
The company is preparing for AI agents that research and buy on people's behalf as well as human visitors. That remains a strategic direction, rather than evidence in the announcement of measured gains from agent traffic.
For buyers, the next useful evidence is a repeatable account of what improves: clearly defined conversion goals, controlled comparisons over stated periods, and gains after deployment costs. The funding expands Coframe's resources to pursue that proof. The customer figures published so far leave its durability and price unresolved.
The Office for Students wants more AI-free assessment and better AI skills teaching, backed by existing quality rules. Its research raises doubts about unsupervised coursework, but universities say assessment reform is already under way.
Finland has recognised Tesla’s provisional Dutch approval after changes to branding and speed controls. Drivers remain responsible, and the national decision does not commit Finland to backing EU-wide authorisation.
Grok Bot can connect to internal MCP servers through Tailscale without exposing them to the public internet. User-based authorization preserves network controls, but administrators must configure how narrowly the assistant can reach into their systems.
Daylight’s Goldman Sachs Alternatives-led Series B backs security agents supported by human engineers. The young company reports rapid growth, but has not disclosed the economics behind its automation promise.
Primordial Labs raised $18 million to widen deployment of Anura, its military command software. Existing contracts and demonstrations give it a foothold, but integration counts do not establish sustained use or lower operating costs.
Northwestern Polytechnical University bought two Nvidia H200 processors for 874,700 yuan and awarded five contracts for rented Nvidia computing, procurement records show. Separate US restrictions on the defense-linked institution remained in place despite January’s conditional opening for H200 exports to China.
The North Dakota Republican says allegations involving a suspected Chinese intelligence operative make Huang vulnerable to coercion. Nvidia’s board is backing its founder while independent directors investigate, with investors facing both security and leadership risks.
Scott Bessent says Washington and Beijing are establishing a channel for serious AI incidents, following their September agreement. Its usefulness depends on unresolved reporting procedures and participation by the companies that detect model failures.
Alibaba’s Qwen4-Max preview gives developers a fixed model to test through November 25. API access requires Token Plan, while claimed coding gains remain unquantified in the launch materials.
Railway estimates stolen customer keys generated about $740,000 in unauthorized AI spending after an October intrusion. Secrets from up to about 100,000 workspaces could have been decrypted, and the company says it will cover losses providers decline to refund.
Anthropic expects retail investors worldwide to receive up to 20% of its IPO, potentially $12.9 billion at the proposed upper price. Access through six brokerage channels depends on demand and broker rules, while a separate technology-company investment would help finance vast infrastructure commitments and founders and a special trust retain control.
Anthropic's seven founders plan equal IPO share sales worth about $250 million each at the proposed midpoint. Their sales make up 2.9% of the public offering, while the company seeks much larger financing under a structure that limits public investors' voting influence.
Nvidia’s independent directors retained Wachtell Lipton partner Joon H. Kim for an investigation into reports of a foreign intelligence effort targeting co-founder and CEO Jensen Huang. The board previously expressed confidence in Huang; the appointment provides no findings or timetable.
RRE Ventures led the funding for Graphon’s enterprise memory infrastructure. The startup is pitching connected, persistent context across different agents, but benchmark gains leave deployment costs and commercial scale unresolved.
Goldman Sachs, a lead bank on Anthropic’s IPO, reportedly forecasts $685 billion in annual revenue by 2030. The Claude developer has substantial cash reserves and rapidly growing sales, but rising compute costs, competitive pricing and delivery risks complicate the enterprise growth case.
Memories Upload pays contributors for accepted first-person video and sells annotated data to robotics developers. Its million-hour collection claim leaves open how much footage improves robot performance, while contributors surrender ownership without royalties.
OpenAI has acquired AI travel concierge Odessia and is bringing its founder, former Sonder chief Francis Davidson, into its travel effort. The standalone service has closed, leaving its future role alongside ChatGPT’s existing travel partners unclear.
Anthropic says expected advances from future models have triggered its research-acceleration threshold. An outside assessment finds faster progress but does not certify compliance with that threshold, while the voluntary policy leaves consequential decisions about safeguards and continued development with the company.
OpenAI’s free business-profile beta adds branded agents for eligible U.S. ecommerce and online service firms. The agents can also serve as paid-ad destinations, while OpenAI controls distribution and businesses bear responsibility for outputs.
