Whering's AI Fashion App and Its Google Dependency

At Google I/O last week, a startup founder sat across from the person whose fund had invested in her company, on a stage hosted by that investor's employer, to explain how she'd built her entire product on that employer's AI models. The conversation was warm, enthusiastic, and technically interesting. It was also a case study in the structural dynamics that regulators and competition watchdogs are increasingly watching in the AI ecosystem — dynamics the participants had no obvious incentive to name.

The startup is Whering, a fashion app that lets users photograph their wardrobes, receive AI-generated outfit suggestions keyed to mood, weather, and occasion, and eventually participate in community-driven styling. The founder is Bianca Rangecroft, who says she came to the idea through a finance career that included work on fashion tech transactions for large private clients — though the precise nature of those roles, as she described them briefly in the interview, is worth taking at face value rather than as a credentialed biography. The interviewer is Arthur Soroken, identified in the session as a co-founder of Google's AI Futures Fund — though the exact structure and leadership of that fund is not widely documented, and the "co-founder" designation is unusual for what appears to be an internal investment vehicle. Soroken noted that Whering is an AI Futures Fund portfolio company.

That context matters for everything that followed.

The Architecture Is Interesting. The Dependency Is the Story.

Rangecroft described three AI capability pillars for Whering: a conversational chat layer that advises on personal style evolution, resale timing, and outfit elevation; a wardrobe digitization pipeline that uses computer vision to auto-tag, crop, and categorize items from a user's camera roll or Instagram; and an experiential layer covering virtual try-on, color analysis, and contextual styling recommendations. For the wardrobe pipeline in particular, she described a tiered approach: lightweight inference — likely Gemini Nano, which can run on-device — for high-frequency, low-complexity tasks, and heavier cloud-based models for more computationally demanding work.

Rangecroft was notably circumspect about which models are actually in production. When pressed, she said "a bunch of them" and deferred to her team on specifics, though she confirmed using what Soroken described in the conversation as Gemini 2.5 and a photo processing model. That designation — "Gemini 2.5 Photo" — does not correspond cleanly to any publicly documented Google product name as of this writing, and may reflect either a product in limited preview or imprecision in the live conversation. The point is not to catch anyone in an error; it's that Whering's entire inference stack, by Rangecroft's account, runs on Google models, accessed through a relationship that includes early model access shaped partly by feedback from Whering's team.

That is a meaningful arrangement. Whering is not just a customer choosing Google Cloud from a menu of options. It is a portfolio company of a Google-affiliated fund that has apparently shaped product decisions through PM-level conversations. Rangecroft described her "aha moment" as coming from a conversation with a Google product manager she specifically thanked on camera — a moment that reoriented Whering's core onboarding model. Early model access, pricing terms, and product roadmap visibility are structural advantages that Whering's competitors on Google Cloud do not automatically share.

None of this is necessarily improper. But it raises the question that antitrust regulators in both the EU and the U.S. have been asking about the hyperscalers' AI investment practices: when a cloud provider also funds the startups building on its infrastructure, what leverage does either party actually have in the commercial relationship, and what happens to the startup's users if that relationship changes? Rangecroft described cost efficiency as central to her engineering philosophy. What she did not describe is what Whering's cost structure looks like if Google adjusts its model pricing — a question that matters more when your investor and your infrastructure vendor are the same entity.

Invisible by Design, Visible to Whom?

Rangecroft's description of Whering's layered AI visibility is worth sitting with. For one user cohort — sustainability-focused, fashion-first, less technically oriented — AI operates silently: computer vision crops and tags wardrobe items, color analysis runs in the background, the "enhance" feature improves item photos without fanfare. For a younger, more tech-literate cohort, the AI is surfaced directly: a chat interface sits in the main navigation bar, explicit AI recommendations are foregrounded.

"We've kept parts of the app very AI behind the scenes," Rangecroft said, framing this as a UX decision calibrated to demographic preference. Users who want the stylist without the machinery get the stylist without the machinery.

The EU AI Act, which entered phased enforcement in 2024 and 2025, includes transparency obligations for AI systems that interact with natural persons — obligations that are not satisfied simply because a user prefers not to know the mechanism. The FTC's guidance on algorithmic decision-making has similarly emphasized that consumer preference for opacity does not eliminate a company's disclosure obligations, particularly where AI is making recommendations that influence purchasing or financial decisions. Whering's chat layer explicitly advises users on resale timing and pricing. That is a financial recommendation, generated by a model, delivered to consumers — the kind of system that disclosure frameworks are designed to reach, regardless of whether the user has opted for "wardrobe zen" aesthetics.

Whether Whering's current implementation is compliant with the AI Act's Article 52 transparency requirements, or with analogous FTC guidance, is not something I can determine from a product demo. What I can say is that "invisible by design" is a UX choice with a regulatory surface area that the company does not appear to have addressed publicly — and that surface area will grow as the app's AI recommendations become more consequential.

The Democratization Problem

Rangecroft used the word "democratize" three times across the interview, in three distinct senses: democratizing access to personal styling, democratizing wardrobe data, and democratizing community-driven fashion input. The company claims a user base of 10 million globally — a figure sourced entirely from Rangecroft in a promotional interview and not independently verified; for a startup three years old, it warrants the caveat.

But the more interesting tension is structural, not numerical. Toward the end of the conversation, Rangecroft described Whering's plan for managing compute costs: gate the most processing-heavy AI features behind paywalls. The heaviest multimodal inference — virtual try-on, presumably real-time color analysis, the more sophisticated styling outputs — will cost money to access.

This is a rational business decision. It is also in direct friction with the democratization vocabulary. "Democratize" implies access regardless of means. A tiered model where the most capable features require payment describes something else — a freemium product with a capable paid tier, which is a legitimate and common product design, but which means the quality of your AI-assisted personal stylist will be correlated with your ability to pay for it. Users who cannot afford the subscription get the background-invisible, lighter-weight version. Users who pay get the features that actually change behavior.

That is not a mission problem necessarily, but it is a vocabulary problem — and vocabulary in product positioning is not trivial when you are also collecting wardrobe data at scale, building behavioral profiles, and operating in markets that are beginning to regulate data practices in fashion and retail specifically.

What to Watch

The more consequential question for Whering — and for the broader category of AI-dependent startups embedded in hyperscaler investment portfolios — is whether the EU AI Act's transparency rules, as they reach full enforcement, will require companies to disclose AI involvement to users who have affirmatively chosen not to see it. If the answer is yes, the "invisible by design" architecture requires a rethink. And if Google's AI Futures Fund portfolio companies receive preferential model pricing or access that is not disclosed to users or competitors, that is the kind of commercial arrangement that the European Commission's ongoing investigations into cloud market dynamics were specifically designed to examine.

The clothes on Rangecroft's app are already digitized. The regulatory frameworks governing what happens to the data, and to the users who generated it, are still being written.

---

Samira Barnes covers technology policy and regulation for Buzzrag.