TechCrunch reported on October 8 that Zach Yadegari, the teenage co-founder of the calorie-tracking app Cal AI, has raised $10 million for a new AI agent startup called Persona. The round was led by Vine Ventures, and among the participants were Collective Global and Z Fellows founder Cory Levy. A second co-founder, whose identity has not yet been announced, is working with Yadegari to build the company.
The funding follows a notable exit. MyFitnessPal acquired Cal AI in March for an undisclosed sum, after the younger app overtook the older calorie tracker in app store rankings. MyFitnessPal told TechCrunch at the time that Cal AI had generated more than $30 million in annual revenue in under two years, and that the deal was large enough to persuade Yadegari and his co-founders to sell.
Yadegari and his co-founders initially stayed on with the acquirer to keep developing Cal AI, but he left the larger company in June and moved straight to his next idea, he told TechCrunch. Persona is the result: a personal AI assistant in the same broad category as Instinct or Meta's Muse, but with a hardware component — a wearable band priced at $179 that the company says will be available soon.
That places Persona closer to Bee, the wearable AI device Amazon acquired in 2025, though Yadegari points to several differences. Persona is not designed as ambient AI: it will not listen or record continuously. Instead, the band carries a button that activates the assistant, and it can also be configured to trigger with a flick of the wrist.
On privacy, Yadegari states that the device neither stores user data nor runs the AI model. The model operates within Persona's cloud, to which user data is transmitted. According to the company, data is encrypted both in transit and at rest, and users have the ability to delete their data. Persona therefore gives its assurance that conversations between users remain confidential and will never be sold to advertisers or data brokers.
Safeguards against phishing and prompt-injection attempts are also being built by the team — two problem areas for agents that AI labs such as OpenAI have flagged. Persona's agent cannot see a user's credit-card details, Yadegari further promises, and the user's approval will be required for purchases, with payment handled by external providers such as Stripe Link, Stripe's secure wallet intended for agent use.
An advertising-based business model sits alongside those privacy claims. Advertised wares will be shown by the company during shopping research, Yadegari says, but he argues the agent will give unbiased recommendations because which items are sponsored will not be known to it. In his example, chairs matching their stated preferences would be seen by someone researching office chairs, some of which are sponsored results.
The bands are expected to be available in December, with an app in development. In the meantime, Persona has launched as a free beta in which users interact with the assistant over iMessage. Yadegari said the beta has attracted "a few thousand" users, and that the band has already generated "five figures" in preorder revenue.
Functionally, Yadegari describes Persona as capable of the tasks common to consumer AI assistants: ordering food or rides, taking notes, booking travel, and answering questions and emails, among others. He frames the ambition broadly, arguing that the everyday utility apps people use — shopping, Uber, DoorDash, email — are collapsing into a single AI-powered super app.
Longer term, he says he wants Persona to become more proactive, offering help and solving problems before a user recognizes them, and to be seamless enough that people do not need to stare at a phone all day. Those are stated intentions rather than shipped features, and the evidence does not include a timeline for them.
For freelancers, designers and developers, the most concrete signal here is not the assistant itself but the shape of the bet: a small, well-funded team pairing a cloud agent with a low-cost wearable and a messaging-first beta. That is a distribution pattern worth watching, because it lets a startup test an agent product through an existing channel — iMessage — before hardware ships, and it keeps the client surface thin while the model runs remotely.
The privacy architecture also carries a practical tradeoff that this audience should weigh carefully. Running the model in the cloud and sending user data there is what makes a $179 band plausible, but it also means the privacy guarantees rest on the company's encryption, deletion and retention practices rather than on data never leaving the device. The evidence describes those practices only as company statements; it contains no independent audit, technical documentation or third-party assessment.
The advertising model raises a second, unresolved tension. Persona says conversations will not be sold to advertisers or data brokers, yet the business model depends on ads shown during shopping research. Yadegari's argument is that the agent does not know which results are sponsored, so its recommendations stay unbiased. That is a design claim about how the agent is built, not a verified outcome, and the evidence offers no detail on how sponsored placement is selected or disclosed to users.
Payment handling is similarly described at a high level. The claim that the agent cannot see card details, combined with approval requirements and external processing through Stripe Link, suggests a deliberate separation between the agent and payment credentials. But the evidence does not specify which purchases are supported, what approval looks like in practice, or how disputes would be handled.
What remains unknown is substantial. There is no confirmed ship date beyond an expected December window, no pricing for the app or subscription, no detail on the second co-founder, and no information on model providers, latency, battery life or supported platforms. The beta user count and preorder revenue figures come from the founder and are not independently verified.
The clearest supported conclusion is that Persona is an early-stage, funded experiment with a defined hardware price, a stated December availability window and a live free beta, built by a founder with a recent, documented consumer-app exit. Whether the privacy and advertising claims hold up will depend on details the current evidence does not contain — which is exactly the kind of gap a developer or designer evaluating the platform should treat as open rather than settled.