Marissa Mayer, the former CEO of Yahoo, is betting that a phone’s photo history may reveal more about a user’s life than email or a calendar does. This is the foundation on which Dazzle is being built, a personal artificial intelligence assistant whose startup developer raised $8 million in a seed round last December.
Instead of gathering context from multiple text-based apps, Dazzle focuses on a single source: the photo library. According to Mayer, the system can analyze photos to extract indications of hobbies, interests, foods, and personal preferences, as well as how users spend their time and the people who appear repeatedly in their lives.
How does Dazzle work?
The assistant can be used through the app or via text messages, and its functions are divided into two main tracks. For direct tasks, it examines recent photos to extract actionable information; it might read an event flyer and add its details to the calendar, or notice a photo of a damaged garage door and suggest finding a repair technician.
The second track uses the entire photo history to build personalized suggestions, such as ideas for vacations, gifts, and family activities. During a test involving a travel request, Dazzle suggested destinations in the Mediterranean, apparently drawing on previous trips to Spain and Greece. It also suggested local activities, including a pottery workshop and a bioluminescent kayaking tour in Tomales Bay.
What distinguishes it, and what are its limitations?
The feature Dazzle is trying to offer is not merely recognizing the content of a photo or searching for a product visible in it, but connecting accumulated photos with lifestyles and interests. Mayer said the system inferred from her family’s photos that they like escape rooms, then suggested several locations in the San Francisco Bay Area that she did not know about.
But the experience also revealed clear limitations. The assistant suggested buying a pair of ski boots for the user’s daughter, even though she already knew how to ski. The system also recalled a trip to Sicily despite it having taken place four years earlier, indicating that its inferences from the archive are not always consistent or understandable to the user.
Why does this direction matter?
Dazzle represents a different direction in personal-assistant design: moving from executing commands based on email and calendars to attempting to build a model of the user’s identity and preferences. That could make suggestions more relevant, but it also increases the sensitivity of the data the system must analyze, since a photo library may contain pictures of children, homes, locations, and private events.
Mayer says Dazzle prioritizes privacy and discards personal information that the artificial intelligence identifies as sensitive. She also believes some users may prefer giving the assistant their photos rather than access to email and messages, particularly amid security concerns associated with competing assistants such as Muse and Instinct. However, this argument does not eliminate the basic question: Is a mechanism for deleting sensitive data sufficient, and how can users verify the accuracy of what the system inferred or that it has not retained additional information?
Mayer has previous experience with photo tools through her company Sunshine, which launched the AI-powered Shine tool in 2024 before later discontinuing it following criticism of its design and limited adoption. Mayer says the experience left behind useful intellectual property. Dazzle, meanwhile, presents a promising vision of a more personal assistant, but it remains less comprehensive than competing assistants, and its practical usefulness must overcome memory and inference errors before it can become a sustainable alternative to assistants that rely on broader data sources.