Adobe Research Scientist Doga Dogan didn’t take the stage at this year’s Adobe Summit alone—he was joined by a team of what he likes to call “AI agents with an attitude.”
Dogan was presenting Project Face Off, a preview of an experimental project inside Adobe Research. Project Face Off uses AI-generated personas to simulate expensive, time-consuming A/B tests for website designs and marketing campaigns in just seconds. Users put forward a few design ideas, and then a group of AI agents predicts which one is most likely to meet their goals. Personas represent people from different demographics with a range of experiences, personalities, and opinions—and they share their likes and dislikes, along with the reasons behind their preferences.
The goal of Project Face Off is to let marketers explore ideas and lower the risk of launching a new campaign or beginning A/B testing with human users. By enabling rapid early-stage design experimentation, Project Face Off could give marketers valuable insights into which designs are likely to perform best in the real world—and why—before they move on to traditional live experimentation. “One of the great use cases of AI is being able to test scenarios that are very difficult to test in real life,” Dogan explains.
Though the product is still in its experimental stage, users can test drive an online demo.
During the Summit Sneak presentations, audience members voted for their favorite Sneaks and watched as the votes rolled in on a live dashboard. By the end, Project Face Off had taken the top spot as best Summit Sneak of the year.
The story behind the Sneak
Dogan and his colleagues on the Adobe Research and Adobe Experience Manager (AEM) teams have been thinking for a while about how AI agents can help marketers and designers validate ideas before trying them in the real world. Dogan worked on the research behind a 2025 Sneak, Get Savvy, that used AI agents to test poster designs. He was excited when two of last summer’s interns, Tim Reider and Marian Schneider, wanted to dig deeper. The two master’s students decided to explore AI agents with in-depth personas, including histories, likes, and dislikes.
During their internships, Reider and Schneider developed the working prototype that eventually became Project Face Off, designed a method for testing it. With Dogan and others, they wrote a technical paper (SimAB) about persona-based simulations as a fast and reliable proxy for early-stage A/B testing, especially when predictions converge quickly. They realized their timing was perfect: it was about to be Summit Sneak season. So Reider (who’ll be returning to Adobe full-time in the fall) and Schneider submitted their project as a potential Sneak.
Dogan saw the Sneak proposal through to acceptance for the Sneak stage. “It was a great feeling when Project Face Off was selected. It’s proof that interns are very valuable. They bring so much energy and synergy to Adobe,” says Dogan.
When Project Face Off won best Summit Sneak, of 2026 Dogan was thrilled. “I could feel that the audience was really rooting for us,” he remembers.
After the win, Dogan says it was like being a celebrity for a couple of days, with attendees and colleagues constantly stopping him to talk about Project Face Off. “A highlight was really connecting with Adobe customers after the Sneak,” Dogan says. “I spent a lot of time in the Community Pavilion, and everyone was so engaged and interested in what we do.”
The future of Project Face Off
Project Face Off is still in the experimental stage, but if it makes its way to Adobe products, the goal will be to help marketers and designers do “pre-flight” checks of their content, or even to give quick insights on how to polish a design.
Dogan has already heard from friends and colleagues who see use cases beyond websites, including testing illustrations for technical papers, social media posts, and lots of other creative projects. “In all of these cases, it could be helpful to use AI to test ideas. This is AI in the co-pilot sense, letting you move forward with more insights and more confidence,” says Dogan.
One of the most exciting elements of Project Face Off is that it can bring in a whole crowd of tailored AI personas, rather than just offering a single AI perspective. Users can create personas, bring in targeted audiences they’ve already developed, or get suggestions for which personas to include.
Once a test is running, the tool predicts a winner, similar to the way a news organization predicts election results as votes are coming in. It detects how quickly the personas are converging on a favorite option based on the statistical significance of the results. On top of that, users get detailed, individual feedback from each persona to guide future revisions.
“It’s important to put yourself in the shoes of your audience, and AI can help us do that a little better, and more at scale,” says Dogan. “For me, it’s difficult to think about what 30 different personas might think, but AI can help me aggregate that, scale it, and put it all together faster. That’s the superpower that AI gives us.”
Learn more about what Adobe Research was up to at Summit 2026 here.
Wondering what else is happening in Adobe Research? Check out our latest news here.
