A grandmother opens Photoshop. Somewhere between the layers panel and the toolbar, the software has a small examination waiting for her. Does she know the vocabulary? Can she recognize the icon? Has she used something like this before? Daryl Budiman has chosen this encounter as a test of what software ought to be able to do. On his personal website, he asks: “Could you imagine if your grandma can use Photoshop?”
It is an unusually domestic question for someone whose career includes autonomous browser agents and a Y Combinator startup. There is no benchmark score in it. No race against another model. There is a person with a task, and a tool whose usefulness depends on whether that person can make sense of it. The grandmother is hypothetical. The design problem is quite real.
Budiman, originally from Indonesia and now based in New York, describes his interest as making software usable for everyone. His work gives that ambition several different settings. There is a student project about cooking and memory. A founding design job at MultiOn. Andoria, an AI onboarding agent developed with co-founder Anirudh Ramprasad. More recently, articles about advertising research. Each involves someone trying to get something done before the instructions consume the afternoon.
A story simmering on the stove
In Stanford’s autumn 2022 CS147 course, Budiman was the designer on StoreaTime. His collaborators were Lisette Malacon, Jaylene Martinez and Hector Angel Rivera. Their project proposed a way for people to share memories about food while they cooked. A dish could become a prompt: record a story, hear someone else’s experience, find people with similar memories or tastes.
The premise had a pleasing respect for dinner. Cooking would continue. Connection would happen alongside it. The project described audio and text as ways to share without making cooks choose between the activity in front of them and the stories attached to it. A kitchen already contains quite enough interruptions, most of them capable of burning.
StoreaTime’s project archive lays out the design work: needfinding, experience prototypes, a concept video, low and medium fidelity prototypes, heuristic evaluation, a higher fidelity version and a final report. The sequence matters. An attractive screen is only one part of making an idea usable. The work also involves deciding what someone needs, trying a version, examining where it fails and changing it.
There is no need to turn a class project into a prophecy. StoreaTime was a particular collaboration with a particular purpose. Still, it offers an early example of Budiman designing around a person who is already occupied. The cook has a meal to prepare. Any new interface must earn a place in that routine. That is a stricter admission policy than simply having an idea worth presenting.
The bicycle footnote
Budiman studied Symbolic Systems at Stanford. The university’s program directory lists him under his full name, Daryl Albert Budiman. Another Stanford record places him in the 2022 Xfund Ethics Fellows cohort. These are two distinct parts of his university record: a course of study and a fellowship affiliation. Neither needs an embellished origin story to be interesting.
His LinkedIn education entry supplies a less ceremonial detail. He writes that he “developed strong calves from all the biking.” Stanford also lists him on its 2021-22 club badminton roster. Between the academic directory and the bicycle joke, a student begins to emerge from the credentials. The campus was a place to study, get around and play a sport, as campuses tend to be.
The joke is also a welcome interruption to the standard founder biography. Such biographies can make university sound like a holding area for future chief executives. Here, at least, some of the time was spent pedaling. The badminton record establishes participation; it requires no invented account of tournament victories or lessons learned on court.
By 2023, his work was taking him elsewhere. MultiOn announced Budiman as its founding designer and said he was taking an indefinite leave from Stanford. The company described its mission as bringing action-based AI assistance to people. His job title put design inside that effort from an early stage.
“Could you imagine if your grandma can use Photoshop?”
Daryl Budiman
When the assistant can do something
A browser agent changes the designer’s assignment. A conventional interface waits for a person to choose an action. An agent can act through the interface itself. The design must make that relationship understandable: what is being requested, what is happening and what the person can do next. Those are practical questions even when the technology behind them is complicated.
MultiOn’s announcement placed Budiman in that setting. Later, he and Ramprasad described themselves as members of its early team, building autonomous AI agents that interacted with browsers. They took that experience into a company of their own. Zeroframe was founded in 2024 and joined Y Combinator’s Fall 2024 batch.
The product introduced in the company’s launch was Andoria. It was aimed at the moment a customer gets stuck inside a web application. The proposed agent would learn the application, consider the user’s background and goal, and generate a customized walkthrough. It could demonstrate actions in the interface. The integration pitch was a single script tag.
That pitch addressed two audiences at once. The customer needed help with a task. The software company needed a way to offer that help without creating a separate project of its own. Installation belongs to the usability problem too. A tool for reducing effort still has to persuade someone to make the initial effort of trying it.
- 01Learn the appUnderstand the interface
- 02Find the goalAdapt the walkthrough
- 03Show the actionHelp inside the task
A product proposal, rather than a measure of customer results.
The distance between an answer and an action
Consider an ordinary software question: how do I complete this task? A written answer can be accurate and still leave several jobs for the reader. Find the right screen. Match the words to the controls. Work out whether the instructions describe the version currently open. Then act. Accuracy is necessary, but the journey continues after the answer arrives.
An assistant that demonstrates a task attempts to shorten that journey. The attraction is easy to understand. So is the responsibility. A person needs to understand a demonstration well enough to follow it, and an action well enough to remain in control. Helping inside an application brings assistance closer to the work; it also brings mistakes closer to the work.
These are implications of the design, rather than claims about Andoria’s measured performance. A launch explains what a team intends to make possible. Judging whether it succeeds requires evidence from use. Keeping those questions separate gives the idea room to be examined on its actual terms.
For Budiman, Andoria belongs in a career that began publicly with design as well as AI. The interesting connection to StoreaTime is the placement of the interface. In one case, it needed to fit around cooking. In the other, help needed to fit inside a web application. Both ask where assistance belongs when a person’s attention is already committed.
A founder’s biography needs a date
Zeroframe is now marked inactive in Y Combinator’s directory, where Budiman and Ramprasad appear as former founders. Budiman’s own website describes building Zeroframe in the past tense. It gives his location as New York. The distinction matters because an old company title can linger long after the person has moved into a different chapter.
Recent work under his name concerns another demanding interface: the machinery of online advertising. Dovey published his guides to ad creative testing and AI-assisted creator research on September 24, 2026. The subjects are practical. How should a small team organize a test? What should it compare? How should it inspect a potential creator rather than rely on a familiar name?
In the testing guide, he recommends grouping creative versions around ideas and reading results against the advertiser’s own account history. The point is to make a comparison informative. A pile of variations can produce a winner without explaining what to do next. An organized experiment has a better chance of answering the question that prompted it.
The creator-research guide draws attention to the information behind an AI answer. It describes inspecting recent posts and engagement measures, and distinguishes drafting an outreach message from actually sending one. The emphasis is on a usable working process: find candidates, examine evidence, make a choice. A fluent answer is only one stage.
These articles add a current subject to Budiman’s public work without erasing the earlier ones. The cooking project, the campus bicycle joke, the founding design role and the onboarding agent each belong to their own moment. Together, they make his opening question more interesting. A person wants to use a powerful tool. How much of the tool’s complexity must become that person’s problem? Somewhere, a grandmother is still waiting to edit her photograph.