Who Is Devi Parikh?
Devi Parikh is an AI researcher who has spent 20 years following curiosity across the field—from computer vision to generative models to the emerging frontier of web agents. She started as faculty at Virginia Tech and Georgia Tech, moved to Meta's Fundamental AI Research organization, and rose to Senior Director of Generative AI. Then, at the peak of her career, she did something that seems counterintuitive: she left to start Yutori, a company building AI scouts that monitor the web 24/7 and surface what matters.
That decision reveals who she is: not someone optimizing for status or title, but someone optimizing for the opportunity to drive the work itself forward. "I started to miss being in the weeds, being in the details," she explained. At Yutori, she's building orchestrated agents that solve a specific, hard problem—how to extract information from behind web forms and clicks, without endless user effort. But underneath the product is a deeper mission: to give people mental spaciousness back, to create technology that lets them spend less time on mundane tasks and more time on what matters to them.
The Archetype: The Creator
The Creator
The Sage
Tests & Allies
Devi is a builder and inventor first. The Creator doesn't execute someone else's vision; she architects systems from first principles. This is visible in every decision Yutori makes. When faced with the problem of building web agents that work across any website, not just a few optimized ones, most teams would try to parse HTML and DOM better. Devi asked a different question: "What's the actual input that a trained model can generalize across?" The answer—screenshots—seems obvious in hindsight. But it required seeing past the surface-level problem to the underlying architectural constraint.
Her secondary archetype is The Sage, the seeker of truth and illumination. She teaches through revelation, not assertion. When Angelina proposed a naive approach to building a scout (search APIs plus similarity matching), Devi didn't dismiss it. She walked through why it would break: non-crawlable databases behind forms, rare insights buried in Reddit threads, the need to balance coverage against precision. She made the invisible visible.
"One clear set of examples of what would work with just Google Search" becomes a structured lesson. "Different websites are built in such different ways that there is so much noise if you just sort of take that raw HTML DOM information" becomes a principle. That's the Sage teaching through showing, not telling.
The Hero Match
Athena
Athena is the goddess of practical wisdom and strategic architecture—not power wielded through force, but through seeing the structure underneath the problem. She invented the olive tree and the loom, tools that solved problems through clever design rather than effort. Devi matches this in three ways:
First, architectural thinking. You can trace Yutori's entire technical stack back to a core insight: the problem isn't "how do we process HTML better" but "how do we design for generality across the entire web." That reframing made the solution obvious. Screenshots. Hierarchical agents instead of a flat tool list. Post-trained models instead of off-the-shelf APIs. All of these flow from one architectural understanding.
Second, mentorship. Athena was a teacher. Devi teaches through her work—publishing detailed blog posts on why screenshots beat DOM, building transparent evaluation systems, providing citations and "Inspect Work" buttons that let users see the reasoning behind results. She makes complex architecture comprehensible.
Third, measured confidence. Athena was never showy. Devi won't claim Yutori will replace human judgment. She's transparent about what scouts can and cannot do. She's building trust through guardrails and explainability. She's willing to ship a smaller product first (read-only for now) because she's optimizing for depth of trust over breadth of capability.
Kate Beckett — Castle
Kate is a detective who sees patterns others miss. She walks into a crime scene, looks at the obvious evidence, and thinks: "What are we not seeing? What breaks the surface story?" Devi does this in technical problem-solving. She sees what others are missing in the information retrieval space: the heavy tail of data that lives behind forms, the need for multiple specialized agents, the cost implications of scaling an agentic loop. Both are pattern-matchers who don't trust surface-level framing.
And the tone matches: Kate is direct, measured, respects smart people, gets impatient with oversimplification, but is fiercely loyal to her team. That's Devi's conversational style exactly. She's not performing; she's thinking through problems with genuine intellectual rigor.
The Story Behind Yutori
The Founder's Journey ↔ The Company's Journey
Researcher exploring the boundaries of what's possible → Leader managing teams and coordination → Founder who wants to get back into the weeds and build something new.
General multimodal research → AI agents for web tasks → Scouts that monitor the web 24/7 for information you care about → (Future) Autonomous agents with guardrails for transactional tasks.
What makes this alignment powerful is the why underneath. Devi doesn't just want to build technology. She wants to build technology that solves a specific human problem: the feeling of being overwhelmed by information, the exhaustion of manual clicking and searching, the lack of time and space for what actually matters. "If it can play a role in giving us more time and space so that we can optimize our lives in ways that we care about, I think that would be pretty amazing."
That's not a marketing tagline. She picked the name Yutori—a Japanese word for the sense of well-being that comes from mental spaciousness—because that's what she's trying to deliver. The product is the external manifestation of a personal conviction about what humans need.
How Devi Leads
Devi leads through clarity and explanation, not authority. She owns her decisions ("I decided to leave Meta," "we decided to stay read-only"), but she frames them as collaborative discoveries. She credits specific people by name. She's comfortable with public uncertainty about tactics (what's the right future distribution mechanism for scout reports?) while maintaining conviction about principles (transparency, trust-before-capability, generality over site-specificity).
In a technical discussion, she doesn't assert; she reveals. When asked why screenshots work better than DOM parsing, she walks you through the architectural problem: different websites are built differently, raw HTML is too noisy, parsing becomes website-specific, and therefore not scalable. Then the principle emerges: "just using the screenshot is what gives you that generality." You understand not just the decision, but why it's the right one.
She's also unusually transparent about what Yutori can and cannot do. While the product is branded as an "AI Chief of Staff," she's very clear: "This can monitor information for you. It doesn't promise to do anything else." That clarity—and willingness to stay in the read-only space until trust is earned—is rare in an AI space that often oversells capability.
Founder Superpowers
Translating complexity into usable systems
Devi takes intricate problems—reliably extracting information from any website without site-specific parsing—and distills them into elegant solutions. But she does this without hiding the complexity from users. Scouts provides citations for every finding, an "Inspect Work" button that shows the reasoning, and feedback loops that let users nudge the product over time. Most founders either hide complexity or overwhelm users with it. Devi explains it clearly enough that people understand why her approach works. That builds trust. When Angelina said, "I don't think I'm going to build it myself. Seems like a lot of work," she meant: "I understand how hard this is, and I trust you did it right."
Seeing the architecture underneath the problem
When Angelina proposed search APIs plus cosine similarity, Devi didn't critique the approach. She revealed the architectural insight underneath: the problem isn't "what technique should we use" but "how do we deliver comprehensive information coverage while maintaining high precision, at scale, with cost control?" Yutori uses screenshots (generality), hierarchical agents (context management), post-trained models (cost), and feedback loops (personalization). Every decision flows from the same architectural thinking.
Building for the humans, not the technology
Most AI founders lead with capability—"here's what the model can do." Devi leads with purpose—"here's the feeling we're trying to enable, and here are the guardrails we need to build trust." Scouts is read-only because she's optimizing for trust over capability. She's willing to ship a smaller product that scales slowly because she's building toward "trust over time and go deeper" instead of chasing capability metrics. That's unusual in a VC-backed space. The result: users feel genuinely cared for, not exploited.
What It's Like to Work with Devi
Devi is measured and thoughtful, the kind of person who pauses before answering and chooses words carefully. She validates ideas even when she disagrees, credits her team by name, and teaches through explanation rather than assertion. She has strong conviction about principles (generality over optimization, transparency over hiding complexity) but genuine openness about tactics. She's comfortable with uncertainty when it's real.
Working with Devi means being in a culture that respects rigor. She runs detailed evaluations across the entire stack—from the model's individual decisions to the quality of final reports. She provides visibility into how the system works (Inspect Work buttons, citations, evals) because she believes in earning trust through transparency. She's willing to stay smaller if it means staying honest about what the product can do.
She's also someone who thinks about the human implications of what she's building. Not as an afterthought or a marketing angle, but as foundational. "We are not telling you that this is your chief of staff. We are very clearly saying, this can monitor information for you." That clarity—and willingness to add guardrails and ask for permission before expanding capability—shows that she's thinking about the humans affected by her technology, not just the technology itself.
Why This Matters (For You)
If You're a Knowledge Worker Drowning in Information
You probably have tabs open right now. Email newsletters you haven't read. Slack channels piling up. Product announcements you meant to track. News about your industry. Job postings that might interest your network. All of it competing for the finite attention you have. Devi built Scouts because she understands that problem viscerally. The product is designed around a specific belief: that you shouldn't have to do the mundane filtering work yourself, and that good filtering (Scouts) combined with natural feedback loops (replying to an email to nudge the scout) is how you actually build something usable. Not something that claims to do everything, but something that does one thing really well and scales that carefully.
If You're Building AI Systems or Products
Devi's approach reveals something important: the "secret" to building generalizable AI systems isn't always more compute or a bigger model. Sometimes it's better architecture. Yutori uses screenshots instead of HTML parsing not because it's simpler, but because it's more generalizable. They use hierarchical agents instead of a flat tool list not because it's easier, but because it solves the context explosion problem. They post-trained their own model not because it's faster, but because cost economics matter when you're serving this product to millions of users. This is the kind of systems thinking that separates products that scale from products that break. If you're building in the AI space, Devi's interview goes deep into the tradeoffs that matter.
If You're Early in Your Career
Devi's arc offers a specific lesson: the most meaningful work often requires you to give something up. She had the best research job in AI—leading teams at Meta's most prestigious organization. But she missed the creative work. So she left. That decision to optimize for "driving the work itself forward" over prestige or title is worth sitting with. It also says something about how long-term thinking works. She spent 20 years exploring the boundaries of AI. That's not optimization for short-term wins. That's commitment to understanding a domain deeply enough that when the time comes to build something, you know what to build and why. If you're early in your career, the question isn't "what's the most prestigious job?" It's "where can I develop the judgment and skills that will matter 10 years from now?"
If You're Considering Joining Yutori
Devi's leadership style reveals the culture: rigorous, transparent, principled. She leads through explanation, not authority. She provides visibility into how decisions are made (showing the reasoning, not just the results). She's comfortable being wrong about tactics but firm about principles. She credits her team. She thinks about the humans affected by her technology, not just the technology itself. If you join Yutori, you're joining a company that's willing to ship smaller to stay honest, that runs detailed evals and provides transparency as a feature, and that's building toward trust over capability. The product is hiring (visit yutori.com/careers).
Go Deeper
The full conversation with Devi Parikh is on its way. Check out other episodes in the meantime.
Join Yutori
Now that you know how Devi Parikh leads, see if there's a role for you.