Inflection AI’s Second Act With CEO Sean White

In 2022, early in the first wave of generative A.I., DeepMind co-founder Mustafa Suleyman and LinkedIn co-founder turned venture capitalist Reid Hoffman launched Inflection AI, billing it as an attempt to build an emotionally intelligent A.I. The company’s core product was a chatbot called Pi, “purely designed for relaxed, supportive, informative conversation,” Suleyman said at the time. The startup quickly attracted backing from Microsoft, Nvidia, Bill Gates and Eric Schmidt and reached a $4 billion valuation when it was barely a year old.

Less than a year later, in March 2024, however, its core team dissolved. Microsoft hired Suleyman and much of Inflection’s founding team in one of the A.I. industry’s first major acqui-hires. Microsoft paid the startup about $650 million, largely to compensate its investors, according to Reuters. The only person who didn’t leave was Hoffman, who remains on Inflection’s board. He tapped Sean White, a computer scientist who formerly led R&D at Mozilla, as the startup’s new CEO.

Taking over a once-high-flying A.I. startup after most of its founding team has left can sound a little like rebuilding a house from the dust of its demolition. Inflection hasn’t raised new funding and now operates with a staff of roughly 60. After the upheaval, the company shifted much of its focus to enterprise customers. But its original consumer product, Pi, has also continued to evolve quietly.

Last month, Inflection signaled a renewed push into consumer A.I. with the launch of Inflection AI Labs and its first experiment, “Pi Journeys,” a chatbot designed to help users navigate different stages of life, from starting a new career to caring for aging parents.

White spoke with Observer about rebuilding Inflection after its dramatic reset, his vision for what he calls “relational A.I.” and how A.I. could one day be truly embedded in our personal and emotional lives rather than displacing human connection.

The following conversation has been edited for length and clarity.

Observer: You took over Inflection at an unusually turbulent moment, just after Mustafa Suleyman and much of the founding team left for Microsoft. How did Reid Hoffman approach you about the role, and what made you think there was still something worth building?

Sean White: Mustafa left cleanly. Most of the conversation was with Reid. I have an incredible amount of respect for Reid, his thoughtfulness and his view of the long arc and horizon in history.

It was pretty clear that we could take something that was nascent—where everybody was focusing on IQ—and lean into what it meant for systems to have emotional intelligence. To be clear, I’m not talking about a virtual boyfriend or girlfriend. That’s less interesting to me than systems that are actually great interfaces because they understand you and can help you achieve what you want. It’s less about replacing people and more about giving people agency.

Hoffman stayed on the board after the transition. What does your working relationship with him look like now? What do you guys talk about when you meet?

It ranges. We actually met just last night. Sometimes the discussions are about what’s happening globally with A.I. and the things I’m seeing. As you know, all of this moves so fast. We want to have as many radars up as we can.

Sometimes we’re down in the weeds. I was showing him a prototype of something playful that we had been building on our own systems. We also talk about what’s happening with the company and the team, some of the long-haul items and what it means when you combine relational A.I. with existing forms of A.I.

Pi originally stood out because Inflection was emphasizing personal intelligence and emotional intelligence at a time when most of the industry was focused on raw model capability. How much of that original thesis still defines the company today?

If you’re really thinking about designing toward human-centered A.I., you step back and ask: What does it mean to be human? What’s important? None of us exists in isolation, and we shouldn’t exist in isolation.

If you look at the architectures and systems from two years ago, they weren’t built with the same harnesses or pipelines for fine-tuning and actually improving EQ in the ways we now know we can. They also didn’t have any of the real agentic capabilities that I believe are the next step in helping people.

We’ve been building out those systems. Almost every one of these systems has been built from the ground up for A.I.-to-person interaction. What we’ve been experimenting with is taking into account the fact that you have other people around you. It uses relational A.I. to understand what those relationships are and to be prosocial.

Most A.I. tools on the market today focus on automating tasks and helping us work faster. You seem interested in something different. What do you think the industry is missing or getting wrong?

So much of what we see is focused on pure productivity. When I have conversations, particularly with my friends in the humanities, they remind me that life isn’t purely about productivity.

You’ll see some experiments from us that explore that idea. That doesn’t mean we won’t also build things that help people in their everyday lives. With Pi Journeys, you’ll see things like life navigation, parenting and caregiving support, personal growth and help with simple daily tasks, from finding recipes to organizing your thoughts.

These ideas don’t come out of nowhere. We have a team doing really interesting user research, including a study of people who currently use A.I. One thing that surprised me was that, on average, people use more than three A.I. systems over the course of a week. It’s not just one system.

For us, that creates an opportunity to focus on what people are actually doing in their daily lives. If you’re organizing your life and need to stay in touch with siblings, coordinate with other people or care for an older adult or a child, these are ways A.I. can help you be more prosocial while also managing to-dos, organization and other practical tasks.

Inflection AI Labs launched as a place to test some of those ideas publicly, and Pi Journeys is your first experiment. What’s the long-term goal of these experiments?

Of course, we still run lots of experiments internally. But I think one of the best ways to build products is to learn quickly with people outside of our bubble. Pi Journeys is one of our first experiments. It helps you understand the people who are important in your life. That can be for different use cases, whether you’re a caregiver, organizing a household or just living your life.

The arc is: You start with IQ, then you have EQ. If it’s just IQ, it’s just search. Then you go from IQ and EQ to AQ, or the agentic ability to do things, and finally to real social intelligence.

You’re building A.I. around the routines of everyday life. How do you use A.I. in work and life? 

I use multiple A.I. systems for everything from organizing my email to discussing and working things out on my drive in to aggregating new information.

As I’m coming in, I use Pi because I like the dialogue and voice aspects of it. If I’m trying to get something straight in my head, learn something new or make sure I remember a couple of things on my drive into work, I use Pi.

If I’m doing something more business-oriented, like slides or spreadsheets, I use Claude for a lot of that, as well as some experimental systems.

I also use some of our own systems to aggregate different headlines, news stories and other information, work across them and see what they show.

On my way back home, I tend to use Pi again. I also use it in some cases for searching. For instance, recently—and this is super mundane—I needed a new belt and was asking Pi for help with it.

The A.I. landscape is moving so fast. When you think about where the industry is now, what feels genuinely important, and where you think it’s headed?

I tend to have a “long now” visualization of this, which is that we’re just at the beginning. I show my team a graph I built that shows search engine market share starting in 1994, which is roughly when I entered the workforce, through today. What’s amazing is how much it changes.

In 1994, my office mate was Srinija Srinivasan, who later became Yahoo’s first “chief ontologist.” It was just a couple of students sitting next to each other in these terrible offices. Everybody thought Yahoo was going to win. Then, somewhere around 1997 or 1998, everybody thought Lycos and Excite were going to win. A few years later, Google started powering through.

What it reminds me is that we are in such early days, such early stages of A.I. I’d really like to make sure that the things we’re building are actually helping people and that we’re focused on the uses, the systems and how those evolve. That’s where I feel like I, our team and our organization can have an impact.