Founder Story: Kanjun Qiu of Imbue
From High-Frequency Trading to the MIT Media Lab
Kanjun Qiu's path into artificial intelligence did not begin with a grand thesis about machine cognition. It began with a practical problem: how to pay for an MIT education. Qiu studied computer science at the Massachusetts Institute of Technology, where she also worked as a graduate researcher at the famed MIT Media Lab. To fund her studies, she wrote high-frequency trading algorithms to trade the stock market, an early and telling sign of someone comfortable turning code into leverage against the real world.
That combination of academic curiosity and hands-on building would recur throughout her career. While at the Media Lab, Qiu co-authored _Sew Electric_, a book that teaches computer science to middle and high school students through sewing and electronics, stitching circuits into fabric to make programming tangible. It was an early expression of a belief that has never left her: that technology should be something ordinary people can pick up, shape, and control, not a black box handed down by experts. Long before she was building AI agents, she was trying to reveal the builder in everyone.
The Dropbox Years and a First Company
Before she was a founder, Qiu learned what it takes to scale a company from the inside. She joined Dropbox as its first chief of staff, helping the company grow from roughly 200 to 1,200 employees. It was a front-row seat to the mechanics of hypergrowth, the organizational strain of scaling, and the difference between a startup's early scrappiness and the machinery required to run a large company.
Her first real venture as a founder came in a less glamorous corner of tech. In late 2014, Qiu co-founded Ember Hardware, a short-lived project she worked on into 2015. Soon after, she teamed up with Josh Albrecht, the technical partner who would become her collaborator for the next decade, to build something more ambitious. That company was Sourceress, and it would teach her the lessons that made Imbue possible.
Sourceress: A Machine-Learning Bet That Didn't Land
Founded around 2015, Sourceress was a machine-learning recruiting startup based in San Francisco, with Qiu as co-founder and CEO and Josh Albrecht as her co-founder. The idea was to use machine learning to surface job candidates that hiring managers wanted but couldn't easily find themselves. The company went through Y Combinator's Summer 2017 batch and raised roughly $13 million from backers including Y Combinator and Dropbox's co-founders. In 2019, Qiu was named a Forbes 30 Under 30 honoree in Enterprise Technology for the 2020 list on the strength of that work.
But Sourceress ran into a wall that would later shape Qiu's entire technical worldview. The machine-learning systems of the mid-2010s were brittle. They could pattern-match, but they could not reliably reason, and recruiting is a domain full of nuance, context, and judgment. Qiu and Albrecht wound down Sourceress in late 2020. Rather than treat the shutdown as a failure to move past, they treated it as data.
The frustration was not that AI was too weak. It was that AI was unreliable in a specific, structural way: it could imitate patterns without understanding them.
That distinction, between imitation and understanding, became the seed of their next and most consequential company.
The Pivot to AI Research: Generally Intelligent Becomes Imbue
In 2021, Qiu and Albrecht launched an AI research lab originally called Generally Intelligent, backed by an initial round of about $20 million. The mission was audacious: to build AI systems that could genuinely reason, plan, and act, rather than merely predict the next likely token. The team ran deep reinforcement-learning experiments, including a 3D research environment they built called Avalon, to probe how agents learn.
The experiments were humbling in an instructive way. As Qiu later explained, RL agents in Avalon could learn mechanical tasks like climbing and opening complex doors, but they "couldn't do any higher level things" consistently. Her conclusion was blunt: "pure reinforcement learning is not a good vehicle for planning and reasoning." Even the most capable models remained fundamentally unreliable. As she put it, "models are still stochastic. The way you prompt them still makes them do random things."
In September 2023, the company rebranded as Imbue and announced a $200 million Series B at a valuation of over $1 billion. The round was led by the Astera Institute, the nonprofit founded by cryptocurrency billionaire Jed McCaleb, with participation from Nvidia, former Google CEO Eric Schmidt, Notion CEO Ivan Zhao, Cruise CEO Kyle Vogt, and others. To date Imbue has raised on the order of $220 million, and the company secured access to roughly 10,000 Nvidia H100 GPUs to train its systems, all while keeping a deliberately small team of around 20 people.
_(A note on figures often cited in passing: as of mid-2026, Imbue's publicly reported valuation remains the $1 billion-plus mark set in the 2023 Series B. Claims of a roughly $12 billion valuation are not supported by verifiable public reporting.)_
The Reasoning-First Thesis
What makes Imbue distinctive is not simply that it builds agents; nearly every AI lab now does. It is why and how Qiu insists on building them. Her core belief is that reliability is the gating factor for useful AI agents, and that reliability comes from reasoning, not scale alone. Agents, in her framing, are systems that "go off on their own and do stuff," in contrast to a chatbot that simply answers a query. And the trouble, she has said plainly, is that today "the only successful agents are the ones that do really small things" — fixing a button color, not shipping a feature.
To close that gap, Imbue treats reasoning as something to be trained explicitly rather than hoped for. Qiu describes generating "very specific things that are like reasoning traces" to optimize the model's ability to reason, drawing on the heuristics humans use, such as the scientific method of checking claims against evidence. Code plays a central role in this thesis, for two reasons she articulates crisply:
- Code is "the most explicit example of reasoning data on the internet," making it ideal training material for teaching a model to think in structured steps.
- "By generating code, it can act on my computer" — code is both how an agent learns to reason and how it takes real action in the world.
Crucially, Imbue rejects the black-box approach. Qiu wants users to be able to "fork it, try out different things, modify the prompt, modify the plan that it is making." The agent is a collaborator to be inspected and steered, not an oracle to be trusted blindly.
Philosophy: Agents in Service of Human Agency
Where many AI founders talk about automation and replacement, Qiu talks about agency. It is the through-line of everything she builds. She has described herself as obsessed with human agency and human potential, and she frames Imbue's entire mission around it. The goal, she writes, is to "empower humans in the age of AI by creating powerful computing tools controlled by individuals."
This is not soft branding; it is a specific worry about power. Qiu argues that "AI makes software systems dramatically more powerful," and that "that power flows, by default, to those who can build and own those systems." Left unchecked, she warns, most people "quietly become less free," because AI agent builders end up controlling the algorithms that make decisions about users' lives. Her answer is to hand the controls back:
Technology's highest purpose is not to replace human capability, but to amplify what is already inside us.
She wants a world where "humans can be actors, not acted upon," where AI software is "tailored to our own goals and values" rather than the interests of whoever built it. It is a vision that connects directly back to the sewn circuits of _Sew Electric_ and the coliving experiments of her San Francisco years: a durable conviction that the point of powerful tools is to "reveal the builder in every human."
Industry Impact and a Rare Seat at the Table
Qiu's influence extends well beyond Imbue's own research. She is a general partner at Outset Capital, the early-stage fund she runs with Josh Albrecht and Ali Rohde, backing founders at the earliest stages, and she has served as a Sequoia Capital Scout. Through Imbue she also co-hosts a podcast — long known as the Generally Intelligent podcast and continuing under the Imbue banner — where she and her collaborators interview leading AI researchers in unusually technical depth, an exercise that doubles as public education about how these systems actually work.
Her position matters for another reason. In an industry where the largest AI labs and best-funded agent startups are overwhelmingly led by men, Qiu is one of the very few women running an AI company of Imbue's scale and ambition. She has been a longtime advocate for diversity in technology hiring, a thread that runs from her Sourceress days into how she thinks about building teams. She treats team members as "creative agents," aiming to "unlock their superpowers" rather than manage them as assets — a management philosophy that mirrors, almost exactly, the way she wants AI to treat its users.
Legacy and the Road Ahead
Imbue is playing a deliberately long game, and Qiu is comfortable with that. She and Albrecht have said they would rather wait until their technology is "robust, safe and great" before releasing it widely, even if that patience costs them attention in a hype-saturated market. In a field addicted to launch-day fireworks, choosing rigor over spectacle is itself a bet: that the winners of the agent era will be defined by reliability, not demos.
If she is right, the implications reach far beyond one company's product roadmap. Qiu's wager is that reasoning-first, human-steerable AI agents will "thin the barrier between ideas and execution," letting anyone turn intention into working software the way earlier tools let people turn thoughts into documents and spreadsheets. Whether or not Imbue is the company that delivers it, the framing itself — agents that reason, owned and controlled by the humans they serve — is one of the more coherent counter-narratives to AI-as-replacement, and it is spreading. Kanjun Qiu built her worldview out of a decade of hard lessons, and she is now spending hundreds of millions of dollars and years of patience to prove it.
Closing Thoughts
Kanjun Qiu's story resists the tidy startup arc. There was no single eureka moment, no overnight unicorn. There was a trading algorithm that paid tuition, a book that taught kids to sew circuits, a chief-of-staff seat inside a scaling giant, and a machine-learning company that failed for reasons she refused to forget. Each stage sharpened the same question: how do you give people more control over the tools that increasingly shape their lives?
Imbue is her most complete answer so far. It fuses a hard technical thesis — that AI must reason and act reliably, not just pattern-match — with a humanistic one about who that reliability should serve. In an era when the loudest AI voices promise to automate humans out of the loop, Qiu is building, patiently and against the grain, toward the opposite: agents that make people more capable, more free, and more in control. It is a founder story defined less by what she has already shipped than by the unusual clarity of what she is trying to protect.
Your Story Could Be Next
Every founder in this series started exactly where you are now — with an idea and the will to build. What has changed is the leverage. Frederick gives you a team of AI agents that take you from idea to a real company, helping you design, build, and ship software on demand — no giant team and no endless runway required.
If Kanjun Qiu's journey resonated, there is no better time to begin your own. Start building your company with Frederick.
References
- https://en.wikipedia.org/wiki/Kanjun_Qiu
- https://www.forbes.com/sites/alexkonrad/2023/09/07/ai-research-lab-imbue-nabs-200-million-for-speculative-bet-to-build-ai-agents/
- https://www.latent.space/p/imbue
- https://ideas.imbue.com/p/empowering-humans-in-the-age-of-ai
- https://www.ycombinator.com/companies/sourceress
- https://www.bloomberg.com/news/articles/2023-09-07/ai-startup-imbue-tops-1-billion-valuation-after-funding-from-nvidia
- https://www.outsetcapital.com/blog/tnai-kanjun-qiu-josh-albrecht