Kanjun Qiu

CEO & Co-Founder — Imbue General Partner — Outset Capital Co-Founder & CEO — Sourceress Chief of Staff — Dropbox

Listen — profile

0:00 / 3:07

Profile

Kanjun Qiu is the co-founder and CEO of Imbue, a San Francisco AI research lab betting that the missing ingredient in today’s models isn’t scale — it’s reasoning. Where most of the frontier labs chase larger context windows and higher benchmark scores, Qiu has staked Imbue on agents that can genuinely reason about code and act on your behalf: systems that know when to ask a clarifying question, revise a wrong conclusion, break a problem into steps, and admit uncertainty instead of confidently hallucinating. For developers watching the “AI agent” hype cycle repeatedly over-promise, her framing is refreshingly honest — she titled one of her best-known interviews “Why AI Agents Don’t Work (Yet).”

Her path to running an AI lab is unusual. Qiu studied computer science at MIT and did research at the MIT Media Lab, funding her degree by writing high-frequency trading algorithms. She was the first Chief of Staff at Dropbox, joining as it scaled from roughly 200 to 1,200 people — a front-row seat to how organizations grow, coordinate, and lose institutional knowledge, themes that recur constantly in how she thinks about intelligence. She then co-founded Sourceress, a machine-learning recruiting startup, before launching what became Imbue (originally “Generally Intelligent,” founded in 2021 with CTO Josh Albrecht) and rebranding it in 2023 to reflect a mission of imbuing computers with intelligence and human values.

What makes Imbue worth a developer’s attention isn’t just the philosophy — it’s that the lab actually ships infrastructure. In 2024 Imbue trained a 70-billion-parameter reasoning model from scratch on 4,032 H100 GPUs and, crucially, open-sourced the tooling: CARBS, a cost-aware hyperparameter optimizer that lets small experiments predict large-model behavior, plus cleaned evaluation datasets, a new code-reasoning benchmark, and 450,000 human judgments about question quality. The lab announced a $200M Series B at a $1B+ valuation, led by the Astera Institute with Nvidia, Eric Schmidt, and others — remarkable for a team that was around 20 people. Beyond Imbue, Qiu is a general partner at Outset Capital, backing early-stage founders.

For someone learning AI today, Qiu is a useful counterweight to the “bigger model solves everything” orthodoxy. She argues for human-centered AI — tools that amplify human agency rather than replace it — and she backs that argument with released code rather than manifestos. Whether the reasoning-first bet pays off is still an open question, but the open-source contributions alone have made Imbue’s research genuinely useful to anyone training or evaluating large models.