Kanjun Qiu

CEO & Co-Founder — Imbue

Profile

Kanjun Qiu is the co-founder and CEO of Imbue, an independent AI research lab betting that the path to useful agents runs through genuine reasoning — not bigger pattern matchers. She started the company with Josh Albrecht in 2021 as Generally Intelligent, then rebranded to Imbue in 2023 when it raised a $200M Series B at a $1B+ valuation, led by the Astera Institute with participation from Nvidia and others. The round came with access to a 10,000 H100 cluster — serious compute for a lab that isn’t chasing the frontier-model arms race.

Her path here is unusual. She studied computer science at MIT, worked at the MIT Media Lab, and paid her way through school writing high-frequency trading algorithms. She was the first chief of staff at Dropbox as it scaled from 200 to 1,200 people, co-founded the VR/laser-projector startup Ember, then founded Sourceress, an ML-powered recruiting startup. That mix — operator, researcher, founder — shows up in how Imbue is organized: deeply technical, but obsessed with the human side of what agents are supposed to do.

The substantive output so far is a 70B reasoning-and-coding model trained from scratch in early 2024 that roughly matched Llama 3 70B on reasoning benchmarks despite using 7x less data. More importantly for developers, Imbue open-sourced the infrastructure scripts, evaluation datasets (including a new CodeComprehension benchmark), and their hyperparameter optimizer CARBS. The 70B research series is one of the most detailed public write-ups of what it actually takes to train a large model — bare metal to benchmarks.

Beyond Imbue, Qiu is a GP at Outset Capital alongside Albrecht and Ali Rohde, backing early-stage AI founders. She writes publicly at kanjun.me about intelligence, institutions, and what it means to build tools that actually extend human agency. Her thesis — that agents today are stuck in the “bare metal phase,” and that reasoning, not scale, is the bottleneck — is one of the more coherent counter-narratives to the pure-scaling view.