Gabriel Hubert

Gabriel Hubert is co-founder and CEO of Dust, the platform for custom internal AI agents wired to a company's own data and tools. With co-founder Stanislas Polu — a former OpenAI researcher — he left Stripe/OpenAI orbits in early 2023 betting that one model would not rule them all, and that proprietary company data would unlock assistants people use at work.

No GPUs before PMF

They refused to raise a frontier-model fortune before knowing which jobs customers needed done. On Sequoia's Training Data podcast Hubert recalled the internal slogan "No GPUs before PMF": explore use cases on existing models first. The thick product layer — access control, observability, connectors, assistant design — mattered more than training another foundation model. Fine-tuning company data, they argued, was usually the wrong hammer; context control was the right one.

The stochastic mindset

Hubert's cultural thesis is that workplaces must leave calculator certainty behind. On Training Data he said: "We've asked the entire world to move from calculator technology—punch the same keys, you'll get the same results—to stochastic technology—ask the same question, you'll get a slightly different result." Builders who thrive treat drafts as time savers, not oracles: "I'm not asking it to be right 100 percent of the time. I'm asking it to give me a draft that saves me time many, many, many times over."

Many agents, not one oracle

Dust makes it easy to specialize assistants — sales enablement, support drafts, personal coaches, cross-functional translators between engineering and marketing. People, Hubert argues, already know a screwdriver from a hammer; they hate when a single bot answers like the wrong tool. Fragmentation is the point: once a company hits critical mass of assistants, usage skyrockets beyond the original pilot use cases.

Augmenting humans, not replacing them

A product motto he returns to is "Augmenting humans, not replacing them" — not naivete about jobs, but a bet on exoskeletons for people who will still be employed. The through-line is human-driven collaboration with non-human agents: proposals to swipe left or right on, co-editing interfaces, and keeping a finger on customer pulse instead of fire-and-forget automation that hides what support tickets were teaching the company.