Dust

Gabriel Hubert and Stanislas Polu met at Stanford in 2007. They co-founded TOTEMS, a data analytics company, which Stripe acquired in 2014 — their first exit together. Both spent five years at Stripe building products and scaling teams. Polu then joined OpenAI as a research engineer on Greg Brockman's team, co-authoring papers on AI reasoning with Ilya Sutskever. Hubert became Chief Product Officer at Alan, the French health technology company. Two tech careers, one common thread: they had always built things together.

In September 2022, Polu left OpenAI with what he later described as a founding thesis: the models were already powerful enough to be economically transformative, but the product layer required to deploy them usefully inside organisations did not exist. Most enterprise AI was "single-player": one person prompted an assistant, got an answer, and the context disappeared into a private chat window. There was no compounding. No shared memory. No organisational intelligence. Dust incorporated in February 2023 to build the missing product layer — a model-agnostic platform that would not lock customers into any single AI provider.

Dust's differentiation was conceptual before it was technical. Rather than each employee having a separate AI assistant with separate context, Dust created shared workspaces where human employees and AI agents operated with the same projects, conversations, documents, and goals. Agents built on Dust could be created by non-technical employees — operations managers, sales leads, support teams — without writing code. The platform connected to over 100 data sources including Slack, Notion, Google Drive, and Salesforce. Sequoia Capital led the seed round of €5 million in 2023, followed by a $16 million Series A in June 2024.

By 2025, Dust served more than 3,000 organisations with 300,000 agents deployed across the platform. Weekly active usage across customers exceeded 70%. Net Revenue Retention was 240% with zero customer churn across the entire year. Customers included Vanta, whose 46-person revenue team saved more than 400 hours per week; Qonto, which estimated 50,000 hours saved annually across 50 specialised agents; and Doctolib, which had made Dust central to its company-wide AI strategy for 3,000 employees.

In May 2026, Dust raised $40 million in a Series B led by Abstract and Sequoia Capital, with Snowflake Ventures and Datadog participating. Total funding exceeded $60 million. The company had approximately 98 employees in Paris and San Francisco. The founding thesis — that the models were ready but the product layer was missing — had been proven correct. The broader market for agentic AI platforms had surpassed $9 billion in 2026 and was projected to reach $47.8 billion by 2030. Dust had built the operating system for that market before most enterprises understood what they were buying.