Global – October 27, 2026 -- Dust, a collaborative AI platform, is targeting three forms of enterprise AI dependency: reliance on a single model provider, loss of control over agent-generated data, and limited visibility into costs and behavior, according to Co-Founder and CTO Stanislas Polu.
Dust builds a shared layer between staff, company data and multiple models
Polu said Dust lets teams work with AI, switch among supported frontier and open source models, and govern the context, agents and traces their work generates. The platform is multimodel by design, so companies can change models without rebuilding their context and workflows.
Polu argues that agent traces, which record instructions, context, model and tool choices, outputs and human feedback, may become one of a company's most valuable AI assets. He said that corpus could later help companies evaluate or train models for their specific work.
Snowflake serves as a connected structured-data source for Dust agents
Organizations can connect Snowflake to Dust so teams can query warehouse data in natural language and combine it with context from a CRM, Slack, Notion or call notes. Snowflake remains the source of truth, with the company's permissions applied. Polu said data teams can encode trusted access and business logic while more colleagues use the data.
Founders carry lessons from Stripe and TOTEMS into product strategy
Polu and co-founder Gabriel Hubert saw coordination problems at Stripe as it grew from roughly 150 people into a much larger organization, with information lost and decisions harder to trace. At their earlier company TOTEMS, they did not test early enough how large the market could become.
That experience produced an early Dust principle: "No product-market fit, no GPUs." The company chose the product layer over model training.
Polu's operating rule for the platform: agents propose, humans dispose.