Startup Spotlight: Dust’s Multi-Model AI Platform
Dust’s collaborative AI platform helps companies centralize and govern AI workflows, preserving control over data, context, and reusable traces while supporting multiple models.
Stanislas Polu, Dust’s Co-Founder and CTO, describes how the startup addresses a core challenge of modern AI adoption: balancing abundant intelligence with organizational control. Large language models have made AI widely accessible, but fragmented adoption risks creating dependencies on single providers and losing visibility into costs, data, and workflow traces. Dust aims to serve as a shared layer that unifies company data, existing systems, and AI models, allowing teams to collaborate without surrendering ownership of their operational context.
Polu highlights three dependencies that workplace AI often introduces: reliance on a single model provider, loss of control over generated data and traces, and limited oversight of costs and agent behavior. He argues that the traces—records of instructions, model choices, tools, outputs, and human feedback—can become a company’s most valuable AI asset. Dust enables teams to capture, govern, and reuse these traces, turning them into reusable company capabilities while avoiding lock-in to any single intelligence source.
The platform integrates with systems like Snowflake to combine structured data with broader company context, such as CRM records or Slack conversations. By bringing these elements into a shared system, Dust allows AI operators to analyze successful workflows, refine agents, and make improvements reusable across teams. This approach preserves Snowflake as the structured data foundation while enabling natural-language queries and repeatable workflows without rebuilding data strategies around AI vendors.
Polu emphasizes that companies should not treat AI as a zero-sum race with a permanent winner. Dust supports multiple frontier and open-source models, allowing organizations to adapt choices as capabilities, costs, or policies change. The durable asset, he argues, is the company’s accumulated context, decisions, and workflows—not the model itself. This principle underpins Dust’s design: enabling sovereignty through control over how intelligence is used, rather than isolation from external providers.