Zero Churn, 300,000 Agents: Why Dust’s Multiplayer AI Sticks

01 /The Concept

When Dust published its 2025 metrics, investors noticed a number that should not exist in enterprise SaaS: zero churn. Not low churn – zero. Three thousand organizations, 51,000 monthly active users, 300,000 agents deployed, and not a single customer left.

The reason is architectural. Gabriel Hubert and Stanislas Polu founded Dust in Paris in 2023. Hubert had previously co-founded a data analytics company that Stripe acquired in 2014; Polu joined OpenAI as a research engineer, co-authoring reasoning papers with Ilya Sutskever. Their observation from that vantage point: every enterprise AI product treats agents as individual assistants. Nobody had built the collaboration layer.

What Dust built is what they call multiplayer AI – a platform where AI agents and human teams share the same workspace, context, task queues, and governance. Instead of each employee getting a personal AI assistant that operates in isolation, teams get a shared environment where agents hand off to other agents, pull from the same knowledge bases, and work toward shared goals under unified oversight.

Sequoia and Abstract led a $40 million Series B in May 2026, with Snowflake and Datadog also investing – bringing total funding to over $60 million.

02 /Why It Matters

Most enterprise AI deployments fail not because the models are bad but because context does not travel. A sales rep asks their agent to draft a proposal; the agent does not know the deal history, the pricing exceptions approved last month, or what the customer’s security team flagged on the last call. The proposal is wrong. Someone has to fix it – which means the AI consumed time and introduced new errors without saving any.

The standard fix is retrieval-augmented generation: connect the model to company data and let it search. Dust does this, but the non-obvious insight is that retrieval solves the knowledge problem without solving the collaboration problem. Teams do not just need agents that know things – they need agents that know what the team is working on right now, including tasks in progress, decisions pending review, and work assignments that changed this morning.

Dust’s architecture treats agents as team members with shared state rather than isolated assistants with individual memory. The Snowflake and Datadog investments signal something specific – both companies built enterprise infrastructure platforms that became deeply embedded in customer operations. Their participation suggests they see the same trajectory: build the workflow layer early enough, and switching costs make it permanent.

03 /Opportunities

The 100+ data source integrations Dust ships are a switching cost in plain sight – every integration is a new connection that teams build workflows around, and every workflow is a reason not to migrate. The moat compounds with usage.

The competitive risk is distribution. Salesforce, Microsoft, and ServiceNow are all building agent orchestration layers with the advantage of already being embedded in the enterprises Dust needs to reach. Dust’s counter is speed and product focus – they are not protecting a legacy product line.

The entry angle for builders: Dust has proven enterprises pay for agent orchestration that preserves human context. The verticals where that proof is thinnest – manufacturing floor operations, logistics coordination, regulated healthcare workflows – are where purpose-built alternatives could win on domain depth before horizontal players catch up. Going deeper on the operational specifics those sectors actually need, rather than competing with Dust directly, is the move with the highest expected value.