Multiplayer AI startup Dust raises $40M to help enterprises move beyond

Your salesperson spends an hour using AI to research a key account on Monday. On Tuesday, a solutions engineer sits down and does the exact same research — with their own AI agent, in their own chat window. Neither one knows the other already did this. The work doubles. The insight evaporates. And the AI that was supposed to make your team smarter just made it slightly more fragmented.

This is the problem that just attracted $40 million in funding. And it’s one that anyone thinking seriously about agentic AI — at any scale — should pay attention to.

What Dust Just Announced

Dust, officially known as Permutation Labs SAS, closed a $40 million Series B led by Abstract and Sequoia Capital on May 18, 2026. Snowflake and Datadog both participated — two companies that live and die by organizational data infrastructure, which tells you something about how seriously they’re taking this problem. Total funding now exceeds $60 million.

Dust’s Series A was $16 million, raised in June 2024, also led by Sequoia. That’s a more than 2x step-up in round size in under two years — and the traction numbers make it easy to see why. The platform is now used by over 3,000 organizations globally. More than 300,000 agents have been deployed across it. And in 2025, Dust reported zero customer churn alongside a 70% weekly active user rate.

Zero churn in enterprise SaaS. That’s not a statistic you see often. It’s worth sitting with for a moment.

Why the Single-Player AI Era Is Running Out of Road

The case Dust is making isn’t subtle. Most organizations today are running what its CEO Gabriel Hubert calls ‘single-player AI’ — every employee with their own chatbot or copilot, operating in their own private context. What one person’s agent discovers stays locked in that person’s chat window. It doesn’t compound. It doesn’t transfer. It just disappears when they close the tab.

Here’s the structural reason this happens: traditional enterprise tools were built for one-to-one interactions. Email is one-to-one. Chat is mostly one-to-one. The AI systems layered on top of these tools inherited the same architecture — which means they don’t create shared organizational memory. That’s not a bug in any specific tool. It’s a design assumption baked in from the start.

And the uncomfortable truth, surfaced in Dust’s own writing on the problem, is that introducing individual AI assistants can actually make collaboration worse before it gets better. Give every team member their own optimized AI workflow and they retreat further into independent silos — each one running faster, none of them talking to each other.

Dust’s answer is a shared collaboration surface where humans and agents coexist in the same projects. Conversations, artifacts, and to-do lists are visible across the organization — not siloed in individual windows. An intelligence layer connects to more than 100 enterprise data platforms including Slack, Notion, and Salesforce, giving agents access to full organizational context rather than just the last 10 messages you typed.

There’s also something technically interesting here that goes beyond the collaboration story. Dust’s Deep Dive research agent doesn’t use a single AI model — it uses OpenAI’s GPT-5 for planning, Gemini 2.5 Flash for web page summarization, and Claude 4.5 Sonnet for synthesis and writing. Each model handles the subtask it’s best at. The platform is explicitly model-agnostic by design, letting customers choose which underlying model powers each agent and redeploy agents to different models without rewriting any logic. If you’ve been watching the AI agent platform space, that kind of orchestration-without-lock-in is increasingly what separates platforms from point solutions.

What This Signals for the Agent Ecosystem

The Sequoia and Snowflake signal matters here. Sequoia led both rounds. Snowflake’s data infrastructure play is a bet that organizational memory — the thing Dust is trying to create — becomes a product category of its own. When a company whose entire business is making data queryable writes a check into a ‘shared AI memory’ startup, that’s a thesis, not just a financial bet.

Dust’s founders bring relevant credibility. Co-founder Gabriel Hubert helped scale AI adoption at Stripe. Co-founder Stanislas Polu was a research engineer at OpenAI, where he co-authored papers on AI reasoning. These aren’t enterprise software veterans bolting an AI story onto a CRM. They’re people who’ve thought about how AI reasoning actually works at production scale.

The practical implication for anyone evaluating enterprise AI infrastructure: the era of ‘deploy chatbots to individuals and declare AI transformation complete’ is ending. The next wave of AI ROI is going to come from shared context — agents that know what other agents have already done, build on previous work instead of starting from scratch, and create knowledge that compounds across the organization rather than evaporating when the session ends. You can read more about how this connects to the broader shift toward collaborative agent networks in our piece on agentic AI companies building the future in 2026.

What to Do With This

• Audit your current AI tool stack for context leakage. Map where insights from AI sessions go after the session ends. If the answer is ‘nowhere,’ you’re running single-player AI — and that’s the problem Dust is built to solve.
• Track Dust’s AI Operator feature closely. Non-technical employees in marketing, sales, and support can already build and deploy specialized agents without engineering involvement. This is the vector by which multiplayer AI enters organizations — not top-down mandate, but bottom-up adoption by the people who feel the collaboration pain most acutely.
• Watch the model-agnostic trend. Dust’s architecture — orchestrate across multiple models, no vendor lock-in, redeploy without rewriting — is becoming a design pattern across serious AI agent platforms. Platforms that tie you to a single model provider are making a bet that might not age well.
• Reframe your success metrics. The Dust blog frames it clearly: measure success less by minutes saved per task and more by whether decisions that used to require three Slack threads and a meeting now resolve in a single conversation. That’s a fundamentally different KPI than ‘productivity improvement.‘

What This Means for Enterprise AI in 2026

• Dust raised $40M in a Series B led by Sequoia Capital and Abstract, bringing total funding past $60 million, with Snowflake and Datadog participating.
• The platform serves over 3,000 organizations, has seen 300,000+ agents deployed, achieved 70% weekly active user rates, and reported zero customer churn in 2025.
• The core thesis: single-player AI (isolated chatbots per person) doesn’t create shared organizational memory, causing duplicated work and knowledge that never compounds.
• Dust connects to 100+ enterprise data platforms, uses multiple AI models simultaneously on the same task, and allows non-technical employees to build agents without engineering support.
• The broader signal: the next wave of enterprise AI ROI comes from shared context and collaborative agents, not more sophisticated individual tools. Platforms built for organizational memory are the infrastructure bet of this cycle.

The companies that get ahead of this shift aren’t the ones with the most AI tools. They’re the ones where AI-generated knowledge actually sticks — where one agent’s work becomes the foundation for the next. Every organization still running single-player AI is paying a compounding tax on every project: the cost of rediscovering what was already known. The technology to stop paying that tax exists now. The question is whether your AI infrastructure is built to take advantage of it.