Dust CEO pitches shared AI agents over solo workplace chatbots

Dust is positioning its workplace AI platform around shared agents that teams can use together, rather than individual employees prompting isolated chatbots. No financing, valuation, revenue or customer numbers were disclosed, so the significance rests on the operating model Dust CEO and cofounder Gabriel Hubert described to Sifted, not on a new transaction.

Hubert said many companies still use AI in a one-person pattern: an employee gives a chatbot context, asks for an output and applies the result alone. That may speed up the employee, he argued, but it does not necessarily change how the wider company works or preserve the better process for others.

Dust’s pitch is that AI agents should sit inside shared company workflows, draw on company data and be reused across teams. Hubert described agents that can be mentioned inside collaborative workspaces, assigned a task and then hand work to another agent. One example he gave was a writing agent producing a blog post, followed by a social-media agent drafting LinkedIn copy from the same context.

Sales is another use case Hubert cited. In his example, a sales representative could ask an agent to collect lead information, check it against the company’s qualification rules, update the CRM and route the lead. The claim is that the process becomes consistent and reusable, rather than depending on how each salesperson handles research and admin.

Governance is the constraint

The model raises familiar enterprise concerns around permissions, auditability and employee use of unapproved AI tools. Smarsh, a data and intelligence platform, found that 55% of large enterprises in the EU use AI, while only a quarter of those companies say their governance systems are fully ready for implementation.

That gap can create what the industry calls shadow AI, where staff use AI tools outside central IT oversight. The risk becomes more acute when agents are connected to internal systems such as Google Drive, because an agent with broad access could expose information to people who should not see it.

Hubert said Dust handles this through administrator-controlled access. Administrators decide which data sources connect to the platform and how information is grouped. Some spaces can be available across the company, while others can be limited to specific teams or individuals.

According to Hubert, a Dust agent takes on the permissions of the space where it was created, and those permissions do not change based on who uses the agent. He said an administrator may allow broad use of an agent while keeping the underlying data inaccessible to most employees. Dust also says its platform blocks an agent from showing data to a user who lacks permission to view it.

The human role shifts, according to Dust

Hubert also argued that companies need people who redesign processes around AI, rather than treating AI as a helper for one-off tasks. He called this person an “AI operator” and said the role focuses on eliminating recurring work instead of optimizing each task in isolation.

Microsoft research cited by Dust found that organizational factors, including manager support, talent practices and culture, drive more than twice the AI impact of individual employee effort alone. That supports the broader point that adoption is less about access to a chatbot and more about whether teams change the way work is assigned, reviewed and governed.

Hubert said he expects companies by 2027 to spend less time debating whether to use agents and more time managing groups of them. That remains a projection from a vendor in the category. The harder questions are still accountability, reliability and whether agent activity can be inspected by the people responsible for the process.

Dust’s framing reflects a broader enterprise AI shift from experimentation to operational control. The company’s argument is that the value sits in the loop among agents, company-owned context and employees who keep improving the system. The unproven part is how broadly that pattern works once agents move beyond retrieving and summarizing information into taking action across business systems.

This story draws on original reporting from Sifted.