From chatbots to ‘digital teammates’: The shift towards multiplayer AI

For the last couple of years, the default AI experience at work has been lonely by design: open a chatbot, paste in some context, get an answer, repeat. Gabriel Hubert, CEO of Dust, thinks that model is already becoming outdated. He calls what's replacing it "multiplayer AI," where agents don't just serve one person but get passed between teammates and departments, carrying context and improvements with them as they go.

The distinction Hubert draws is simple but has real teeth. In the single-player setup, if one employee figures out a smarter way to use a chatbot, that knowledge stays with them. The company as a whole doesn't get faster, just that one person does. In a multiplayer setup, an agent that drafts a blog post can hand off to another agent that turns it into LinkedIn copy, using the same shared context. Sales reps can tag an agent to pull lead data, apply qualification rules, and update the CRM automatically, a process Hubert says used to eat 30 minutes per lead and varied wildly from rep to rep.

The catch is that shared agents need shared plumbing, and most companies' IT governance wasn't built for this. Smarsh data cited in the piece shows 55% of large EU enterprises are already using AI, but only a quarter feel confident their governance can actually handle it. That gap is exactly what produces shadow AI — employees using unauthorized tools or granting agents access to things like a shared Drive without anyone in IT realizing what's now exposed. Dust's approach is to tie agent permissions to the data space they were built in, so an agent can be used company-wide even if most employees never see the underlying files directly. It's a reasonable fix, though it also quietly confirms how much trust companies now have to place in whoever configures these systems.

Hubert's bigger claim is organizational, not technical. He argues the real bottleneck isn't the AI, it's culture — citing Microsoft research showing management support and internal practices matter more than individual effort in determining whether AI actually changes how a company works. His answer is a new role he calls the "AI operator," someone whose job is to ask not how AI can speed up a task, but whether the task should exist at all anymore. By 2027, Hubert expects the conversation to shift from whether to use agents to how you manage an entire workforce of them — who's accountable, and whether their decisions can be trusted. His closing point, that judgment becomes more valuable as agents take over execution, is the one worth remembering. Multiplayer AI doesn't remove people from the loop. It just changes what they're for.