AI has rapidly transformed the workplace, taking on the repetitive, time-consuming work that once ate into people's days.
But according to Gabriel Hubert, CEO and cofounder of AI company Dust, the workplace is in the middle of being moved on another notch, with a significant shift towards multiplayer AI.
Instead of single employees using isolated tools, multiplayer AI involves agents becoming shared and used collaboratively across different departments. These agents learn from a company's data, allowing teams to automate processes and share workflows.
In an interview with Sifted, Hubert unpacks how companies can shift from single-player AI — where value and productivity is created by the individual — to multiplayer AI used collaboratively across teams.
The shift to multiplayer AI
The typical way AI has been used in many companies is as follows: an employee opens a chatbot, pastes in the context it needs for a task, sends a prompt, gets an answer back and either uses that information to complete a task or moves onto something else.
While this can increase individual productivity, it has its limitations, according to Hubert. “One person learns a better way to do something, but that improvement doesn’t necessarily spread,” he says. “The individual gets faster, while the company continues to work in roughly the same way.”
Inside companies operating with a multiplayer AI model, agents can “start participating in the same workflows as other people and other agents,” he adds. “They can hand work over, reuse what another team has learned and contribute to a shared system rather than starting from scratch every time.”
Multiplayer AI agents essentially act as ‘digital teammates’ — users can ‘@’ mention a particular agent in a collaborative workspace, ask it to perform a task and then hand over to another agent.
The important part is that the workflow is shared, improves over time and becomes available to the rest of the team.
For example: a ‘blog writer’ agent can generate a blog post and then pass it onto a ‘LinkedIn’ agent to draft social copy based on the shared context.
Another area Hubert has seen multiplayer AI implemented effectively is in sales.
“Traditionally, a sales rep spends 30 minutes looking for information, deciding whether a lead is relevant and updating a CRM,” he says. “Different reps will inevitably do this differently.”
He adds: “In a multiplayer workflow, the sales rep can ‘@’ mention an agent that gathers the relevant data, applies the company’s qualification criteria, updates the CRM and routes the lead. The important part is that the workflow is shared, improves over time and becomes available to the rest of the team.”
As agents evolve, they “need to be able to do things, not only retrieve and summarise information,” Hubert adds.
Governance, security and scaling
As organisations transition from single to multiplayer AI, traditional IT governance models aren't always adequate.
Across the EU, 55% of large enterprises are using AI, but only a quarter of these companies believe their governance models are fully equipped to handle the implementation, according to data from Smarsh, a data and intelligence platform.
This can lead to ‘shadow AI’ within an organisation — the unauthorised use of AI tools or features by employees, in a way that isn’t governed by a central IT team.
For example, if an agent is granted access to a company's Google Drive, any user interacting with that agent might accidentally come across confidential records.
An agent inherits the data access of the space it has been built on and that access stays the same no matter who uses it.
At Dust, administrators manage the access control of AI agents, says Hubert. These admins decide which data sources are connected and how that information is organised.
“Some spaces are open to the whole company while others are restricted to particular teams or people. Employees can build and use agents drawing on the spaces they have access to.
“An agent inherits the data access of the space it has been built on and that access stays the same no matter who uses it. An admin can let the whole company use an agent even when most employees can't see the underlying data directly.”
At Dust, if a user doesn’t have access to specific data, the platform prevents the agent from surfacing that data to them.
“The context and operating knowledge of a company should belong to the company. It shouldn’t be trapped inside a model provider, an individual conversation or a series of disconnected applications,” Hubert adds.
Keeping a human in the loop
One of the major barriers to widespread AI use at work is that individual employees are often ready before their organisations are.
Research by Microsoft shows that organisational factors such as manager support, talent practices and culture drive more than twice the AI impact of individual employee effort alone.
At Dust, encouraging widespread use of multiplayer AI relies on “AI operators”.
This person is “the most important new role in the enterprise,” Hubert says. “What they share is a different way of looking at work. Instead of asking how AI can help with one task, they ask whether the process should still exist in its current form now that AI is available.
As agents absorb more execution, judgment becomes more valuable.
“The best operators run an ‘anti-to-do list’, asking after every annoying task, ‘how do I never have to do this again?’"
By 2027, Hubert hopes organisations will no longer be asking if they should use agents, but rather how they manage an agent workforce.
“Once a company has many agents participating in workflows, it needs to understand what they are doing, who is responsible for them and whether their decisions remain reliable,” he says.
He also suspects a more important challenge will remain, which will require human input.
“As agents absorb more execution, judgment becomes more valuable. The agents themselves are not the entire asset. The asset is the loop between the agents, the context the company owns and the people who keep improving both.”
