Within two years, Rob Seaman believes, Slack will have more AI agents using it than humans. Seaman is the general manager of Slack, and he made the prediction at Salesforce's Trailblazer DX conference in San Francisco, in the tone of someone describing an outcome he considers essentially inevitable rather than aspirational.
The platform already handles more than a billion messages a day, a milestone that sounds human-scale until you notice that a disproportionate share of that growth is coming from agents, not people. These are not silent background processes. They are active participants: routing requests, handling queries, coordinating across humans and other systems. What Slack will be called when agents outnumber its human users, Seaman admits he is not sure. "Hub" feels tired. "Operating system" is played out. The word he keeps reaching for does not quite exist yet.
The broader data suggests Seaman is not describing a fringe scenario. A Gartner survey of 469 CEOs and senior business executives, conducted across late 2025, found that 80 percent expected AI to force a high-to-medium degree of change to their operational capabilities within a few years. The shift they anticipate is not simply more automation. Gartner frames it as a move from "digital business" to "autonomous business": a state where self-learning software agents make decisions and take actions with minimal human intervention. Fifty-four percent of those surveyed described their current automation as limited to specific tasks. By 2028, only 13 percent expect to remain at that level.
The economic implications are already starting to shape how companies think about revenue. Twenty-eight percent of CEOs identified transactional revenue as the area most at risk from AI agents, because automated purchasing, pricing, and negotiation systems remove the intermediate steps where transaction fees and intermediaries normally live. Models that worked for decades are being quietly rendered obsolete before most of their practitioners have noticed.
That is the executive view. The operational reality is messier. A survey of 1,000 business decision-makers across the US, UK, Germany, and France, commissioned by enterprise software company Infor, found that while 80 percent of organisations believe they have the internal capability to implement AI, 49 percent remain in early deployment stages, often limited to pilots that have not scaled. The barriers cited most often: data security and compliance concerns (36 percent), lack of internal AI talent (25 percent), and unclear return on investment (23 percent).
The gap between executive intent and operational execution has a structural explanation. A new analysis from Bain describes most AI agents today as "useful but forgetful": capable of completing tasks when prompted, but operating as distinct episodes rather than continuous participants. Each new session starts largely from scratch. Bain distinguishes this from true continuity, where a system maintains goals, accumulates domain-specific knowledge, and builds on the history of prior decisions rather than reconstructing it each time. A new class of persistent, long-running agents is beginning to emerge, designed to close that gap. Whether they arrive fast enough to match the transformation timelines showing up in CEO surveys is the open question.
What the Slack prediction and the Gartner data share is an assumption that the transition is a matter of timing, not destination. The agents are already in the building. The question is how quickly they fill the rooms.