You’re reading an issue of “The AI Economy,” my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.
Slack is about to become a place where most of its users don’t need it — at least not in any human sense.
The platform has long been where work happens. That’s been the pitch since day one. But spend time with General Manager Rob Seaman and you start to wonder whether the work happening there will soon look nothing like what anyone imagined.
He points out that Slack users now exchange more than a billion messages a day, a first in the app’s history. The number certainly sounds like a human milestone, but it isn’t, at least not entirely. The disproportionate growth driving it isn’t people. It’s AI agents.
And that’s just the beginning.
“We’ve seen a tremendous increase in the number of AI agents that have been built by our customers and deployed into Slack,” Seaman tells The AI Economy in an interview on the sidelines of Salesforce’s Trailblazer DX (TDX) conference in San Francisco earlier this month. “We really are seeing it be the place where…humans and agents are working together.”
He believes that won’t last. Within two years, Seaman says, “there will be more agents as users in Slack than there will be humans.” What that makes Slack, he admits, is harder to define. It won’t be seen primarily as a communication platform, he argues—more of a mashup where agents aren’t tools running in the background but active participants in the conversation. Seaman doesn’t have a name for it yet: “hub” feels tired, while “operating system” is played out, even though Salesforce branded Slack as the “agentic OS for the enterprise.”
That said, Slack’s GM concedes the OS analogy holds in one important way: just as an operating system abstracts hardware complexity from the end user, Slack has always abstracted the complexity of its integrations. The difference now is that what’s being abstracted isn’t software. It’s agents.
“It’s a place where people are going to spend most of their time, and there will be more agents there than there will be people,” Seaman concludes.
Slackbot Will Be the Main Way to Use Salesforce
Interacting with traditional software meant learning its language—navigating menus, clicking buttons, hunting for the right field in the right screen. Businesses hired people whose entire job was knowing where everything lived. That’s changing. The interface that once defined how we used computers is giving way to something far simpler: a chat field, a prompt, a request asked in plain English.
Salesforce is repositioning itself around that reality. The CRM that once demanded its own screen, its own login, its own mental context switch is receding into the background—becoming load-bearing infrastructure rather than a destination. The shift became visible at Dreamforce last year, when Salesforce debuted natively accessible apps inside Slack: Agentforce for Sales, IT service, HR, and Tableau Next. The product wasn’t gone. It had just stopped asking to be seen.
Now, it’s gone further. With the move to a headless model, Salesforce is offering the ability to run entirely underneath—no interface, no context switch, invisible by design. But if there’s no Salesforce screen, how will we interact with the applications? That’s where Slack, or more specifically, Slackbot, plays a critical role.
“Slackbot is going to be the primary way that customers use Salesforce moving forward,” Seaman says, referencing the recently upgraded Slack agent wth new capabilities that make it the “ultimate teammate.” He contends that Slackbot will be the main user of Salesforce’s Headless 360 offering.
“We’re already seeing it internally,” Seaman points out, noting that 60,000 employees are using Slackbot weekly to read and update the CRM platform. “The second-highest tool call after Slack search itself within Slackbot is the Salesforce MCP server. We’re seeing that within Salesforce [and]…in our early customers. I absolutely think the best manifestation of the Headless 360 is going to be Slackbot consuming the Salesforce MCP server, but also consuming Agentforce agents as tool calls themselves.”
The capability gap from acquisition to now is visible in something as simple as a slide deck. Seaman describes asking Slackbot to help him visualize a set of ideas—it searched Slack for context, found the Google Slides connector, pulled the corporate template, and produced a formatted presentation. “Previously, I would have had to go do many of those steps myself, or build an automation,” he notes. “That type of stuff just kind of happens magically now.”
It raises an obvious question: if Slackbot is becoming the primary interface for Salesforce, isn’t it just another AI assistant competing for attention alongside ChatGPT or Claude? Seaman is direct about what it isn’t: “We don’t want Slackbot to go be a standalone assistant that competes with Claude Code or ChatGPT,” he says. “That’s not a winning battle or even a smart strategy.” The differentiator, in his telling, isn’t raw AI capability—it’s context. The value of Slackbot comes from wedding whatever you’re doing back to the institutional memory that already lives inside your Slack. Take it out of that environment and you’ve lost the thing that makes it worth using.
Measuring Agentic Success
If agents are about to outnumber humans on Slack, the obvious question is how you measure their impact. Message volume tells you something about human activity—it tells you almost nothing about whether an agent is actually delivering value. Seaman says Salesforce is working on a new concept to address that gap: the agentic work unit. Unlike token counts, it’s “an actual indication of work being completed,” such as a tool call, an action taken in another system, something that shows the agent did something that mattered.
For Slackbot specifically, Seaman says Slack tracks three things: penetration, retention, and intensity. How many users with access are actually using it? Of those, how many come back the following week? And among those returning users, is usage growing over time? “If they’re coming back and using it, is there any intensity of doing so growing over time—are they getting more value from it?” he says. This framework is simple by design. In a landscape where AI features are easy to ship and hard to prove, Seaman argues these three signals cut through the noise. They show whether an agent is creating genuine habits, not just first-time curiosity. Although this approach is used for Slackbot, he believes it can be applied to other agents as well.
Seaman still believes Slack is the place where work happens, albeit the definition of work has evolved, along with its nature, scope, and breadth. The question is no longer whether AI belongs in the enterprise—it’s where it lands. And Seaman’s answer is consistent: it lands in Slack. The institutional memory is already there. The customer data is already connected. The 175,000 companies that run their businesses on Salesforce are already inside the ecosystem.
For an AI agent seeking context, there may be no richer environment in the enterprise than a Slack channel that has accumulated decisions, conversations, and institutional knowledge over the years. Seaman isn’t just predicting that agents will outnumber humans on Slack. He’s arguing that Slack was always where they were going to end up.
Disclosure: I attended Salesforce TDX 2026 as a guest of the company, with my flights and hotel stay paid for. The AI Economy’s coverage is editorially independent from those that it covers. These words are my own.Featured Image: Slack logo on display at Salesforce TDX on April 15, 2026. Credit: Ken Yeung
