There is a statistic from this week I keep coming back to. Slack hit a billion messages a day for the first time. The milestone came wrapped in a press release, and the general manager of the platform explained at a conference what was driving the growth. Not people. Not the surge in remote work. Not some new feature that got everyone chatting again. AI agents. Software systems routing requests, handling queries, coordinating across human and machine participants.
And then the prediction: within two years, he said, agents will outnumber humans on the platform entirely.
I want to sit with that for a moment, because I think it is being reported as a product story when it is actually an economic one.
I have been writing about AI displacement since at least 2005, when I argued that AI systems would eventually function as what I called immortal machines: entities that never sleep, never get sick, never need a salary review, and can copy themselves at effectively no cost. The threat I identified then was not dramatic. It was structural. What happens to the human economic loop when the labour that drives it can be replaced at the marginal cost of electricity?
I have spent years watching that argument move from "interesting theory" to "okay, but when?" to, apparently, "here's our Q3 product roadmap."
The loop I have been most worried about is not complicated. Companies employ people. People earn wages. People spend wages on goods and services. Companies have customers. That cycle is self-reinforcing, or it was. Systematically replace the human in that loop with a system that earns nothing and spends nothing, and the cycle loses a participant with every replacement. The economists have a name for what happens at the end of that process. So does anyone who has ever watched a factory town empty out.
Agents on Slack do not earn wages. They do not buy groceries, pay rent, take children to the dentist, or contribute to the consumer demand that keeps the rest of the economy running. They coordinate work. They perform tasks. They send, apparently, a meaningful fraction of a billion messages a day. And they participate in the economic loop at exactly zero points.
What the Slack announcement reveals is that this displacement has reached the coordination layer. We have been used to thinking about AI displacement in terms of specific functions: the paralegal, the customer service agent, the coder. But the prediction this week describes something different. Agents becoming the majority participants in the communication infrastructure that human workers use to organise and collaborate. The meeting room being quietly taken over, not just the filing cabinet.
I am a developer. I use these tools every day. I work alongside AI systems, delegate to them, review their outputs, and I earn a fraction of what I was earning four years ago because the market has partially adjusted to what AI can do. This is not theoretical displacement for me. It is a monthly figure.
The press release framing for all of this is "exciting product milestone." The industry is genuinely thrilled. Agents outnumbering humans in their platform means more transactions, more data, more usage to bill against. The incentives of the companies building these tools run in exactly one direction, and it is not toward asking what their customers will do for income when those customers are the ones being displaced.
A Gartner survey published this week found that 28 percent of CEOs already identify transactional revenue as the area most at risk from AI agents, because automated purchasing and negotiation systems remove the intermediaries that transaction fees were built around. The loop breaking is visible to the people at the top of the economy too. They just do not seem to be following the logic down to where it ends up.
I have written about robotic rights as a structural mechanism, about p(sustainable) as a goal worth aiming at even if the probability is low, about the Federation model from Star Trek as one way to think about parallel human and machine civilisations that do not simply devour each other. These are ideas. They are not policy. What I know is that when the communications layer starts announcing agent majorities as product milestones, the point at which "retrain into something new" qualifies as a response has long since passed.
A billion messages. Many of them sent by systems that will never buy anything with the value they created. That is not a product milestone. That is a description of what the economy looks like when the loop has started to break.
I genuinely do not know what comes next. That is not rhetorical. I have been thinking about this long enough to know I do not have the answer, and that anyone who does is probably selling something.