Zero. That is the number of people who actually believe a CEO of a trillion-dollar AI company when they say the technology won’t take your job.
Mustafa Suleyman, the head of Microsoft AI, recently spent some time on The Verge AI podcast arguing that superintelligence is looming on the horizon, but that it will act more as a partner than a replacement. The narrative is familiar: the AI handles the drudgery, the human handles the “high-level” thinking, and everyone wins. It’s a cozy picture of the future where we all become managers of digital interns who never sleep and don’t ask for equity.
Suleyman’s take is that we are moving toward a world of personal agents that understand us deeply and execute complex tasks. He frames this as an augmentation of human capability. (Which is a convenient way to avoid talking about the actual cost of H100s). He suggests that the shift won’t be a sudden cliff where millions of people wake up to find their Slack accounts deactivated, but rather a gradual transition.
He talks about the proximity of superintelligence as if it’s a weather pattern we should all just prepare for with a sturdy umbrella, ignoring the fact that the umbrella is being sold by the same person who is seeding the clouds. Why would a company pay for three developers when one developer and a superintelligent agent can do the same work for the price of a few API calls (and probably a very expensive Azure subscription)?
The “augmentation” argument is the standard corporate shield. It’s like a restaurant owner claiming that a high-speed, automated oven doesn’t make the pastry chef obsolete; it just lets them “focus on the art” of the cake. But in the real world, the owner eventually realizes they only need one “artist” and three ovens to produce the same volume. The economics of efficiency always win over the sentimentality of craftsmanship.
The disconnect here is between the technical capability and the corporate incentive. Suleyman is talking about what the technology can do to help us. The CFO of a Fortune 500 company is looking at what the technology can do to reduce the headcount of a mid-level operations team. When superintelligence arrives—or whatever marketing term we use for the point where the model stops hallucinating and starts reasoning better than a senior analyst—the goal will not be to make the analyst “more productive.” The goal will be to reduce the number of analysts required to maintain the same output.
It’s a corporate fairy tale.
We’ve seen this movie before. We were told the ATM wouldn’t kill the bank teller, and while tellers still exist, the ratio of tellers to customers has plummeted. The difference this time is the speed of deployment. Software doesn’t need to be trained in a trade school; it just needs a weight update and a deployment pipeline.
If we look at the trajectory of agentic workflows, the “partner” phase is just a transitional state. Once an agent can reliably handle the end-to-end lifecycle of a ticket—from triaging to coding to deploying—the human in the loop becomes a bottleneck. The pressure to remove that bottleneck will be irresistible.
By Q2 2025, the industry will shift from talking about “copilots” to announcing “autonomous agents” that replace specific entry-level roles in bulk.
The real question isn’t whether the AI will take the job, but who gets to keep the surplus value created by the automation. If the productivity gains from superintelligence only flow upward to the people owning the compute, then the “partnership” Suleyman describes is less of a collaboration and more of a takeover. He can promise us the world from the comfort of a Microsoft executive suite, but the people writing the code know that “augmentation” is usually just the first step toward “replacement.”
