Hello, dear readers!
This is our weekly brief on remarkable AI topics, so you can see where the AI boom is meeting the limits of everyday business reality.
Today's focus — the OpenAI CEO has found one thing he does not want to automate: his own messages. Sam Altman says he went back to typing some emails and Slack messages himself after discovering that people actually care about human interaction. What else might companies be over-automating?
Also in this week's edition:
1. Starbucks and Uber are getting cold feet about AI, rolling back initiatives and rethinking costs.
2. Nvidia and Microsoft are betting on agentic PCs to make AI actually useful.
Even OpenAI's CEO Wants A Human Touch
Sam Altman says AI is unlikely to cause the “jobs apocalypse” some expected — at least not yet. Speaking at a Commonwealth Bank of Australia conference, the OpenAI CEO admitted that while the company had been broadly right about the technology’s capabilities, it had been “pretty wrong” about its near-term social and economic impact.
Altman said he had expected more entry-level white-collar jobs to be eliminated by now. Instead, he now believes many jobs contain a “human part” that is harder to replace than the AI industry once assumed.
His own inbox helped make the point. Altman said he had experimented with using AI to respond to Slack and email messages, clearly labeling the replies as coming from “Sam’s AI.” But the experience convinced him that people still care deeply about whether they are interacting with a person or a system.
So yes, the headline is technically true: Sam Altman stopped using AI. Not entirely, of course — just for some messages where outsourcing the interaction made it feel less meaningful.
Altman says he is “delighted” to have been wrong about the speed of job disruption. That may be true. Though given that OpenAI is still building increasingly powerful systems for automating knowledge work, one suspects the delight has its limits.
Business Gets Second Thoughts About AI
The AI reality check is not limited to philosophical reflections about human connection. In some companies, the question is becoming much simpler: is the damn thing actually worth it?
Starbucks is ending its AI inventory management system only nine months after introducing it. The tool used computer vision to help stores count inventory and was meant to simplify record-keeping, improve accuracy, and reduce stockouts. In practice, Reuters reported that the system miscounted or mislabeled items.
One employee's comment shared by Starbucks thanked the company for “trusting the partners over unreliable spatial recognition.” Another said staff feedback about AI counting had finally been heard. Not exactly the testimonial AI vendors dream of.
At Uber, the problem is cost. COO Andrew Macdonald says the company is struggling to connect rising AI spending with meaningful improvements in consumer products. AI coding tools are now widely used by engineers, but more token consumption does not automatically mean better features.
The issue became harder to ignore after Uber reportedly used up its entire 2026 Claude Code budget in just four months. Macdonald said the company may need to compare token spending with other costs, including headcount: if AI use cannot be clearly tied to useful features shipped to users, the tradeoff becomes harder to justify.
In other words, shipping more code is not the same as improving the product.
That may be the next stage of corporate AI adoption. Not “how many employees are using AI?” but “what are we getting for the money?”
Agentic PCs Enter The Chat
Still, it is not all doom and gloom for AI. If some companies are pulling back from over-automation, Nvidia and Microsoft are pushing a different idea forward: maybe AI becomes more useful when it runs locally, privately, and directly on your computer.
Nvidia has announced RTX Spark, its first full consumer PC chip family for laptops and mini-PCs. The company says the chip can power high-end creative work, gaming, and local AI agents, with enough capacity to run very large models on-device.
The pitch is not just faster laptops. Nvidia describes this as a shift toward “personal AI” — computers where agents can control apps, handle repetitive tasks, and work across the system under user supervision. Microsoft is also preparing new Windows security and containment tools so these personal agents can run more safely and under user control.
That matters because it addresses two of the biggest complaints from the other stories. Local AI could reduce dependence on expensive cloud tokens, and keeping data on-device may make personal agents feel less risky for users and businesses.
There are still plenty of caveats. Nvidia has not shared detailed performance benchmarks, and pricing remains unclear. AI, meanwhile, is still unreliable, expensive, and occasionally useless. Even for emails.
But the next wave of hardware suggests the industry is not giving up. It is trying to make AI feel less like a cloud service and more like part of the computer itself.
Thanks for reading AIport. Until next Monday — by then, AI will almost certainly have generated a very confident business case for something nobody asked it to do.
