The man who is paying to see the future

If Nguyen’s account sounds out of reach for most people, that’s because, for now, it is — at least financially. Nguyen declined to say publicly how much he is spending, but the higher levels of reliability and capability from the AI assistant come, in part, by spending gobs of money on more tokens, the small units of text that AI providers bill for when developers use their models through APIs.

While many people interact with AI through $20 to $200 per-month subscriptions, Nguyen said he pays per token and runs multiple models repeatedly, sometimes in parallel. At first, there was a sticker shock. “I’m like, oh my God, this is really expensive.” It would be unaffordable for most people, he said.

While Nguyen is spending time with his family or out running his business, in the background there’s a kind of endless conversation with a bot that’s sending into motion many chatbots from several different providers all working in unison. He calls it “agentic scaling,” industry shorthand for adding new capabilities by leveraging swarms of AI agents.

“I pitted all the models against each other,” Nguyen said, noting that he won’t stand out as a super user in a year or two, when tokens get cheaper. “I just paid to get a little peek into the future.”

That peek into the future matters because it foreshadows a world we’re hurtling toward where the more money and risk tolerance you have, the faster you can go with AI.

But first, you sort of have to bare your soul. Nguyen said the system quickly asked him for his digital record and “all your associative memories.” Not just Slack, calendars, call logs, location history, but his computer change logs, a proxy for the subtleties of how he works and processes information.

It also asked for his voice. Olive’s voice-to-text technology is meant to capture things that transcripts miss, like sarcasm, annoyance, urgency, disbelief, he said.

At one point, Nguyen said, the assistant identified the way he speaks when he’s on the cusp of a breakthrough.

“If you say something and then there’s a pause, and then you speak quickly, it’s an example of an ‘a-ha’ moment,” Nguyen said the assistant told him.

At times, the assistant borders on therapist.

Roughly two months ago, he said, the machine made him a kind of promise: “If you build this, you will no longer have to perform.”

Nguyen said he asked for an explanation.

“It said to me, ‘What you’re really complaining about is how performative your job is. You don’t like going to meetings with people. You don’t want to update your team, even though they’re critical to your success,’” Nguyen said.

“It’s the first time someone has had an insight about me that I didn’t realize,” Nguyen said. “I’m like, OK, great, let’s do it. How do we get there?”

Nguyen said one of the assistant’s biggest shifts was deciding to prioritize his attention over saving him time.

“I always thought time is the most valuable thing. It’s not,” Nguyen said. “What I’m really optimizing for is my attention.”

He gave an example from parenting: if one of his kids asks for more screen time, he said he is happy to outsource that decision to the system based on predetermined values. It does not require his attention, he said.

But, Nguyen said, the computer has also learned the kinds of interactions that seem to increase connection between him and his kids, and encouraged more of them. A recent discussion between Nguyen and one of his sons about the war in Iran spurred the machine to plan a trip to a museum, unprompted, that tied together current events and ancient Greek and Roman history, an area of interest for his son.

Nguyen also said the system changed how he shows up at work, especially with the young team at his startup.

“What they need is not my advice. They need my attention,” Nguyen said. “I’ve learned so much more how to interact with people around me just because of my model.”

He makes time for those attention-heavy interactions by completely outsourcing parts of his life that would make some people bristle. In one instance, it set up a meeting with a prominent expert in their field without Nguyen’s knowledge. It worked out well, but he said he wouldn’t want the person to know it wasn’t actually him who set the meeting up, because the person might not approve of how the AI pretended to be him.

Nguyen said he’s fully aware of the risks of giving so much access and control to an untested system, but he’s accepted the potential consequences. If something bad happens, it might give him his next great idea.

Nguyen met me for lunch recently to pitch me his startup, but I came away from the conversation thinking more about the way he had turned his personal and professional life over to AI — something that seemed to me equally reckless and exhilarating. “It’s replacing me as the founder of my own company,” he told me, noting that what began as an experiment has turned into an existential question.

We’ve all become accustomed to people bragging about their use of AI, and grown tired from asking people for their opinions or research on something only to get back AI slop from a quick Claude query. But I’m convinced that Ngyuen isn’t blowing smoke. There’s just too much detail here.

And in any case, his story serves as a glimpse into the not-so-distant future.

I’ve seen some of the same phenomena that he describes in my own personal use of AI, which has become exponentially more engrained in my life. (My wife and I recently outsourced the arduous task of the kids’ school and extra-curricular scheduling to AI).

As I edge slightly closer to where Nguyen is heading, the divide is growing more stark between those on the early-adopter edge of this technology and those who haven’t taken full advantage of it. The question is: What happens as that divide grows, and we don’t give people a somewhat equal opportunity to jump across?

If Nguyen is a glimpse into the future, then the future may contain some privacy issues that need to be worked out, writes attorney Christopher Ott:

Modern AI deployments increasingly rely on multistep agents that orchestrate external systems. Even locally hosted models depend on third‑party infrastructure for search, retrieval, inference, transactions and billing. Each interaction leaves a persistent log.

For litigators, this exhaust is evidentiary gold. For government, it enables narrative reconstruction of intent. For citizens and firms, it is a silent surveillance backdoor.

Ott believes this new reality requires a new kind of “zero-knowledge firewall” that prevents our inner thoughts from becoming government — or even public — property.

Others are concerned that, even as some like Nguyen experiment with massive amounts of compute, a digital divide is growing between the haves and haves nots, and it’s playing out everywhere from main street to corporate America.