When the head of Axios spent the past year turning himself into a full-time AI test subject, he wasn't just playing with a chatbot between meetings. He loaded his medical records into Claude, built an always-on computer running multiple AI agents, and spent one to two hours daily wrestling with systems designed to know him better than he knows himself.
The experiment has surfaced a stark reality that separates genuine AI capability from the hype: the technology is radically more powerful than most people understand, but deploying it at corporate scale remains a nightmare of friction, security concerns, and agent-to-agent chaos.
Axios CEO Jim VandeHei's deep dive into AI across the company reveals something that tech evangelists rarely admit. AI is smarter than 95% of people on 95% of topics 95% of the time. His own experience with medical diagnosis showed an AI system that matched or exceeded his doctor's clinical judgment on complex cases. But that same system hits concrete walls the moment it needs to play nicely with other AI agents inside a corporate environment or access restricted company data.
The work required to extract real value is not optional. Casual users who prompt an AI once or twice and get boring results assume the technology is oversold and move on. That's the wrong conclusion. Success demands feeding the system copious amounts of information, persistently telling it what works and what fails, and rebuilding that feedback loop daily. Only then does what VandeHei calls "super-knowledge" emerge.
The personal toll is real too. VandeHei found himself jumping out of bed at 3 a.m., wired and creative, accomplishing more at 55 than in previous decades. But the same intensity that drives innovation appears to trigger anxiety and disrupted sleep patterns. He suspects causation, not coincidence.
Job losses and efficiency gains are not the story anymore, at least not the biggest one. A year ago, VandeHei expected AI to be a subtraction game: automate ruthlessly, cut headcount, reduce cost. Axios has done exactly that. But over the past three months, his thesis flipped. The real opportunity is building new business lines that were economically impossible before AI. Three new revenue projects at Axios would not exist without it. This suggests long-term hiring could exceed long-term cuts, despite near-term job displacement.
The scaling problem is where theory hits reality hard. Axios installed chief-of-staff AI agents for executives, but the moment two agents need to coordinate, the system breaks down. What can they know? What can they share? What decisions can they make together? These questions remain mostly unanswered across corporate America. Until agent-to-agent collaboration works reliably and securely, the transformative potential that VandeHei experiences individually will remain out of reach for most companies at scale.
One unexpected discovery has been easier to act on: some rank-and-file workers are wired for AI in ways that are obvious once you see it. They are not technologists or coding specialists. But they intuitively understand how to partner with AI to amplify their own output and their team's capacity. Spotting and training these natural accelerators has become simpler than expected, and it suggests that AI literacy, not AI expertise, is the real competitive edge.
The bottom line, according to VandeHei, is straightforward. For $20 a month, anyone can experiment with world-class AI models. The obligation is to be clear-eyed about what actually works, what fails, and what the real tradeoffs are. Curiosity and daily use matter more than hype or fear.
Author James Rodriguez: "VandeHei's honesty about both the superhuman capability and the absolute mess of enterprise deployment is refreshing, and it exposes why so many corporate AI rollouts are flopping while individual power users are seeing genuine transformation."
Comments