The Signal #006 — Dakota’s read on the AI news that actually matters to people running a business.

There’s a gap between playing with a tool and understanding a process. Most of us who run field service businesses know this instinctively. You can watch a tech swap out a blower motor on YouTube and think you understand HVAC. You don’t. Not until you’ve done it in a crawlspace at 95 degrees with a customer calling you every 20 minutes.

That gap is now playing out at the top of the tech industry, and the consequences are landing on real people.

What happened

Box founder Aaron Levie posted something unusually candid on X last week. “CEOs are uniquely prone to AI psychosis because they’re sufficiently distant from the last mile of work that still has to happen to generate most value with AI,” he wrote. His point: executives play with AI, get a working prototype or a drafted contract, and then make the leap to believing agents (AI systems that take action on their own) can handle the actual job. They skip over the bug reviews, the hallucinated references, the edge cases, the days of combing through fine print.

This isn’t coming from an AI skeptic. Levie backs AI startups and posts AI positivity to 2.7 million followers. He’s saying the hype is outrunning reality, and the people driving decisions are the ones least equipped to see the difference.

The numbers back him up. In just the first five months of 2026, 115,430 people were laid off from 152 tech companies, according to Layoffs.fyi. That’s nearly as many as the 124,636 let go across all of 2025. A lot of those cuts were credited to AI. ClickUp’s CEO publicly laid off 22% of his staff after deploying roughly 3,000 AI agents, calling it the foundation of a “100x org.”

Meanwhile, a meta-analysis published in UC Berkeley’s California Management Review found “no robust relationship between AI adoption and aggregate productivity gain.” MIT researchers concluded that agents aren’t doing human-quality work yet in many cases, and predicted models will handle most text-related tasks at a “minimally sufficient quality level” somewhere between 80% and 95% success rates by 2029. That’s three years away, and that’s for text tasks. Not field diagnosis. Not customer judgment calls. Not the nuanced stuff that keeps a service business running.

Why it matters for operators

You are not a tech CEO. You don’t have a board to answer to, and you’re probably not laying off a quarter of your staff based on a demo. But the same logic Levie is warning about can creep into smaller businesses too.

Someone sells you a voice AI that handles inbound calls. You test it on a Tuesday afternoon, it books three jobs cleanly, and you think: I can cut my dispatcher. Or someone shows you an AI that writes estimates, and after one clean run you start wondering if you need your office coordinator.

That’s the happy path. It’s real. And it’s also not the whole picture.

The whole picture includes the customer who said “Thursday” but meant the Thursday after the holiday. The job that needed a follow-up call because the address was a suite number. The estimate that was missing a materials line because the description was vague. The AI handled none of those. A person did, quietly, and you didn’t notice.

Levie’s advice applies here too. Use AI a lot. See what it actually does. “Come out the other side with an appreciation for both the upside and the real work.”

What most people get wrong

They treat AI as a replacement decision before it’s proven itself as a support tool.

The research published in the Harvard Business Review makes a point worth sitting with: when everyone uses AI to produce more output, the bottleneck shifts to whoever has to review and approve that output. You don’t eliminate work. You move it. If your AI is booking jobs, generating estimates, and sending follow-ups, someone still has to catch the errors before they become problems. That job didn’t disappear. It just got less visible.

The MIT researchers also note that even at 80% to 95% success on text tasks by 2029, that remaining gap is where your reputation lives. One bad job booked wrong, one estimate sent to the wrong customer, one follow-up that sounded off. In home services, those aren’t data points. They’re Google reviews.

The takeaway

AI is a real tool. The productivity gains are real, sometimes. The automation is real, in specific, well-defined spots. What isn’t real yet is the version where you hand over a process you don’t fully understand and trust it to run cleanly without supervision.

Know your processes first. Automate the parts that are actually repeatable. Keep a person close to anything where a mistake has a name and address attached to it.

If you want to think through where AI actually fits in a field service operation, without the hype, start at xovionlabs.com.