OpenAI's workplace data tells a familiar story, if you've been paying attention

This is the self-driving car hype cycle all over again in a weird way. Remember when we were told autonomous vehicles would eliminate human drivers by 2020? Then 2025? The technology worked great in demos and controlled environments, but the messy reality of actual roads, actual weather, actual unpredictable humans, that's where things got complicated. Same thing's happening here. ChatGPT works great when you're asking it to draft a marketing email. It gets shakier when the task requires judgment, context, or knowing that your company's CEO absolutely hates the word "synergy" and will reject any document containing it.

The education angle is where things get really interesting. OpenAI just announced they're rolling out ChatGPT access to the entire California State University system, that's 500,000 students and faculty. It's the largest deployment of ChatGPT in education to date, which sounds impressive until you think about what it actually means. Half a million people suddenly having access to a tool that most of their professors don't fully understand, that can write passable essays but also confidently state things that are completely wrong, that's going to be... messy.

I talked to a few professors (via email, because I'm a dinosaur and that's how I prefer to communicate) and the reactions ranged from cautious optimism to barely concealed panic. One computer science instructor told me she's already redesigned her entire assessment structure because take-home coding assignments are basically meaningless now. An English professor said he's actually excited because it forces him to assign more interesting prompts, but what do I know, maybe he's just putting a brave face on it.

The broader adoption data from OpenAI's consumer research shows something that surprised me, I'll admit. The gap between early adopters and mainstream users is closing faster than I expected. We're not just talking about tech bros in San Francisco anymore. Teachers in rural districts, small business owners, retirees trying to write better emails to their grandkids, the user base is genuinely broadening.

But here's where I get skeptical. OpenAI's economic analysis makes big claims about productivity gains and economic impact. And sure, the methodology looks reasonable at first glance. But we've been here before! Every major technology platform has commissioned research showing how transformative their product is. Microsoft did it with Office. Google did it with search. The studies are never wrong exactly, they're just measuring what's easy to measure while ignoring the harder questions.

What's the cost of workers becoming dependent on a tool that might hallucinate critical information? What happens to the skill development that comes from doing hard cognitive work yourself? How do you measure the productivity loss when someone spends 45 minutes trying to get ChatGPT to write something correctly when they could've just written it themselves in 20? These aren't hypothetical concerns, they're things I've observed in actual newsrooms, including this one.

The company's approach to data and AI document tries to address some of the thornier questions about training data and content ownership. They've launched something called Media Manager for creators and content owners, which is basically an acknowledgment that the original "scrape everything and ask forgiveness later" approach wasn't sustainable. Whether it's enough to satisfy publishers and creators remains unclear, and frankly, I'm skeptical it will.

So what's the actual takeaway here? I think we're in the awkward middle phase of this technology adoption cycle. The hype has cooled slightly from the peak insanity of early 2023, but we haven't reached the productive plateau yet either. Companies are still figuring out what these tools are actually good for versus what they were promised they'd be good for. That gap, between marketing claims and practical reality, is where all the interesting stuff happens.

The kids building these systems, and I say that with genuine affection for the young founders I've met in this space, they're smart and they believe in what they're doing. But they also came of age in an era where software could solve almost any problem with enough compute and clever engineering. AI is different. It's bumping up against the limits of what purely technical solutions can achieve, limits that require understanding human judgment, institutional knowledge, social context, all the messy stuff that doesn't fit neatly into a training dataset.

I've been covering tech since the 90s, and robotics is my third vertical, so maybe I'm just pattern-matching too aggressively here. But the pattern I see is this: transformative technologies take longer to change things than the optimists predict, and they change things more profoundly than the skeptics expect. Both things are true simultaneously. The question isn't whether AI will reshape work, it's whether the people making decisions about AI adoption, the executives, the policymakers, the educators, understand what they're actually dealing with.

Based on what I'm seeing? Some do. Most don't. And that gap is going to matter a lot more than any usage statistics OpenAI can publish.

If you want to argue with me about any of this, my email's on the about page. I promise I'll read it, even if I don't promise I'll agree.

Mark Kowalski · 27 mins ago · 5 min