New Report Shows Real AI Use Cases

DEEP DIVE
New Report Shows Real AI Use Cases

OpenAI dropped a new report last week, and it’s one of the rare ones that’s actually worth reading.

It’s real usage data, which makes it far more interesting, but also a little harder to interpret cleanly.

Before getting into the numbers, a quick reality check.

A lot of this data is self-reported, some of it is difficult to verify from the outside, and OpenAI obviously has a stake in how these results come across.

With that in mind, here are the stats that stood out to me and how I’m thinking about them.

20% of enterprise messages now run through Custom GPTs or Projects

Also weekly usage of these features is up 19x year over year.

That feels very believable to me, mostly because it mirrors how I use ChatGPT.

Any work I actually care about lives inside a Project rather than in random, disconnected chats.

Once you have persistent context — files, instructions, previous decisions, half-baked ideas you haven’t deleted yet — the tool starts to feel less like a chatbot and more like a workspace.

I wrote last week about why long-term memory might be one of the most valuable features an LLM can have, and this data backs that up.

After you’ve invested time setting up context, switching to another model suddenly feels like work, even if that model looks better in a demo.

That friction makes people stick around, and honestly, it probably should.

If OpenAI wants to lean harder into anything, Projects feel like the obvious place to do it.

75% of workers say AI lets them do tasks they previously couldn’t

This headline sounds dramatic, but the details underneath it are where things get more grounded.

The report shows that coding-related messages outside of engineering, IT, and research roles increased by 36% in 6 months. That’s a much more useful signal.

It means someone in marketing can tweak CSS without opening a ticket.

An intern can write a custom Excel formula without apologizing five times before asking for help. These aren’t heroic feats, but they remove a lot of small, annoying bottlenecks.

Most white-collar work already happens on a computer, and a surprising amount of it involves logic, structure, or syntax, even if nobody calls it “coding.”

AI doesn’t turn people into engineers, but it does make technical-adjacent work feel less intimidating.

This doesn’t reduce the need for IT teams, but it does explain why AI feels immediately useful to so many non-technical roles.

The caveat, of course, is that this is self-reported. I’m sure there are plenty of stories where someone confidently “fixed” something and quietly created more work for the people who actually know what they’re doing.

Enterprise users report saving 40–60 minutes per day

This is the stat that gets quoted everywhere, and it’s also the one I’m most skeptical about.

Once again, these numbers come from self-reporting, which introduces a very human bias.

If you spend time learning and using a new tool, you naturally want to believe it’s paying you back in time saved.

The more you use AI, the easier it becomes to feel productive, even if the actual clock doesn’t fully agree.

OpenAI includes a chart showing that heavier users report higher productivity gains.

What this chart clearly shows is that people who use AI more tend to feel more productive.

What it doesn’t show is whether they’re genuinely saving an hour a day or just redistributing effort in a way that feels better.

From OpenAI’s point of view, that distinction almost doesn’t matter.

Usage is easy to measure and optimize for. Productivity is messy, subjective, and hard to audit, so it’s no surprise which axis gets more attention.

That doesn’t invalidate the data, but it does mean it’s better read as directional rather than precise.