OpenAI's New 'Academy' Is Teaching Enterprises to Build Their Own AI Agents

Then there's "workspace agents," which is where things get interesting and, honestly, a little concerning. These are autonomous-ish systems that can connect to external tools (your CRM, your project management software, your calendar) and execute multi-step workflows without constant human oversight. The workspace agents guide talks about "scaling" these agents across teams, which sounds great until you think about what happens when one of them goes sideways at 2 AM on a Friday.

The rest of the Academy covers more familiar territory:

• Projects for organizing ongoing work with persistent context
• Brainstorming tools that help you structure messy ideas into something actionable
• Role-specific guides for operations teams, customer success folks, and presumably more to come

The operations materials are particularly detailed, lots of stuff about standardizing processes and driving "faster execution." The customer success content focuses on reducing churn and improving communication, which is enterprise software 101.

What this actually means

Look, I'm not going to pretend I know exactly where this is headed. It's too early to say whether these workspace agents will actually work at scale or whether they'll turn into the same kind of half-broken automation that made everyone hate enterprise software in the first place. But a few things seem clear.

OpenAI is betting that the future isn't about individual users playing with a chatbot, it's about organizations embedding AI into their core workflows. That's a fundamentally different business, one that requires implementation consultants, enterprise sales teams, and the kind of boring infrastructure work that doesn't make for good demo videos. The Academy materials are basically a self-serve version of what other enterprise AI companies charge consulting fees for.

The other thing that strikes me is how much this resembles the early days of the cloud computing transition. Back then (and I covered this, so I remember), companies had to be taught that "the cloud" wasn't just a place to store files, it was a different way of building software entirely. OpenAI seems to be making the same argument about AI: it's not a tool you use, it's infrastructure you build on.

Whether that argument lands depends on factors we can't really assess yet. How reliable are these agents in practice? What happens when they make mistakes? How do you audit an AI system that's making decisions across your entire customer success pipeline? The Academy materials don't really address these questions, which is either because OpenAI doesn't have good answers yet or because they're saving that for the paid enterprise tier. Probably both!

I reached out to OpenAI for comment on their enterprise adoption numbers and didn't hear back. The company hasn't disclosed exact figures on how many organizations are using these more advanced features, which makes it hard to know whether this is a real trend or just marketing.

The skeptic's take

Here's what I keep coming back to: every major tech transition I've covered, and there have been several, call me old, goes through a phase where the technology companies try to turn users into builders. Sometimes it works (think Salesforce's AppExchange), sometimes it doesn't (remember when everyone was supposed to build their own apps with no-code tools?). The difference usually comes down to whether the underlying technology is reliable enough to bet your business processes on.

We don't know if AI agents are there yet. I've used ChatGPT plenty, and it's impressive, but it also hallucinates, loses context, and occasionally produces output that's confidently wrong in ways that would be embarrassing if a human did it. Scaling that across an organization's customer success team seems... ambitious.

But what do I know. The young founders building this stuff are smarter than me, and they've certainly got more energy. Maybe workspace agents will work beautifully and I'll be writing about how wrong I was in a year. I've been wrong before.

What I do know is that OpenAI is making a very clear bet with this Academy launch: they think the enterprise market is where the real money is, and they think the way to win it is by teaching companies to build their own AI infrastructure rather than just using off-the-shelf chatbots. That's a big swing, and it's going to be interesting to watch whether it connects.

If you want to argue about this, my email's on the about page.

James Chen · 1 hour ago · 6 min