You're Probably Using AI Like an MBA (And That's the Problem)

I saw a tweet from Andrej Karpathy last week that's been sitting with me. He said he's never felt this much behind as a programmer. The profession is being dramatically refactored, and he has a sense he could be 10X more powerful if he just properly integrated what's become available over the last year.

This is Andrej Karpathy. Former Director of AI at Tesla. Built autopilot neural networks. Now at OpenAI. One of the most technically sophisticated people in the world.

And he feels behind.

Boris Cherny from Anthropic responded with a similar experience. He was debugging a memory leak in Claude Code the old way when his coworker just asked Claude to do it. Claude shotgunned the solution and put up a PR. His point was that newer engineers who don't make assumptions about what the model can and can't do are able to use it more effectively.

I've been thinking about this through the lens of something completely unrelated: the marshmallow challenge.

If you haven't seen this, it's a team-building exercise where groups compete to build the tallest structure using spaghetti sticks, tape, string, and one marshmallow that has to sit on top. Someone ran this experiment across different groups and found something unexpected. Business school graduates performed terribly. They'd plan extensively, assign roles, architect the perfect structure, and then watch it collapse when they finally put the marshmallow on top at the end.

Kindergartners beat them consistently.

The kids didn't plan. They just started building. Put the marshmallow on immediately. Watched it fall. Built it again. Iterated fast. The MBAs spent their time optimizing for looking smart and ran out of time. The kids optimized for learning.

I think most of us are the MBAs right now with AI tools.

I've been lucky in one respect. I'm naturally inclined to just try stuff. I don't wait to understand something perfectly before I use it. That tendency has kept me on the cutting edge with AI tools, not because I'm smarter about them, but because I'm willing to look dumb while I figure them out.

But even with that advantage, I still catch myself slipping into MBA mode. Reading another article about prompt engineering. Waiting to see what best practices emerge. Hesitating before I just build something and see what breaks.

We're seeing this pattern play out at scale at elvex. We work with enterprises adopting AI, and there's a clear pattern in who accelerates fastest. It's not the people with the most AI experience. It's not the ones who've taken courses or read all the documentation. It's the ones who just start building.

The marketing manager who spins up an agent to analyze customer feedback without asking permission first. The operations lead who automates a workflow before checking if there's a "right way" to do it. The junior employee who hasn't learned yet that they're supposed to be cautious.

These people aren't more technically sophisticated. They're just more willing to be kindergartners. They build, break, learn, rebuild. They don't optimize for looking competent. They optimize for figuring it out.

The gap between them and everyone else widens every week. Not because the tools are getting harder to use, but because the tools are evolving so fast that the planning mindset can't keep up. By the time you've figured out the "right approach," the tool has changed and your knowledge is outdated.

Here's what I think happens next. The companies that win aren't going to be the ones with the most AI expertise on staff. They're going to be the ones that turn everyone into builders. The ones that give their entire organization permission to experiment, break things, and learn fast.

That's the bet we're making with elvex. We're not building a platform for AI experts. We're building a platform that turns every employee into someone who can build with AI. Because the constraint isn't access to the technology anymore. It's the willingness to use it like a kindergartner instead of an MBA.

Stop waiting for the manual. It's not coming. Start building.