The Missing Step Between AI Hype and Real Profit

Back in February, I grabbed a flyer at an anti-AI march in London. I’m not sure if the folks at Pause AI intentionally riffed on South Park’s underpants gnomes, but they nailed the vibe anyway. It read: “Step 1: Grow a digital super mind. Step 2: ? Step 3: ?”

For anyone who missed the 1998 episode, the gnomes steal underpants at night with a three-phase business plan: collect underpants, question mark, profit. It’s become a classic meme format for mocking anything that skips over the hard part. Elon Musk even used it to explain his Mars funding plan once.

Right now, that’s exactly where AI is. Companies have built the tech—Step 1. They’re promising salvation, transformation, all that—Step 3. But Step 2, the part where you actually figure out how to get from here to there, is still a giant question mark.

Pause AI wants regulation to fill that gap. Fair enough, but what regulation and who enforces it? Nobody agrees. Meanwhile, the boosters are convinced Step 3 is some kind of utopia and mostly ignore the messy middle. OpenAI’s chief scientist Jakub Pachocki told me recently we’re racing toward an “economically transformative technology.” He knows where he wants to go, sort of. It’s hazy and still far off. And everyone’s taking a different route.

I don’t think they all make it.

Look at the studies. Anthropic put out a paper predicting which jobs LLMs would hit hardest: managers, architects, media folks should brace themselves; groundskeepers, construction workers, hospitality, not so much. But those predictions are basically educated guesses based on what LLMs seem good at in a lab, not what they actually do in a real workplace.

Then there’s the study from Mercor, an AI hiring startup, that tested AI agents from OpenAI, Anthropic, and Google DeepMind on 480 tasks bankers, consultants, and lawyers do regularly. Every single agent failed most of the work.

Why the disconnect? First, who’s making the claim matters. Anthropic has skin in the game. Second, most of the people telling us something big is about to happen are basing that on how fast AI coding tools are improving. But not every job is coding. Other research shows LLMs are terrible at strategic judgment calls.

And even when you deploy these tools, they don’t land in a clean room. They land in messy offices with existing workflows, cranky humans, and legacy systems. Sometimes adding AI makes things worse. Sure, maybe you need to tear up those workflows and rebuild them around the tech, but that takes time and guts. Most companies don’t have either.

That big hole where Step 2 should be creates an information vacuum. Without any real agreement on what’s coming or how, every wild claim of the week fills the void. One social media post can shake markets. We’re that unmoored.

We need fewer guesses and more evidence. That means transparency from model makers, coordination between researchers and businesses, and new evaluation methods that tell us what actually happens when AI rolls into the real world.

The entire tech industry—and a lot of the global economy—is betting that AI really will be transformative. But that’s not a sure thing yet. Next time someone starts talking about the coming revolution, remember that most businesses are still standing there holding a pile of underpants, wondering what the hell Step 2 is.