What a Mocked CEO Got Right About How to Use AI 

In the late 2000s, Microsoft CEO Steve Ballmer always came off a bit unhinged. Sweaty and loud, he became a meme while ranting about Microsoft’s future. There’s an apocryphal story about how Steve figured out something important, that ended up being a prescient take about an AI-first world. I think about this story once a week; I hope it is equally impactful for you in how you use AI.

The story goes that Steve wanted Microsoft to build a tablet. The iPhone had changed the direction of the tech industry, and Ballmer saw where things were headed: mobile, touch-based computing was the future. But instead of placing one big bet, he did something only a CEO in his position could do. He had the money. He had the talent. So he funded two entirely separate teams to take a shot at creating a tablet. 

This is staggering. Most companies cannot afford to allocate thousands of exorbitantly salaried tech workers to build moonshot hardware products. But do it twice? Most companies couldn’t afford that kind of redundancy. But Microsoft wasn’t most companies.

One division was a secretive group led by J Allard, the creative force behind the Xbox. They were building Courier, a sleek dual-screen device with no keyboard, no Windows, and no trace of Microsoft’s legacy software. It was a complete reimagining of how people could interact with a digital notebook. The other project came from Steven Sinofsky’s Windows division. It was more conventional and built to extend the Windows platform into a tablet format, keep compatibility, and stay within the Microsoft stack.

Both teams worked in parallel. Both knew they were being judged against each other. And both were preparing for the same thing: a final presentation to Ballmer, who would choose which vision would move forward.

Ballmer never spoke publicly about this move so we don’t know his exact reasoning; I assume he did it because it increased his odds of creating a great product. With this strategy, he didn’t have to hope one team nailed it. He’d borrow a navigation pattern from one team, an onboarding flow from the other, and synthesize the best ideas into a stronger final product.

Ballmer was practicing discernment. He wasn’t designing the tablets; his role was to recognize what made the tablet good and select the best options. He focused his energy not on producing, but on choosing. That was the leverage point.

Discerning > Prompting

Which brings me to generative AI. Most people use generative AI like a vending machine. They ask for one thing, take whatever comes out, and move on. But Ballmer wouldn’t do that. He’d want five answers. And so should you.

He was a CEO with resources, power, and a massive R&D budget. But generative AI puts you in a similar role, without needing to be the CEO of Microsoft.

When you ask ChatGPT, Claude, or any other tool to help you solve a problem, don’t just ask for an answer. Ask for three. Or five. Or ten. Tell it, “Give me three versions of this outreach email.” Or “Give me five go-to-market strategies if we were to pivot our product to target midsize law firms.” Or “Give me three strategies for identifying our next round of investors.” Whatever the challenge, generate your options up front. Then compare and contrast the outputs, hunting for the best.

Because your job isn’t to produce anymore. Your job is to choose. Pull the strongest ideas and throw out the weak ones. The AI is your team. You’re the one who picks the winner.

Ballmer didn’t build the tablet. He created the conditions to choose the best one, and his strategy put Microsoft on the path to create the Surface, a product that generates $7 billion every year. You may not be Microsoft’s CEO, but generative AI gives you the ability to operate as if you are.

Discernment is a powerful skill for getting the most out of AI.

Every time you prompt AI for multiple answers and choose the strongest one, you’re sharpening that ability.

So don’t just prompt. Practice discernment.

P.S. This essay is an example of discernment in practice. I prompted GPT-4 and Claude to give me multiple takes, angles, and drafts. If you want to see one of the ChatGPT threads where I practiced discernment to write this essay, see here. I judged like Ballmer. You should too.

If you’re into the intersection of AI, innovation, and actually doing the work—join my newsletter. I share stories, tools, and experiments from the front lines: camhouser.com/newsletter

Thanks to previous readers: Matt Beebe, Rik Van Den Berge