Question: How did the turtle win the beauty contest?

Answer: It had a shell of a body.

AI has a promise problem. The story we’re being told about it, mostly by people who benefit from believing their own hype, isn’t lining up with the evidence. Companies are acting as if it already has, and that gap between belief and reality is where anyone who depends on a paycheck needs to pay attention.

Now more than any time in the past, companies are looking to do more with less. The recent wave of “restructuring” shows us they think they've found a way to do that. X/Twitter and Facebook/Meta set an example by cutting headcount while keeping their services running and watching revenue climb. That was the starting gun and other companies have joined the race. The vision is leaner companies and more output, with AI promising to be a silent workforce.

The key word here is promise. Having vision in the business world means convincing yourself you can see the future. The best storytellers then convince others that investing in that future is the path forward. AI has become the perfect vessel for vision and the perfect narrative for the story.

The evidence this future will materialize is scant. Mike Judge (the data guy, not the animator) has a data driven takedown of the AI productivity myth. He points out that if AI were really boosting software creation by the 3–10x some are claiming, we’d expect to see unmistakable signals. There would be a surge in new apps, domain registrations, GitHub repositories and new games. Instead, the numbers look more or less like they always have: increasing slowly, steadily, and predictably. There is no evidence of a productivity explosion.

Video games make the clearest case to me. Making a game is a multi-disciplinary creative undertaking. It combines art, audio, design, engineering, and a million tiny decisions into something people might love for forty hours or abandon after ten minutes. Engineering alone is expensive and slow, and I say that after fifteen years as a game developer at multiple companies. If AI were dramatically accelerating development by meaningfully improving code, art, and audio iteration speeds we’d be drowning in new games by. Designers would be prototyping playable worlds without engineers. Artists would be releasing solo projects without studios. Everyone with an idea and a weekend would be shipping something.

That’s not what the data shows. Let's look at releases on Steam, a popular online video game store:

Steam releases are up. November 2022 saw 1,129 new games and October 2025 saw 1,846. But it's the same gentle upward slope we’ve seen for a decade. It’s not the sudden spike you’d expect if production had just gotten 5x cheaper and faster. Capitalism is very good at rewarding people for being efficient. If anyone had cracked the code on making games at superhuman speed, we’d see it in the data.

This isn’t proof that AI is useless. AI is helpful every day in small, tangible ways. People use it to draft text, automate tasks with scripts, and getting past the blank-page moments. I’m using it to help edit this newsletter. Those small improvements are real, but they don't add up to the large scale transformation being promised and advertised. Maybe the bottlenecks in creativity aren’t the parts AI solves. Maybe the hard parts of making things remain just as hard as before. Things like the human judgment involved in business strategy, marketing, and final polish. Maybe adoption is slower than we thought. Or maybe AI doesn’t deliver the magnitude of productivity boost that people this it does.

In the end, the explanation matters less than the effect. Companies aren’t waiting for the evidence. They’re reorganizing with the expectation that it will happen. This puts workers in the position of preparing for a world that doesn’t exist yet, because their employers insist that it soon will.

So what do we do? How do you stay solvent when the decision makers are sure the future will be automated? These aren't new questions. Computers have been good at crunching numbers and generating repeatable reports for decades. They are still bad at reading human handwriting. That's why there are hundreds of thousands open jobs for manual medical information entry.

I'm thinking of preparing for this future like a self defense class. The purpose of self defense isn’t to turn you into a superhero, it’s to teach you how to get through a situation you didn’t choose. The current climate with AI requires a similar mindset. You don’t need to become an AI expert or compete with the machine at its strengths. You need to understand what parts of your work are irreplaceably human.

One way to explore that understanding is tabletop gaming. One of the topics of this newsletter is games, after all. Last year, I ran a Dungeons and Dragons game with an AI assistant. Feeding it prompts while playing got me room descriptions, dialogue, monster reactions and more. It was useful for stringing together a couple of coherent sentences describing something. But it also dropped context constantly, missed narrative cues, and struggled to integrate events that happened five minutes earlier.

The parts of running a game AI fails at were obvious. Keeping continuity, shaping the story on the fly, reading the room, managing pacing, noticing when someone is uncomfortable, knowing when to cut a combat short or let a roleplay moment stretch. None of this is execution. It’s judgment. It’s social sense. It’s taste. It’s the ability to understand not just what is happening, but what should happen so that everyone is having fun in the moment, based on tiny signals no machine can yet interpret.

AI didn’t introduce me to the friends at the table. It didn’t build those relationships or make game night something to look forward to. Those connections remain solely human.

The qualities of continuity, curation, and connection show up differently in every workplace. They’re the parts of work that resist automation not because they’re complicated, but because they require being a person working alongside other people to make something that people are going to enjoy. They require presence, memory, intuition, taste, patience, and the ability to sift meaning out of ambiguous human behaviors. If you understand which parts of your job rely on those skills, you know where to double down.

The good news is that this preparation will help you no matter how AI evolves. If the promise turns out to be unfulfilled, you’ll have sharpened the skills that already matter. And if AI really does become a force multiplier you’ll be the kind of person who knows how to do the things it can't do rather than someone waiting to be replaced by it.

Either way, the work worth doing is still the work only you can do.

For more on the unfulfilled promise of AI check out the links below

The wizard Thandros the Cautious sought to create a guardian that could endure centuries without rest. He blended the iron sinew and petrifying breath of a gorgon with the impervious shell and patience of a colossal tortoise. From alchemical fumes and runes stitched in molten basalt, the Gorgoise lumbered forth: a living fortress, its shell veined with stone, its eyes black as midnight.

As always, you can find more on the Gorgoise for free over on Patreon.

Some links on this site are affiliate links that may give me a kickback if you buy something. There’s never an extra charge to you.