Buffalo Wings for King Thamus

On outsourcing, atrophy, and playing a new game with AI

In the Phaedrus, Plato tells the story of the Egyptian god Theuth, inventor of writing, presenting his creation to King Thamus. Theuth pitches it the way any founder would: this will make your people wiser, a recipe for memory itself. Thamus is unmoved. Writing, he argues, will do the opposite. Men will stop exercising their memories and lean on external marks instead. They will have the appearance of wisdom without the substance, reciting things they no longer truly know.

Twenty-four centuries later I was sitting in a restaurant in Round Rock, Texas, having this exact argument over a plate of buffalo wings.

My friend Mac Bordreau and I worked together years ago, and after I moved to the Austin area we kept meaning to sit down and hash out how life had gone since. Sunday was finally the day. Somewhere between the catching-up and the wing sauce, I brought up AI (I am maniacally obsessed with using it to solve real pain points in my life, and I had seen too many benefits in my own engineering work to stay quiet about it). Mac’s response was Thamus’s, almost word for word with the vocabulary updated: this stuff is going to make us dumber. Especially in creative work, the argument goes, outsourcing the craft means losing the craft.

And here is what struck me as I sat there eating grilled chicken slathered in this genuinely great house-made buffalo sauce: I am a pretty good cook. I was, at that very moment, enjoying the fruits of having outsourced my food preparation to a kitchen full of strangers. My afternoon plans were to hit the grocery store, where I would outsource my hunting and gathering for the week. I had driven to lunch in a Tesla that handled ninety percent of the attention that driving used to demand, and the automobile itself (never mind the self-driving part) has made me a far less prolific walker than my great-grandparents were.

Nobody at that table was worried about any of this. Nobody accused me of forgetting how to cook because I ordered the wings.

Both things are true

I will grant that my analogy doesn’t line up perfectly with the problem, and yet the resonance with the Phaedrus keeps humming in my head. Because I think both things are true at once. If you outsource your creative mind completely to the machine, you will atrophy. That is not a hypothetical; it is how bodies and minds work. If I never walked because the car exists, my health would deteriorate. If I never cooked because restaurants exist, I would eventually forget how to craft my own food. The tool doesn’t decide whether you atrophy. Your relationship with the tool decides.

There’s a running joke in my family about this. Years ago my mother was sitting with me while I agonized between two possible solutions to what was, for me at the time, a genuinely hard problem. She was “listening” the way you listen to a topic you have no knowledge of and no stake in, which is to say she was doing her crossword and nodding. When I finally finished presenting my two options, she looked up and delivered her verdict: “Well, there are positives and negatives to both sides.”

She meant it as a joke, a way of gracefully exiting a conversation she’d never entered. I was young enough to receive it as revelation. And honestly, decades later, it has become a pillar of how I measure the world. Everything is a trade-off. The problems worth having opinions about never live in a binary, black-and-white world; they live in gradient shades of gray. Writing cost us some feats of memory (nobody recites the Iliad from heart at parties anymore) and bought us libraries, science, and the accumulated compounding of every mind that ever wrote something down. Thamus was right about the cost. He just couldn’t see the ledger.

The question I’d rather you ask

So my question to you is not a call to arms. I’m not asking you to race into the future bolting AI onto everything you touch. I’m asking you to sit with a different question than the one everyone is arguing about.

Not “will this make me dumber?” but: what is the buffalo-wing-shaped part of your work? What are the tasks where, like me at that restaurant, you have no ego investment in doing the slaughtering and sauce-making yourself, because what you actually want is the meal? And then the follow-up, which is the one that matters: if AI handed you back four extra hours every day, what would you do with them?

If that question interests you at all, here is the reframe I’d offer. Stop cataloging what AI gets wrong. Get curious instead about what AI is, and what you could give it to produce better results. The people getting the most out of these tools are not the ones with the longest list of its failures. They are the ones playing a different game.

What the machine actually is

Strip away the marketing and AI is a prediction engine operating over vectors. Picture the model as a giant circle containing the knowledge, and artifacts of knowledge, that humanity has produced since the dawn of the digital age. Now draw an inner circle inside it: that’s the slice bounded by the context and prompt you provide. The “AI” is, at its simplest, a prediction of which tokens come next inside that inner circle.

Hold that picture and a lot of frustrating behavior becomes legible. When you get a sub-par answer, it might be the model you chose (and models get better every single day). But it might also be that your prompt and context carved out the wrong slice of the Venn diagram, and the prediction was bounded against the wrong region of everything the model knows.

The picture also reveals the real limitation, the one that matters more than hallucination discourse: the machine is bound to its training data plus whatever you hand it. It cannot sit with the absence of something. It doesn’t know what it wasn’t given. The novel idea, the constraint nobody wrote down, the thing that makes your problem your problem rather than a generic one, all of that has to come from you, supplied as context, so the prediction gets bounded against something true.

The tacit knowledge trap

Here is the pattern I keep seeing, and if you’re a senior or staff-plus engineer, this may sting a little. The people who struggle most to get useful outcomes from AI tend to be the people who are best at tacit knowledge work. Many of the strongest engineers I know feel their way through a problem. They can’t fully articulate how they’d solve it until they’ve solved it. Their expertise lives in their hands, and it has served them so well for so long that they’ve never needed to externalize it.

That skill profile is almost perfectly wrong for this tool. The machine can’t read your hands.

The fastest single upgrade I know of is embarrassingly low-tech: practice writing down (or speaking aloud to a voice-to-text tool, if writing makes you itch) your understanding of the problem and the outcome you actually want. Not the implementation. The problem and the desired outcome. You can hand the machine exacting implementation instructions, sure. But the mode I’ve drifted to almost entirely is giving it patterns and examples rather than solutions. I treat it the way I’d treat a promising junior engineer I’m hoping will learn from what I give them, not a ticket puncher I’m asking to do the work so I don’t have to.

Don’t say “kill the chicken”

Which brings me back to the wings, because the kitchen metaphor goes further than I first realized.

Say “kill the chicken” to a system with all the world’s knowledge and none of your context, and it might clobber the bird with a three-ton stone, rendering no usable meat. Technically compliant. Tell it to cut the chicken into quarters and you may get four random pieces of equal weight, because you never mentioned that we split chickens at the joints, and joints are exactly the kind of thing an expert stops noticing they know. And nobody who cares about the result asks for “buffalo sauce” cold. You’d share the recipes you’ve tried, the version your friend makes that you’d steal if you could, which parts of each you loved and why. Those pieces of context will render you a dramatically better wing than “quarter the chicken, cook it, put buffalo sauce on it” ever will.

Every one of those failures is the same failure: an expert assuming that what’s obvious to them is in the inner circle, when it was never handed over at all.

If you have raised children, you already know this shape. A child’s mind is taking in a world that is vast and overwhelming, and if you don’t help bound it properly, you will often get very odd results. Not because the child is broken. Because the framing was.

A new game

Thamus looked at writing and saw only what it would take. He wasn’t wrong about the cost; he was wrong about the game. And there’s an irony he never got to appreciate: the only reason we know his objection at all is that Plato wrote it down.

So here is the whole ask. When you think about what’s going right or wrong in your use of AI, consider whether you can play a new game with the system. Instead of gripping every step (kill the chicken, cut the chicken, cook the chicken, make the sauce), try treating it as a partner with access to a world’s worth of knowledge, one that can make your favorite dish if you give a great explanation of the outcome you desire. Keep cooking sometimes, because the craft is worth keeping and atrophy is real. Keep walking, for the same reason. But stop auditing the tool for failures and start auditing what you’re feeding it.

There are positives and negatives to both sides. My mother, doing her crossword, was righter than she knew.