Let's Do Something, Even If It's Wrong

Doug was from Atlanta. He had the easy confidence of someone who'd never spent much time worrying about what other people thought, and when a group of us stood around in college unable to agree on how to spend an evening, he'd cut through it the same way every time: "Let's do something, even if it's wrong."

He died in a car accident in Gainesville, Florida in 1986. His best friend from childhood, Todd, was with him. Todd survived. Doug did not. I have been repeating that saying ever since, in his memory, in situations that keep finding it relevant. It is not a productivity philosophy. It is not the rallying cry of a type-A personality who cannot tolerate stillness. It is something more specific: a recognition that the paralysis of too many options is its own kind of choice, and usually not the good kind.

The conversation about AI automation has been standing in that same paralysis for years now.

The Value Was Never in the Difficulty

Here is the fork in the road that most of that conversation ignores. When AI makes something easier to do — writing code, drafting contracts, generating analysis — it does not make that thing less valuable. It makes the labor less valuable, which is not the same thing at all.

I worked at Timeplex, a division of Unisys, in an era when a dedicated T1 line cost roughly thirty thousand dollars a month. Gigabit ethernet now costs practically nothing. Nobody argues that network connectivity became less valuable when it became cheap. The value was always in what the connectivity enabled, not in the difficulty of provisioning it. The difficulty was just friction — real, expensive friction that employed a lot of people to manage it, but friction nonetheless.

The same logic applies to code, prose, and legal documents. The output is what was always valuable. The hard part was just the price of admission. AI is lowering that price, which means the admission is now available to more people for more purposes — which looks, from here, like an increase in value rather than a decrease.

The Jobs Nobody Will Miss

The second fork is thornier, and the one that doesn't get enough honest attention.

A significant portion of the workforce is doing work they actively dislike, for organizations that treat them badly, producing output that adds little to anyone's life including their own. Filling out forms. Sitting in meetings that exist to schedule other meetings. Working alongside sociopaths who have learned to game performance reviews. Getting paid, in the most literal sense, to complain — about colleagues, about management, about the gap between what is expected and what is humanly possible.

The anxiety about AI displacing this work carries an assumption worth examining: that the displaced work was good work worth preserving. Some of it was. Much of it was soul-crushing drudgery that people endured because it paid and because they had not yet found a way out. If AI eliminates the drudgery and replaces it with nothing, that is a policy problem, not a technology problem. The technology, in that framing, did these people a favor.

The get-rich-quick corner of the internet has its own answer to the freed-time question, and it is not an inspiring one: arbitrage. Find the spread between what a thing costs in one market and what it fetches in another, extract the difference, repeat. This is presented as entrepreneurship. It is mostly rent-seeking — the identification and exploitation of structural inefficiencies that exist because someone built a cartel around them, not because they represent genuine value creation. AI will commoditize this too, once it learns to navigate the institutional impediments that keep those spreads alive. The people celebrating arbitrage as liberation are describing a waiting room, not a destination.

The Two Things That Are Actually Yours

Here is what remains after you strip away the labor that was merely friction, the work that was merely drudgery, and the arbitrage that was merely extraction.

Learning. And creating.

These are the two activities where the doing is not a means to an end but the end itself. Not instrumentally valuable — not valuable because they produce income which you then spend on the things that make life worth living. Directly valuable. The loop closes without money as the intermediary.

You cannot outsource learning a language to a machine. I can ask Claude to translate a menu in Seville, and it will produce something more fluent than most native speakers could manage, in seconds. What it cannot do is acquire Spanish on my behalf. The acquisition happens in my brain, through exposure, through the patient accumulation of comprehensible input, through the particular frustration and small triumph of finally understanding something you have heard a hundred times. There is no shortcut. The machine is already fluent. I still have to do the work, and the work is the point.

Creating is the same. You can use AI to draft, to iterate, to find the word that was hovering just out of reach. What you cannot delegate is the judgment — knowing when it is right, knowing what you actually want to say, knowing when the paragraph has gone slack. Those are yours. They require the accumulated weight of everything you have read and experienced and thought about, and no model has your particular version of that.

Doug's saying was never really about evening plans. It was about the only honest response to the question of what to do with a life: pick something, and go. Not because any particular choice is guaranteed to be correct, but because the alternative — standing around waiting for perfect clarity — is a choice you made without admitting you made it.

The AI is running. The tokens are spending. The question, as it always was, is what you do with the rest of it.