Lots of people have stories about the one formative event that set their life on its track. Sometimes, this is a class they had in school that convinced them to study a new field. Sometimes it's a book they read, or a movie they saw (which sounds like it couldn't be that important, but Vladimir Putin decided to join the KGB because he saw a spy movie). For a lot of people in tech, that moment was the first time they saw source code for something they'd used and realized: I could make something like this.

Those are important catalysts, but they also change the shape of your future information consumption: way more depth in one place, but, time being finite, a bit less breadth anywhere else. This is just the way things work; being slightly better than mediocre across a wide range of skills is not as interesting as being excellent in the eyes of peers whose judgement you respect. The problems are twofold. First, there's the unhedgeable risk of premature optimization; if you decide at 13 that it's your destiny to be a neurosurgeon, you lower your odds of figuring out at 15 that your true calling was writing pop history. Not because you're locked in to a career track that early, but because you'll radically shift your information diet once you find a good obsession, and you'll tend to reframe other things you learn about in light of the main thing you're focused on. And that points to the second risk: once you've locked on to a particular career, your information diet is going to reflect your conception of that career, and over time the way you do things will drift away from the optimal way those things ought to get done.

There are forces pushing in opposite directions that make this effect invisible. Practice goes a long way towards improving performance, and getting good at something means assembling mental checklists (or literal checklists), habits of mind, processes for breaking out of a loop to reconsider your goals, etc. All well and good! But that means that what you're practicing is whatever process you started with. The better you get at programming-without-LLMs, or doing-journalism-without-promoting-it-on-social-media, or investing-without-using-alternative-data-and-expert-calls, the more you're optimizing a process that has a lower ceiling than it otherwise could.¹

Many of the best performers in different domains are doing different things from what they started with, though often using adjacent skills. The analyst who was best in the world at digging through obscure filings to find dirt cheap microcaps in the 1950s is still at it today, but he's a bit more focused on high-quality large-cap stocks. The guy who got famous writing the Game of Thrones books has been incredibly productive at the set of all activities that a) require the same skillset as writing Game of Thrones, and b) are not completing The Winds of Winter. Tom Wolfe spent a few decades as a journalist before he started writing novels.

The Wolfe example is a tad unfair but in a helpful way: Wolfe wrote very journalistic novels—his research for Bonfire of the Vanities involved lots of time spent in New York's courtrooms and trading floors. And before that, he was an unusually novelistic journalist, who was braver than most when it came to quoting his subjects' inner monologues and the like. So one reason Wolfe didn't get locked into an inflexible career is that he defined his career in a flexible way: he was telling stories, and started out by telling implausible-but-true ones before switching to the plausible-but-made-up variety.

So that's one option: define your career in a much broader way than what you specifically do, and then think about the connection between these two things. That also means defining your career in terms of what other people value: nobody wants "code," but plenty of people want the specific things code does. People often dread interacting with lawyers, but they sleep better knowing that the contract they just signed works the way they think it should. Periodically switching from the game to the metagame is a good way to ensure that you're still playing the right game.

But that's one of those habits of mind that's easy to recommend but hard to implement. Some people have a knack for it, and it's unclear how many people can really develop it. There are more practical steps, though. One of the easiest: post your email address where people can find it. Or, more generally: make sure there's a way for random people to reach you, especially for people who are in the earlier stages of the same career you're in. Because these people are working on similar things to you, but haven't had nearly as much time to develop habits, good or bad. It's also a good habit to move one level higher and lower in abstraction. If you spend all day programming, there's a good chance that learning a little bit of linear algebra, graph theory, or queuing theory can save you a lot of heartache; these are abstract enough that you don't know that you're missing them, but sometimes they're incredibly handy. And it's also helpful to have a better-than-average understanding of how Python turns into specific ones and zeroes (there's no time like the present for this, since programmers get further from the metal with each passing year—there's so much vibe-coders don't need to know, which also means that when they get stuck, they don't know where to start).

And it's also a good exercise, if you're mid-career, to talk to people who are much older, especially about what's changed. A lot of business history is hard to parse until you have a specific understanding of the day-to-day limitations that people used to face, but don't any more. LLMs are okay at this kind of history, and it can be a good exercise to think of the equivalent of your job in some older time period and ask the LLM what people with that job would do all day, what tools they'd have, and to try to figure out what was missing.

All of this is a search for context, by way of a search for randomness. The better you get in any one domain, the more you have implicit mental models that guide the way you think about it. Especially in fields that change fast, these mental models drift away from reality, and suddenly they end up being actively misleading—a great deal of money was made in software by realizing that software was a product, not an add-on, and then by realizing that it didn't have to come in the form of a CD and could be accessed by a browser, and then from realizing that a server did not have to be a specific box in a closet but instead could be an abstraction, at least from the user's perspective. These shifts are hard to spot when you've only been through one cycle, but if you don't develop the ability to spot them, it's hard to thrive through multiple cycles.

In The Diff, we’ve written a bit about the general question of information acquisition in a few different contexts:

• We’ve looked at how company org charts are implicitly designed around the CEO’s communication quirks ($), especially for founder CEOs.

• Companies are also an endless struggle against entropy.

• Why is email a great communications medium ($)?

• And why, specifically, is there a (perceived?) correlation between quick email turnaround time and success ($)?

One reason it’s gotten easier to stay mentally adaptable: the talent surface area has broadened.

Join the discussion!

1  All of these are generalizations, of course. There are journalists who are great at longform, terrible at Twitter, and who have found their way to organizations that let them focus on what they're good at. There are some programming problems where LLMs don't help much, or even get in the way, though in that specific domain they're getting harder to find—to make LLMs actively counterproductive, you probably need to be programming under some unique constraint, or doing something that the LLM makers actively want to discourage. (So, yes, state-sponsored backdoors in widely-used telecom equipment are probably mostly written the old-fashioned way.) There are plenty of investing strategies that don't rely on alt data, but that, too, is a diminishing pool: if you're a long-term oriented buy-and-hold investor, being able to slice up data can give you a unique understanding of a company's unit economics. And if you're a short-term trader, you at least want to know that volatility will be high around particular alt data releases, even if in the context of your business model it doesn't make sense to have a directional view.