Only What You Love Lasts

Some people say doing what you love is a luxury—in the survival phase, in the competition phase, what choice do you have? I used to think so too. Until I recently reread Hume and ran the idea through a programmer’s lens, and realized this isn’t a romantic question at all but a cost question: if you want to keep doing something over the long haul, only love can sustain it. This piece argues that one thing.

Hume’s Theory

In Book II of A Treatise of Human Nature (1739), David Hume (1711–1776) wrote a line that has been quoted endlessly and ignored just as endlessly:

Reason is, and ought only to be, the slave of the passions.

Here “passions” does not mean what English-speaking ears might first hear as “burning passion”; it’s closer to what we’d call today emotion, desire, motivation—everything that makes you want, want to avoid, want to draw near, want to leave. The core argument he’s making has three layers:

1. Reason on its own cannot drive action. Reason can tell you that A entails B and B entails C, but it cannot tell you that you “ought to” want C.
2. The real driving force comes from emotion. The reason you take some action is that you want some thing, and the “wanting” itself is not produced by reason.
3. Reason only helps emotion find a path. Its job is—once a goal is given—to find the means to reach it.

What Hume is doing here is drawing a very clean distinction: reason can only judge whether a means is effective; it cannot judge whether a goal is one you “ought” to have. In one line—what actually drives behavior is always emotion; reason carries no power of its own.

In the Language of Computers

I come from software, and every time I reread this passage I can’t help translating it into more familiar terms. The closest analogy goes like this:

1. Emotion is the machine instruction set, the thing the CPU can actually execute. You’re hungry and want to eat, tired and want to sleep, you see someone you like and want to draw near—these aren’t “decided,” they’re triggered directly by the underlying hardware.
2. Reason is a high-level language. The code you write in C++, Python, or TypeScript is not executed directly by the CPU; it must first be compiled into machine instructions before it can run. A chain of rational reasoning works the same way—it ultimately has to bottom out in “what you care about” before it can run at all.

There’s an especially important corollary here: any rational argument, to actually make you act, must at some step connect to an emotion you already have. If it doesn’t connect, even the most airtight logic is just words in a PDF.

“You should go to bed early; it’s good for your health.”—a perfect piece of high-level-language code. The CPU doesn’t execute it.

“There’s someone I want to see tomorrow, and showing up to a first meeting with dark circles under my eyes would be too humiliating, and I hate being humiliated.”—a piece of code with an emotional anchor. The CPU executes it.

The difference isn’t in the semantics; it’s in whether it can compile down to the bottom layer.

Spell out this relationship and you have the entire argument of this piece:

• Driving a behavior with reason means running a long chain that “compiles down to emotion”: “I should memorize vocabulary” → for the exam → to get into a good university → for a better life → some happiness I want. The chain only runs once it bottoms out in emotion, but it is long.
• Driving a behavior with love is itself the instruction: the act connects directly to something you care about right now—no layer-by-layer compilation needed. A short path.

It comes down to cost. Running a long chain has overhead every single time. Do something once and it doesn’t matter—the overhead is amortized; but the moment something has to be done long-term and at high frequency, you have to rerun the whole chain every single time, overhead multiplied by count, until sooner or later you can’t sustain it.

This is one of the iron rules every programmer knows best: a hot loop can’t be written in Python. A piece of cold code that runs once in a lifetime—write it in Python, however comfortable you like; but an inner loop that has to run millions of times per second is a disaster in Python, and must be written close to the hardware, in the lowest-level instructions.

Life is the same: anything you want to keep doing long-term is, in essence, a “hot loop.” A hot loop can’t run on reason (the long chain), just as a hot loop can’t be written in Python—it can only run on love, driven directly by emotion. That is why only what you love lasts.

Case Studies

Run this model in reverse and it explains a lot of stories where “willpower ultimately fails.”

The Job That “Looks Right”

Many people pick a job they don’t actually love but that “looks right”—stable, respectable, decent pay—and then get through each day on “I should.” From a software view, this is running an uncompiled long chain on the hot path: reason can issue the instruction, but the CPU executes it painfully slowly, burning enormous “compute” every single day just to perform an action that should have happened naturally. Ten years, twenty thousand hours, every hour high-friction. What finally crushes a person is usually not the difficulty of the work itself, but the act of “forcing it”—willpower depletion, emotional numbness, essentially the CPU endlessly software-emulating an instruction the hardware doesn’t support.

Conversely, people who genuinely love their craft can go deep in the same direction for ten or twenty years and still enjoy it. Buffett, in his nineties, still says he “tap-dances to work”—not because his willpower is superhuman, but because his path is driven directly by love and needs no forcing.

The January Gym Membership

Every January gyms are packed; by March they’re half empty. The reason isn’t complicated: most people start on “I should exercise, for my health”—a textbook long chain (exercise → health → fewer illnesses down the road → …). The chain is long, you have to talk yourself into it afresh every time, and it snaps after a few runs.

The people who stay active long-term almost never rely on willpower; they’ve found something they genuinely enjoy: a ball game, hiking, dancing, cycling. The enjoyment itself is the direct instruction—no need to recompile “I should” every time. Same activity, but one runs on the long chain and the other on the hardware—and only the latter lasts.

The Writing and English You Can’t Keep Up

“I should blog,” “I should memorize vocabulary”—how many people set this resolution and then stop three days later? Same problem: driving on “should” is running a long chain, at high frequency, and it can’t be sustained.

The people who post daily for years, or take a language to near-native level, are almost never “persevering”—they’re genuinely hooked: to keep up with the latest in a field, to watch a show without subtitles, or for the sheer pleasure of expressing themselves. When the drive comes from love, the word “perseverance” doesn’t even apply—because it costs no effort.

When You Can’t Avoid It

What I’ve described so far is “when you can choose, choose what you love.” But in reality there are always things you can’t get out of—things that objective constraints force you to do long-term anyway. Then what?

The model already gives the answer: a hot loop won’t run because it’s running the long “I should” chain. So don’t let it run the long chain—graft a direct emotional anchor onto the thing by hand, and rewrite the long chain into a short path. Satisfaction and a sense of achievement are themselves emotions—instructions the CPU can execute directly. A few common ways to wire it up:

• The stamp-collector’s satisfaction of a streak. A running record you don’t want to break—turn the process itself into a “don’t break the chain” collection, with an instant small reward each time you complete it.
• The achievement of beating your past self. Give progress a quantifiable, certain scale (50 more words memorized, 10 seconds faster pace), so that “I’m better than yesterday” itself becomes the kick.
• Gamification, ritual, company. Progress bars, ranks, people checking in alongside you—at bottom, all of these solder a temporary line straight to emotion beside that tedious hot loop.

This isn’t quite the same as genuine love: a grafted emotion is often weaker and less durable. But it does the right thing—it moves the drive off the long chain and onto a short path, so the daily cost drops sharply. If you truly can’t do only what you love, then at least find a way to make the unavoidable thing connect directly to emotion somewhere.

Closing

So the line “do what you love” can be pried loose from romanticism: it’s not feel-good chicken soup, it’s a prerequisite for the long game. Anything you want to keep doing over the long haul—work, exercise, writing, a relationship—reason can get you started, but only love can carry you to the finish.

This also gives a very practical test: before committing to something long-term, don’t only ask “is it right? is it worth it?”—ask one more thing: “Can I come to love the process of doing it?” If you can’t, either don’t take it on as a long-term task, or at least graft a direct emotional anchor onto it—otherwise you’re running, in Python, a hot loop that’s slated to run for ten years.

Reason is useful, of course—it helps you pick the direction and judge what’s worth doing. But once you’ve decided to do something long-term, the driving has to be handed to love: reason does the choosing, love does the lasting.

Your life is a program that will run for many years. The hot loop has to hug the hardware—only what you love lasts.

References

• David Hume. A Treatise of Human Nature. John Noon, London, 1739.
• David Hume. An Enquiry Concerning Human Understanding. A. Millar, London, 1748.
• Ulrich Drepper. What Every Programmer Should Know About Memory. Red Hat, Inc., 2007. akkadia.org/drepper/cpumemory.pdf