HNHacker News
TopNewBestAskShowJobs

tikhonj

16,934 karma · joined September 12, 2011

Tikhon Jelvis

I am interested in programming languages, functional programming (especially Haskell), domain-specific languages, static analysis and program synthesis.

I am working on apply programming language ideas to supply chain planning.

I worked on supply chain optimization at Target for several years, and co-authored a textbook on reinforcement learning based on that experience.

contact: http://jelv.is, tikhon@jelv.is

submissionscomments
tikhonj··on Baumol's Cost Disease
If that's all you need, there are free alternatives that should be more than sufficient today.
tikhonj··on The programmers who live in Flatland
> If it was as great as its advocates say, surely it would have taken over the world by now.

That is a big assumption about the way popularity contests work.

tikhonj··on Using LLMs at Oxide
That's because embarrassingly bad writing is useless, while embarrassingly bad code can still make the computer do (roughly) the right thing and lets you tick off a Jira ticket. So we end up having way more room for awful code than for awful prose.

Reading good code can be a better way to learn about something than reading prose. Writing code like that takes some real skill and insight, just like writing clear explanations.

tikhonj··on Technical Deflation
The overall trend has been the opposite though, hasn't it? People used to buy a new phone (or new laptop/etc) every couple of years because the underlying tech was improving so quickly, but now that the improvements have slowed down, they're holding onto their devices for longer.

There was an article[1] going around about that recently, and I'm sure there are more, but it's also a trend I've seen first-hand. (I don't particularly care for the article's framing, I'm just linking to it to illustrate the underlying data.)

[1]: https://www.cnbc.com/2025/11/23/how-device-hoarding-by-ameri...

tikhonj··on Trillions spent and big software projects are still failing
There's a high switching cost with substantial information asymmetry. Good places are hard to find in some absolute sense and it's hard to evaluate whether a new team is actually going to be good or not. And even if you do find a good team, there's no guarantee it'll last past the next reorg.
tikhonj··on Make product worse, get money
All of the pizza examples are about reducing cost. The argument about dating apps is about increasing retention. The dynamics are qualitatively different.

The argument with pizza is more like "people like salty, fatty food, so pizza places are incentivized to make their pizza less healthy so that people come back more often"... which is exactly what happens!

So why doesn't a legitimately healthy restaurant come along and take the whole market? It's partly because restaurants aren't just in the business of selling (healthy) food: it's also about convenience and satisfaction and experience. More importantly, that just doesn't fit with how people largely make day-to-day decisions.

The same thing happens with dating apps. People get drawn in for all sorts of reasons that don't necessarily map to getting married, even if finding a long-term relationship is explicitly their goal. Tinder competes with Tiktok more than it competes with other dating apps.

The other problem is that making a really effective dating app is just hard. It's fundamentally difficult to help people find compatible partners, especially without in-person contact. That's compounded by cultural and demographic issues. It doesn't matter how well your app is designed when there's a massive imbalance in genders!

tikhonj··on Things I don't like in configuration languages
In the world of human-readable data formats (ie not programming languages), the best one I ever used was Jane Street's sexplib[1] s-expression format.

It was concise and expressive. There was a direct way to describe variants (types with multiple constructors), which is always awkward in JSON, but the format was still surprisingly low-noise for reading and editing by hand. I remember you could even use it as a lightweight markup format:

    (here is some text with (em formatting) information)
(The format leaves the interpretation of things like (em ...) up to you; you could use it as a slightly more verbose Markdown, but you could also use it to structure readable text with other sorts of metadata instead.)

And, unlike certain other formats I won't name, it has comments!

It also helps that Emacs with Paredit makes editing s-expressions flow. The tool doesn't need to know anything about the sexplib format specifically; just relying on basic s-expression structure gives us fluid but simple structural editing.

I am constantly sad that nobody else uses this sort of format, and I have to deal with a mixture of JSON, YAML, TOML and other ad-hoc formats instead.

[1]: https://github.com/janestreet/sexplib

tikhonj··on Collaboration sucks
> Imagine you are driving a car.

Rally races sometimes have two people, a navigator and a driver. The dynamic between them is absolutely fascinating. I recommend watching some videos because, before I saw it, I would not have imagined that driving a car like that would have been possible much less effective.

tikhonj··on Ireland is making basic income for artists program permanent
There is a lot of value in reducing risk, variance and up-front costs for artists. Enough value that, to somebody barely scraping by, it makes the difference between a show/exhibition/etc being doable and not.

It also changes the distribution. I'd say it's a net positive if a bunch of artists get enough money to live on their art rather than the vast majority not making enough and a tiny fraction making the most. It's just a matter of correcting for structural factors that otherwise push towards an exponential distribution of income.

tikhonj··on The illegible nature of software development talent
It's used here as a specific term from Seeing like a State[1][2], which is one of those books that got popular with a specific set of online tech folks. "Legibitility" in that particular sense is a very useful concept with no other convenient word, so I'm not too surprised it caught on.

[1]: https://en.wikipedia.org/wiki/Seeing_Like_a_State

[2]: available free to read online: https://files.libcom.org/files/Seeing%20Like%20a%20State%20-...

tikhonj··on Category Theory Illustrated – Natural Transformations
Category theory gives us a nice, high-level set of conceptual tools to try to understand and generalize over things that are hard to connect otherwise. Some people find that useful directly, other people just enjoy it for its own sake, or even for aesthetic reasons. (I think all three are totally reasonable!)

At the same time, it's actually rather more accessible than most other areas of pure math—at least at the level that people talk about it online. Basic category theory can be hard to learn because it's so abstract but, unlike almost any other are of math from the 20th century onwards, it has almost no hard prerequisites. You can reasonably learn about categories, functors, natural transformations and so on without needing a graduate degree's worth of math courses first. You might not understand the most common examples mathematicians use to illustrate category theory ideas—but it's such a general framework that it isn't hard to find alternate examples from computer science or physics or whatever else you already know. In fact, I expect most of the articles that get talked about here do exactly that: illustrate category theory ideas with CS/programming examples that folks on HN find relevant and accessible.

tikhonj··on Show HN: Autism Simulator
I've also worked on some relatively senior/high-trust teams using roughly scrum-style processes, and the experience was decidedly worse. Even with the best setup and intentions in the world, the process still makes focusing on small, individual tasks the path of least resistance, which leads to less collaborative, more short-term focused work with less flexibility and ownership.

A sufficiently strong team can make even a bad process work out okay, but that just means they're strong enough to compensate for the process, not that the process had any latent merit.

tikhonj··on Show HN: Autism Simulator
Working on somebody else's goals, and being systematically micromanaged are two very different things.
tikhonj··on Show HN: Autism Simulator
My experience has been 100% the opposite. Daily public status check-ins, top-down decisions, every work interaction mediated through artificial structure? The points are made up, the deadlines are obviously fake, but everyone acts as if they are real? Except when they're not?

That, on its own, would make it clear the environment wasn't built for me.

The fact that the environment was very obviously built for management—for information to flow up so that decisions can flow down—but also that nobody is willing to acknowledge that? That just makes it even clearer.

I've worked in an environment that did feel like it was built for me, and it was pretty much the opposite of scrum/agile/etc. I had real trust with a clearly defined area of ownership. I was responsible for managing the interfaces and interaction points around my area and, occasionally, for real deadlines (with real context!), not a slog of fake short-term deadlines that exist just to create pressure. I didn't have to break down or justify my work in terms of bite-sized tasks that could roll up into somebody's spreadsheet.

And the best part? We got more done, faster, than conventionally managed teams.

If the culture hadn't been totally ruined by a reorg, I'd still be there. I'm still sad I haven't been able to find anything similar since. But, having experience that, I am only more confident that scrum et al are absolutely not built for me.

tikhonj··on David Lynch LA House
Sure, but there's a gap between being poor and being wealthy. $15 million is a lot more than most people have and it's certainly enough to be comfortable and not have to worry about money, but it's still categorically different from having the sort of money that gives you institutional power.
tikhonj··on This map is not upside down
For whatever reason, the Berkeley campus maps[1] have East at the top. And, for whatever reason, it always felt natural—probably because it matches the terrain (East is up-hill).

It's still surprising how much that has colored my mental orientation in Berkeley even 15 years later. I now consciously know to correct for it, but still think of campus "oriented" in that direction. For years I had to really think about it to remember that up-hill was not, in fact, North.

[1]: https://www.berkeley.edu/wp-content/uploads/2025/06/campus-m...

[1]: https://www.berkeley.edu/wp-content/uploads/2025/06/campus-m...

tikhonj··on We can’t circumvent the work needed to train our minds
> You have to remember EVERYTHING. Only then you can perform the cognitive tasks necessary to perform meaningful knowledge work.

If humans did not have any facilities for abstraction, sure. But then "knowledge work" would be impossible.

You need to remember some set of concrete facts for knowledge work, sure, but it's just one—necessary but small—component. More important than specific factual knowledge, you need two things: strong conceptual models for whatever you're doing and tacit knowledge.

You need to know some facts to build up strong conceptual models but you don't need to remember them all at once and, once you've built up that strong conceptual understanding, you'll need specifics even less.

Tacit knowledge—which, in knowledge work, manifests as intuition and taste—can only be built up through experience and feedback. Again, you need some specific knowledge to get started but, once you have some real experience, factual knowledge stops being a bottleneck.

Once you've built up a strong foundation, the way you learn and retain facts changes too. Memorization might be a powerful tool to get you started but, once you've made some real progress, it becomes unnecessary if not counterproductive. You can pick bits of info up as you go along and slot them into your existing mental frameworks.

My theory is that the folks who hate memorization are the ones who were able to force their way through the beginner stages of whatever they were doing without dull rote memorization, and then, once there, really do not need it any more. Which would at least partly explain why there are such vehement disagreements about whether memorization is crucial or not.

tikhonj··on 996
Performative hours are, fundamentally, a symptom of a lack of trust. A lack of trust in others but also a lack of trust in yourself.

Not trust others is pretty obvious: leaders push for long hours because they don't trust people to be intrinsically motivated or to work in the most effective way for them. If you assume people are inherently unmotivated and lazy, well, trackable hours and artificial pressure seem like the obvious consequence.

But it's also a sign of not trusting yourself. Being judged on outcomes—never fully under your control—is scary. Being judged on anything fuzzy or arguable—taste, experience, skill, insight—is scary. If you're the sort of person who is content to "grind", the best way to win competitions is to turn them into grinding competitions. You can't be confident that you are more skilled, more intelligent or have better taste than others, but you can always just "grind" that extra hour. For a certain personality, time spent is by far the easiest metric to control.

If you grow up constantly being praised for how "hard" (read "long") you work, constantly out-competing people by doing more rather than better, the inherent value of "hard" work over everything becomes fundamentally ingrained in your personal story. And, unfortunately, our culture tends to put those people into positions of power, so this tendency gets reinforced and propagated.

Taking a step back, doing something good with less effort ought to be more impressive than doing it with more effort. That's what real potential looks like.

More importantly, even if working more hours purely increased your effectiveness and productivity—and we absolutely know that it doesn't—it would still be a weak form of leverage. Maybe you can work 2× the hours, but you can never work 10x. On the other hand, with taste and experience, you can absolutely come up with a 10× better design, or a 10× better understanding of what you're doing, and, unlike long hours, those 10× advantages compound.

If you trusted your own taste and creativity to give you the leverage you need, you wouldn't work ridiculous hours because you'd know it's counterproductive.

But when you don't, long hours are an easy, socially accepted fallback.

tikhonj··on 996
performative hours ≠ excitement

if folks were actually excited and motivated, you wouldn't need forced hours, you'd just trust people to work in the best way for them

tikhonj··on Purposeful animations
Delight sounds similar to what game designers call "juice" and, done well, it really does make a game feel delightful beyond its pure gameplay.

I've had the same feeling with more utilitarian interfaces, but it's pretty rare. I don't know why. I expect it's partly because we have different expectations for programs than we do for games, partly because the context and the interactions are pretty different, and partly because most organizations do not have the will or the ability to make interfaces that satisfying. (After all, it's the worst sort of thing for most organizations: something that requires taste, time and experience and cannot be managed, measured or executed by committee.)

tikhonj··on How to Give a Good Talk
It's "meaningless" because if you broaden out the definition of "value" to "anything that will make people interested in what you're doing", the advice turns into a tautology.
tikhonj··on You don't want to hire "the best engineers"
One way is for senior engineers to give junior folks well-defined areas of responsibility, and then not worry about assigning and tracking small tasks within that. The more senior folks explain and motivate what we need and talk through what the interfaces around the area are and how it fits into the broader work of the team, then help as-needed.

Instead of trying to "track" work in terms of tasks, you keep up with the state of the system by understanding the state of each area. Which is pretty natural to do since you'd be helping with design discussions, code review and some pair programming and debugging.

You can get a lot done by talking and by looking.

tikhonj··on You don't want to hire "the best engineers"
There's a massive difference between "don't write anything down" and "don't use tickets". There are lots of (much better!) ways to write things down and to communicate than tickets.

More importantly, there are categorically better ways of understanding what we're working on than trying to break work down into bite-size linear "tasks".

tikhonj··on You don't want to hire "the best engineers"
One thing I learned from working on a remarkably good team—at Target of all places—is that you can hire great people as long as you can demonstrate (show-not-tell) that you're doing something different and interesting. How you demonstrate this can take lots of different forms (the whole point is that there is no one way to do this!) but just saying "we have high standards and work hard" is never enough on its own. Everybody says that.

You'd think that startups would be the best-positioned organizations to do things differently in the name of hiring. When you start your own company, you don't have to answer to management or HR, and you don't have to follow trends for political reasons. ("We can't let your team do X differently from everybody else, people are going to talk...") But I rarely see this in practice. Most startups seem intent on having a pretty standard approach to management and work; there's clearly some pressure, whether directly from investors or just purely sociological, to be like every other startup.

tikhonj··on You don't want to hire "the best engineers"
Eh, with good leadership who knows how to support different people with different motivations, you really don't. But good leaders are even harder to find and hire than good engineers.

And they probably won't use Jira. Or tickets.

tikhonj··on The cost of transparency: Living with schizoaffective disorder in tech
Slight variations on "This isn't X—it's Y." have been popping up all over the place, almost definitely because it's a pattern that ChatGPT has been tuned to (over-) use.
tikhonj··on Malleable Software
The best software around is Emacs. Does that count as "opinionated" in your view?

In some ways it is—Emacs does a lot of things its own way, completely unbothered by mainstream conventions—but, at the same time, it's also totally malleable in the sense of this article. What makes Emacs great is a consistent and coherent conceptual foundation coupled with remarkably flexible code, letting you adjust Emacs to your needs rather than adjusting your needs to Emacs.

Or maybe the best software around is situated software. Software that's built by and for a specific set of people in a specific social context. Situated software is qualitatively different from product software, and it works so well because, again, it gives its users real agency and control. Instead of trying to create software that knows better than its users, we can create software that supports its users in whatever ways works for me. The result is still opinionated, but it's opinionated in a categorically different way from what you're describing.

So perhaps the best mainstream software is Excel.

And, while I don't think they're there now, it seems like LLMs are likely to be the foundation for the next Excel.

tikhonj··on We put a coding agent in a while loop
There's a classic Peter Naur paper about this from 1985: "Programming as Theory Building"

https://pages.cs.wisc.edu/~remzi/Naur.pdf

tikhonj··on Valve Software handbook for new employees [pdf] (2012)
I mean, you can provide support and structure without imposing it or telling people what to do. People who need the structure can rely on it; people who don't can work their own way.

Top-down, controlling management vs unstructured chaos is a completely false dichotomy.

tikhonj··on The joy of recursion, immutable data, & pure functions: Making mazes with JS
If you enjoyed that, I have a blog post on generating mazes in Haskell (from over a decade ago!). The algorithm is very similar, but the code is written using "inductive graphs" and the post is really more of an intro to working with graphs in a purely functional style.

https://jelv.is/blog/Generating-Mazes-with-Inductive-Graphs

← PreviousPage 7 of 34Next →