Fascinating to see how heavily Ruby and Python feature. The things that surprised me by their absence and near absence were languages I love: JS and C#
That said, this is taking now well established startups that will have begun their lives 5+ years ago in general. If I was picking a back end stack then I'd probably hesitate before picking the .NET Framework. But these days I'd easily pick .NET Core.
Likewise, I'd want a statically typed back end and 5 years ago I'd probably have hesitated before using TypeScript and node together. Now I do it regularly.
Would be super interesting to see the same chart in 5 years with companies starting now.
I think this is because, in the 2000s, when a lot of these companies were founded, Ruby and Python hit an pretty much empty sweetspot for web application development in between quickly thrown together scripts and the super heavy enterprisey frameworks.
But since then, that space has gotten a lot more crowded from all sides. There is now a pretty big selection of mature technology to pick from depending on your needs.
In the case of Python, it got also a massive boost from its popularity in datascience. Couple of years ago I thought Python web development is dying out slowly, but seems to have swung back quite a bit.
I think most frameworks/languages swing in and out of the lime light. Using Ruby as an example, it was great then got really hot, then as you said Python got the data science boost and moved a lot of people there. I'm curious if Matz makes good on his promise to 3x Ruby that will result in Ruby getting some more data science libraries.
That said, last I checked regarding performance Rails and Django were pretty much neck and neck with a negligible difference with rails being very slightly faster (iirc).
I think also a big reason why Ruby is feature so heavily in this list is because of it's focus of developer productivity. It's a nice mesh of flexibility and convention. Generally all projects are setup and organized the same but you can still write a conditional if statement a dozen ways. Compared to my personal experience working with Django and Python are the woes of getting everything configured and setup for each project which is frankly a pita. Also, Python is fairly regimented on a lot of flexibility which has its perks too.
I'd caution people outside hyper-growth startups from drawing that much from this chart. People in that ecosystem have very different utility functions than most software engineers.
Comparatively, I'm sure a chart with database swapped out for language would show a huge over-representation of NoSQL products like Mongo and Couchbase. Yet for the overwhelming majority of projects the right answer is just use a solid SQL product like Postgres.
Again, for SV hyper-growth startups the calculus is different. Mongo has tons of downsides and pitfalls. But the one thing it has going for it is, you can get started fast without having to think about it. Just throw shit in a giant nested key-val map and pull it out later. You don't have to design schemas or provision ahead of time. Change the JSON on the fly, and if you hit performance issues, just throw more hardware at it. Move fast and break things.
For 99% of projects that kind of attitude comes back to bite you in the ass. For every hour that you save today, you'll end up spending ten hours in a year or so from now when you're eventually forced to detangle a crusty sloppy mess.
But for a unicorn-aspiring SV startup, that tradeoff works. Almost any startup would gladly spend a hundred man-hours in a year or two, to get back one man hour today. And that works, because the growth rates are astronomical. In a few years, you'll hopefully be a billion-dollar unicorn with tons of resources to throw at the minefield of quick and dirty technical decisions you made at the seed stage. (More realistically, you simply won't be around anymore, at which point the pitfalls in waiting also don't matter.)
The point is don't necessarily pay too much attention to the decisions made by YC startups. This is true even for regular startups with more prosaic aspirations. (Some of us are more than happy to own $10 million companies, and aren't aiming for billionaire or bust.) If your business model is rooted in 1000% hyper-growth rates, that encourages many tradeoffs that are otherwise deeply pathological.
I use Python by choice and the more there is growing need for Natural Language Processing, Machine Learning, Data Mining, the more I can not leave Python.
I try Rust, Ruby and others on the side. But Python is just so heavily fortified now that I constantly suggest startups who are going to build non-trivial tech to select Python. Also the whole Linux, deployment, etc. space is filled with Python.
Biology, Physics, Math folks also use more Python than any other language.
It's also faster than Node. Significantly faster. I was troubleshooting a connection issue to a db from a .NET Core API so I set up a NodeJS API hitting the same db to isolate it to the connection. After I was done fixing things, I compared the API response speeds for the same thing. .NET Core outperformed Node dramatically just like in benchmarks.
Hmm. If you believe in microservices then I think it's fine to use python for your ML pipelines and API's, and some other language for the rest of your application software.
I recall the bootcamp craze that happened a few years ago and most of them were Ruby (with Rails) based.
The result of that is that if I lookup basic concepts in web dev today I'll still find a lot of tutorials and courses aimed at total beginners for Ruby. Meaning that even today, Ruby would be a good choice for CS students in a program where they have no web dev course if they wanted to get a small website running.
Now Python is gaining adoption as an introductory language (MIT ditched Scheme for it!). My bet is that a lot of times python was chosen simply because everyone in the team knew it.
If I knew the scope of the project right when running git init I wouldn't pick these two, but my gut feeling is that a lot of these started as "let's get a demo working, we won't have more than 10 users ever anyways unless we get funding!"
I started learning Python some 7 months after starting to learn C (my first language). I didn't really grok logical program structure until the burden of mem management was behind me.
So, relieving that burden allowed me a playground to pseudocode a program and then translate that logic to a more optimized inplementation.
I still use that strategy in forming mental models.
But now it's a matter of placing weak joints on a pipeline and accepting bottlenecks where they arise.
Can I MVP in 2-6 weeks on those 10 (the startup) alpha users? Fine. Can I leverage my engineering strengths and understand replacement priorities? That's my goal and I run very far from startups that want to build panickedly with toothpicks and glue.
After building two companies with Python, and then one with Javascript, I can see a huge difference. With Python, you get a predictable development velocity. With Javascript, there's lots of unpredictability (this is rapidly improving) because of thrash in the ecosystem. It's hard to overstate the value of predictable timelines and predictable development velocity when building a business.
I use python because its just so much easier to develop. I think for startups iteration speed is magnitudes more important than running efficiency of code
Node will probably start showing up more, though I'd say its biggest weakness is that there's no full framework (that takes care of everything) that's truly dominant. Working in node, feels like working in Java, especially if you're using Typescript. Unlike Ruby and Python, Javascript / TS suffers from the paradox of choice https://en.wikipedia.org/wiki/The_Paradox_of_Choice
Javascript's ecosystem isn't just fragmented, it is downright capricious. In all my decades in tech, I've never seen an environment where dependencies rot out of maintainability faster.
Ruby on Rails in particular made it really easy to get up and going in a few hours or days. I think Python with Django and Flask has occupied that spot.
And for ML and AI-focussed startups, Python is a no-brainer.
The page doesn’t list the year the startups were founded. I wonder if you’d see a good correlation to the founding year.
More startups on the list use Ruby than any other language, and it accounts for over fifty percent of the valuation. It was very popular with startups ten, twelve years ago, but may since have slipped?
On the other hand among younger companions es one might expect to see more Go, Node and maybe Rust?
The author added a note under the main table about that, which goes in the direction of what you're saying:
> Note: Ruby and ruby on rails was a popular choice for YC startups around 2010 - 2012. Anecdotally, ~40% of YC startups used ruby during its peak popularity.
Yeah, Ruby and specifically Ruby-on-Rails was (and still) is very popular in SV startups. Only one mention of Elixir, I'd expect this to change in the future as Phoenix replaces Rails as the easiest framework to build MVPs.
This serves as an interesting counterpoint to Paul Graham's Beating the Averages[0] essay (which argues for using a powerful programming language as a secret weapon to allow a startup to outperform its competitors).
Reading this list, I'm struck by just how mainstream the languages are. I don't have anything against Python or Ruby, but it'd be hard to describe either as a secret weapon — indeed, about the only "secret weapon" languages on that list are Lisp and Elixir, each of which shows up only once.
He wrote that essay nearly 2 decades ago. It was a different world. At the time, the only language with a truly comprehensive open source ecosystem was perl, so, if you were going to work on anything else, there was a decent chance you'd end up writing almost everything yourself. That puts you in a spot where your secret weapon is the language itself.
Nowadays, I'd argue that the secret weapon is never the language itself. Ruby's popularity was never about Ruby itself, it was about Rails. Python is, IMO, an ugly hack of a language, but it's still the one I'm always pushing for at work, because its unbelievably lush open source ecosystem means that choosing Python means you'll end up having to do a lot fewer things yourself. Java's was originally about cross-platform deployment, but, now that it's been a decade or so since anyone was actually that worried about interpreted languages, I'd guess it's now more the fact that Java developers are cheap and easy to hire.
I'm also honestly not super impressed by Graham's comments on programming languages. Sure, Viaweb sold for a lot of money. But then it turned out that it was an unmaintainable mess that needed to be rewritten. Graham has done a good job of pitching the idea that this is because people can't understand the obvious genius of Lisp, and I'll admit, as a Lisper, that that story once beguiled me. But, now that I've been around the block a few times, I realize that code that can only be maintained by its own author is never good code, and that sale price is rarely a good proxy for quality (especially when the purchaser is Yahoo!), and that sheer dumb luck plays a much larger part in entrepeneurs' success than any business essayist cares to admit, least of all the ones writing autobiographical essays.
> Viaweb sold for a lot of money. But then it turned out that it was an unmaintainable mess that needed to be rewritten
I think that's because the code after the sale had to meet different requirements than the code before the sale.
Before the sale, the chief requirement, at least from what I gather from pg's essays, was fast implementation of new features. That was the secret weapon, and Lisp was a key enabler for it. The number of people touching the code was very small, so having the code be understandable by others was not a high priority.
After the sale, the chief requirement, I suspect, was maintainability while running at scale. The before-the-sale period had already sufficiently explored the feature space that fast implementation of new features was no longer a requirement, so the advantages of Lisp were no longer crucial. But having code that lots of people could understand and modify reliably was crucial.
So the code had to be rewritten after the sale. That doesn't mean the code before the sale was bad, just that it was tailored to different requirements.
Code should be obvious the the point that anyone taking a glance can say: "I can take it from here". It happened to me once. I was let go from a gig because the code was so darn easy to maintain (Django plugins), that I coded myself out the job.
There's a tendency to optimize for development speed in startups these days, which usually means gluing together lots of dependencies rather than writing things yourself. That pushes developers towards languages with well-established ecosystems.
When he wrote that, he considered Python and especially Ruby as outside of the mainstream and a bit more powerful, too. The mainstream he was comparing against was Java.
I'd say the ecosystem has move considerably in the direction he advocated. Even Java has moved considerably from 2001 Java.
Also there's a large selection bias: people using less common or more esoteric/demanding languages tend to be better engineers who have invested the time in learning something off the beaten path. That absolutely does not mean it's the right decision since most employees will statistically be average and you want to have a stack that appeals to a large enough talent pool.
Just depends entirely on what you are doing. If you are ecommerce you can basically stick to really slow languages forever as the request volume is so small it should be a small fraction of your application. For consumer scale you end up scaling with caching and eventually migrate or severely optimize - like writing your own language runtime like FB. None of this accounts for all the analytics they are doing which is mostly written in Java and C++.
Even outside of e-commerce -- I would say anything on the web.
Run something like Ruby/Python for the majority of your web services, and deploy Elixir/Go/Kotlin/C++/Swift whenever you hit a data-munching need that Ruby/Python can't handle.
This doesn't fly if you're doing games, or augmented reality (AR), or mobile apps, or embedded stuff... but the web is still a pretty big place to build.
Seems like age is a factor: ruby had its day and companies of that vintage have been around long enough for some of them to be worth bucks. Python is more perennial (over the lifetime of YC itself).
In my annecdotal experience (could be completely wrong) Python and Django are much more common for the same vintage of startups in Europe as when Ruby and Rails were popular in the USA, I don't exactly know what drove it's adoption here.
Even though it says "initial back-end language(s)" it looks like the information is often pulled from the startup's current career pages. Mixpanel is listed as using Go as one of their initial back-end languages, even though Mixpanel was founded in June 2009, several months before the first version of Go was released.
Readit News
That said, this is taking now well established startups that will have begun their lives 5+ years ago in general. If I was picking a back end stack then I'd probably hesitate before picking the .NET Framework. But these days I'd easily pick .NET Core.
Likewise, I'd want a statically typed back end and 5 years ago I'd probably have hesitated before using TypeScript and node together. Now I do it regularly.
Would be super interesting to see the same chart in 5 years with companies starting now.
But since then, that space has gotten a lot more crowded from all sides. There is now a pretty big selection of mature technology to pick from depending on your needs.
In the case of Python, it got also a massive boost from its popularity in datascience. Couple of years ago I thought Python web development is dying out slowly, but seems to have swung back quite a bit.
That said, last I checked regarding performance Rails and Django were pretty much neck and neck with a negligible difference with rails being very slightly faster (iirc).
I think also a big reason why Ruby is feature so heavily in this list is because of it's focus of developer productivity. It's a nice mesh of flexibility and convention. Generally all projects are setup and organized the same but you can still write a conditional if statement a dozen ways. Compared to my personal experience working with Django and Python are the woes of getting everything configured and setup for each project which is frankly a pita. Also, Python is fairly regimented on a lot of flexibility which has its perks too.
Comparatively, I'm sure a chart with database swapped out for language would show a huge over-representation of NoSQL products like Mongo and Couchbase. Yet for the overwhelming majority of projects the right answer is just use a solid SQL product like Postgres.
Again, for SV hyper-growth startups the calculus is different. Mongo has tons of downsides and pitfalls. But the one thing it has going for it is, you can get started fast without having to think about it. Just throw shit in a giant nested key-val map and pull it out later. You don't have to design schemas or provision ahead of time. Change the JSON on the fly, and if you hit performance issues, just throw more hardware at it. Move fast and break things.
For 99% of projects that kind of attitude comes back to bite you in the ass. For every hour that you save today, you'll end up spending ten hours in a year or so from now when you're eventually forced to detangle a crusty sloppy mess.
But for a unicorn-aspiring SV startup, that tradeoff works. Almost any startup would gladly spend a hundred man-hours in a year or two, to get back one man hour today. And that works, because the growth rates are astronomical. In a few years, you'll hopefully be a billion-dollar unicorn with tons of resources to throw at the minefield of quick and dirty technical decisions you made at the seed stage. (More realistically, you simply won't be around anymore, at which point the pitfalls in waiting also don't matter.)
The point is don't necessarily pay too much attention to the decisions made by YC startups. This is true even for regular startups with more prosaic aspirations. (Some of us are more than happy to own $10 million companies, and aren't aiming for billionaire or bust.) If your business model is rooted in 1000% hyper-growth rates, that encourages many tradeoffs that are otherwise deeply pathological.
We have used Postgres from day 1 in March and are now 1200 people. Managing growth with a JSON blob sounds absolutely insane.
You need _more_ guarantees, not fewer.
I try Rust, Ruby and others on the side. But Python is just so heavily fortified now that I constantly suggest startups who are going to build non-trivial tech to select Python. Also the whole Linux, deployment, etc. space is filled with Python.
Biology, Physics, Math folks also use more Python than any other language.
https://docs.microsoft.com/en-us/aspnet/core/host-and-deploy...
It's also faster than Node. Significantly faster. I was troubleshooting a connection issue to a db from a .NET Core API so I set up a NodeJS API hitting the same db to isolate it to the connection. After I was done fixing things, I compared the API response speeds for the same thing. .NET Core outperformed Node dramatically just like in benchmarks.
or Scala if you need bigger scale?
The result of that is that if I lookup basic concepts in web dev today I'll still find a lot of tutorials and courses aimed at total beginners for Ruby. Meaning that even today, Ruby would be a good choice for CS students in a program where they have no web dev course if they wanted to get a small website running.
Now Python is gaining adoption as an introductory language (MIT ditched Scheme for it!). My bet is that a lot of times python was chosen simply because everyone in the team knew it.
If I knew the scope of the project right when running git init I wouldn't pick these two, but my gut feeling is that a lot of these started as "let's get a demo working, we won't have more than 10 users ever anyways unless we get funding!"
So, relieving that burden allowed me a playground to pseudocode a program and then translate that logic to a more optimized inplementation.
I still use that strategy in forming mental models.
But now it's a matter of placing weak joints on a pipeline and accepting bottlenecks where they arise.
Can I MVP in 2-6 weeks on those 10 (the startup) alpha users? Fine. Can I leverage my engineering strengths and understand replacement priorities? That's my goal and I run very far from startups that want to build panickedly with toothpicks and glue.
And for ML and AI-focussed startups, Python is a no-brainer.
More startups on the list use Ruby than any other language, and it accounts for over fifty percent of the valuation. It was very popular with startups ten, twelve years ago, but may since have slipped?
On the other hand among younger companions es one might expect to see more Go, Node and maybe Rust?
Still, an interesting analysis.
Ruby is still popular with startups, and there's no falloff in Rails usage, if the gem download graph is anything to go by.
> Note: Ruby and ruby on rails was a popular choice for YC startups around 2010 - 2012. Anecdotally, ~40% of YC startups used ruby during its peak popularity.
Deleted Comment
Reading this list, I'm struck by just how mainstream the languages are. I don't have anything against Python or Ruby, but it'd be hard to describe either as a secret weapon — indeed, about the only "secret weapon" languages on that list are Lisp and Elixir, each of which shows up only once.
[0]: http://www.paulgraham.com/avg.html
Nowadays, I'd argue that the secret weapon is never the language itself. Ruby's popularity was never about Ruby itself, it was about Rails. Python is, IMO, an ugly hack of a language, but it's still the one I'm always pushing for at work, because its unbelievably lush open source ecosystem means that choosing Python means you'll end up having to do a lot fewer things yourself. Java's was originally about cross-platform deployment, but, now that it's been a decade or so since anyone was actually that worried about interpreted languages, I'd guess it's now more the fact that Java developers are cheap and easy to hire.
I'm also honestly not super impressed by Graham's comments on programming languages. Sure, Viaweb sold for a lot of money. But then it turned out that it was an unmaintainable mess that needed to be rewritten. Graham has done a good job of pitching the idea that this is because people can't understand the obvious genius of Lisp, and I'll admit, as a Lisper, that that story once beguiled me. But, now that I've been around the block a few times, I realize that code that can only be maintained by its own author is never good code, and that sale price is rarely a good proxy for quality (especially when the purchaser is Yahoo!), and that sheer dumb luck plays a much larger part in entrepeneurs' success than any business essayist cares to admit, least of all the ones writing autobiographical essays.
I think that's because the code after the sale had to meet different requirements than the code before the sale.
Before the sale, the chief requirement, at least from what I gather from pg's essays, was fast implementation of new features. That was the secret weapon, and Lisp was a key enabler for it. The number of people touching the code was very small, so having the code be understandable by others was not a high priority.
After the sale, the chief requirement, I suspect, was maintainability while running at scale. The before-the-sale period had already sufficiently explored the feature space that fast implementation of new features was no longer a requirement, so the advantages of Lisp were no longer crucial. But having code that lots of people could understand and modify reliably was crucial.
So the code had to be rewritten after the sale. That doesn't mean the code before the sale was bad, just that it was tailored to different requirements.
I think the ultimate test of code is whether it generates money. That is the sole reason the world cares so much about code.
It’s an unpopular definition, and I don’t like it, but it seems correct.
I'd say the ecosystem has move considerably in the direction he advocated. Even Java has moved considerably from 2001 Java.
Looks like HN brought it down.
Run something like Ruby/Python for the majority of your web services, and deploy Elixir/Go/Kotlin/C++/Swift whenever you hit a data-munching need that Ruby/Python can't handle.
This doesn't fly if you're doing games, or augmented reality (AR), or mobile apps, or embedded stuff... but the web is still a pretty big place to build.
The premature optimization adage has been touted around since forever but it can be widely applied.
He's on HN too.