Constraints that make developers faster
The Stack Overflow Podcast
October 2, 2026
Ryan chats with Julien Verlaguet, CEO at Skip Labs, about finding the balance between human tolerance and tooling constraints, the spectrum of typed programming languages, and building cost-effective tooling for AI agents.
Speakers Ryan Donovan, Julien Verlaguet
TopicsTechnologySociety & Culture
Ryan Donovan (0:07)
Hello, everyone, and welcome to The Stack Overflow Podcast, a place to talk all things software and technology. I am your host, Ryan Donovan.
And today, we are talking about creating tooling to constrain the programmer and programs to make sure they're not making mistakes, whether that is through a language or through an agent harness. My guest for that is Julien Verlaguet, who is the CEO over at Skip Labs. So welcome to the show, Julien.
Julien Verlaguet (0:33)
Thank you. Thanks for having me.
Ryan Donovan (0:34)
Before we get into the tooling topic, can you tell us a little bit about how you got into software and technology?
Julien Verlaguet (0:39)
TLDR, I got really into it because my parents bought a computer when I was a kid.
And back in those days, I wanted to play video games really bad. And the only way to play video games was to copy the source code of some games that was at the end of some book to learn basic. And that's how I ended up debugging basic back in those days, because none of those programs was ever working, at least not from the get go. So that was early on. Then at university, I got very quickly involved into programming languages and type theory and proofs and all this kind of field. And I think it's a French thing. There's a lot of French people interested in type systems somehow. And so I started programming languages in Paris. And then throughout the rest of my career, I was always involved in anything regarding DevTools, correctness, programming languages, etc., etc.
Ryan Donovan (1:36)
In your involvement in programming language, you've created a programming language or two.
Julien Verlaguet (1:40)
I created several, but I think the only ones that are worth mentioning are Hack and Skip. So there was one called LinearML, which was more experimental, but it was a bit, think of it as Rust for ML. So it was Rust before Rust existed.
This was not a major implementation. The mainstream languages would be Hack and Skip. So Hack was created before TypeScript, but I guess the best way to describe it these days would be to say it's TypeScript for PHP. And Skip is a programming language designed for reactive programming. So it's domain-specific language, if you will. It's not a general programming language.
Ryan Donovan (2:21)
These programming languages are essentially tooling for creating software, right? How did you think about designing them to prevent mistakes?
Julien Verlaguet (2:29)
Things have changed now with AI, obviously. But back then, the balance had to be found between what a human being would tolerate and what the tooling could bring to the developer. Because, I mean, we could go all the way, right? Going all the way would be, you have a dependent type system, ACDA, Coq, or this kind of stuff. And then you can write whatever you want in your type, and be as specific as you want, and enforce that. The problem is people are not doing that because the cost of doing that for a human being is just way too high. And so I think back then, the question was not so much about what can we enforce statically. We knew already back then that we could do and enforce pretty much whatever we wanted statically.
The question was more, what do we want to enforce statically that will be bearable for a human being in that context. And the context is going to matter a lot. So for example, when I started my career, I was working on safety critical embedded software. So the kind of software that runs on a plane or nuclear plant or this kind of stuff. So the kind of software where when something goes wrong, people are going to die, right?
And in that kind of context, developers are willing to put up with a lot of hurdles, like you can give them to link that is very constraining, and they're going to jump through the hoops and make it work until it's correct. In other contexts, such as web development, eh, not so much. Like what they're willing to put up with is they will typically take anything that makes them gain time. So if you give them a tool that prevents mistakes, that they were going to have to debug anyway, then they'll do that. So go back to your original question. I think for hack and for any system of the sort. So hack back then, the trade-off was, we have a bunch of PHP developers at Facebook. They are shipping a ton of code. They start to slow down now because the code base is becoming large enough that maintaining a large code base written in dynamic language comes with a set of challenges.
22 more minutes of transcript below
Thousands of transcripts fetched by people building searchable podcast archives
Fetch the whole transcript
The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000793191416