2026-09-21 · monday Disclaimer: I wrote this but Codex running on Sakana Fugu did the styling and added the links/references/etc following my instructions over a few prompts.
I have been meaning to write about what it takes, other than great engineering, to launch a successful product and I find the TypeSafe frenzy a great timing to share my views.
As a technical founder, it’s so easy to fall into the trap of believing that because your engineering is sound it will need no introduction. Engineers suffer from the curse of knowledgeNote 1Curse of knowledge. Named by Camerer, Loewenstein & Weber, Journal of Political Economy 97 (1989): better-informed agents cannot discount what they know when modelling someone who does not know it., it can be frustrating to be sitting at the innovation of technology and balancing having to consistently stay disciplined enough to practice self-awareness outside your bubble. To remind yourself, whoever is sitting behind the screen and about to make a decision on if they should trust your work may not know enough about the subject as you do.
I believe this is why we see a lot of lag between academics and applied technology in the form of businesses. Though that’s not what I want to write about. Forget the academics or business angle aside, let’s talk about communication.
If you wrote code for a living you certainly might have come across compiled programming languages at least once in your life. The thing with compiled programming languages, and especially strongly typed ones, is that they allow a certain kind of “prediction” over behaviours in the code. There is something magical about having a compiler tell you when you made a mistake and preventing bugs from making it into your software before it executes in front of the end user or program. Then there’s static analysis tools, and things like IntelliSenseNote 2IntelliSense & Roslyn. IntelliSense is an editor feature and dates to 1996. Roslyn is the compiler platform, open-sourced in 2014, that lets other tools query the compiler’s own model of the code. or Roslyn APIs for code analysis (using C# as an example here, but that is not really the point you should focus on). Tools that let you operate on top and around the code, aided by a compiler and by data contracts that can be executed very fast, allowed something even more magical. Now you don’t need to wait for the compiler, you can have something operating in a file telling you right then and there a suggestion on command. First I didn’t know what to do with that, when writing code, code can be about anything and everything. What suggestion is the ideal one? Well, the naive idea would be to think “anything that is allowed by the compiler, of course”, but you quickly would realize some, despite being sound to the compiler, are just dead wrong. You see, not everything is relevant or useful and that the context of your program is what determines the relevance and applicability of accepting a suggestion.
This brought us to better tools, they could scan and understand project context, and even operate under set of rules defined by the user at the project space, so that they get better at giving you what you need when you need it. Rider’s ReSharperNote 3ReSharper & Rider. ReSharper launched as a Visual Studio extension in 2004. Rider is a separate IDE that runs the ReSharper engine as its backend. is one example here, it could produce on the fly code quality analysis that are sensitive to your project and smarter than the C# compiler, it helped with things like null referencing, list enumeration, and even large scale refactoring accounting for imports. Boy how glorious it felt using it. Programming was so fun, you could focus on the core logic of what you were doing, and not stress out about forgetting some verbosity of the language design in how it had chosen to give you certain types of guarantees.
What does any of this have to do with Jev? I will get there, but let’s talk about Cursor first. Heh!
Cursor came along and basically reinvented all of the things I shared above, for everything, not just C#. They really were just an autocomplete UX reinvented with modern tooling. The technology behind it was already there. People were already copy-pasting text and copy-pasting it back. I mean, for those of you that were not on compilers, and strongly typed languages, maybe you had your own version of that in a homegrown way, like coding snippets, StackOverflow gotchas, etc.
Why was Cursor so successful in overtaking everyone’s workflow overnight? Because they nailed the approach to requiring no explanation when selling you on the underlying magic. You didn’t have to understand how it worked; you just had to use it, and they really lowered the bar on using it.
We had some Cursor-like products in IDEs before them as well, such as tools that allow you to have an inline REPL over your program so you can watch computation happen as you write code.Note 4The canonical ancestors. Bret Victor’s Inventing on Principle (CUSEC, 2012) argued creators need an immediate connection to what they make; Chris Granger’s Light Table (2012, Kickstarted) implemented it, and Quokka.js still ships the inline-value idea into VS Code today. Granted this isn’t the same, but it’s a sort of similar design to a similar problem. Maybe the problem of answering a question “in the moment” and unblocking your next step? Anyways, not important, hopefully you get what I mean here. The point is, the innovation wasn’t really a new technology per-se, but a new approach, a new medium, and a way to consume and use it that fit better with the mental model of the consumer.
Now let’s talk Jev.
The mental model of a consumer is often the biggest overlooked thing by engineers. Especially so when they also happen to be in the intersection of customers of their own products. They will often fail to realize how packaging, marketing, and distribution of a product have less to do with an engineer’s mental model for building the product itself than it does for selling it to a larger audience.
Right, how does this have anything to do with Jev? Well, they had a really amazing launch. Time is yet to tell if it is going to be a Cursor-like phenomenon, but the trend is certainly showing great signs for them. However, just like when Cursor launched, lots and lots of engineers were bringing on the “another day, another vscode fork” type of hate towards it, we are seeing the same with Jev. Most of the strong opinions seem to come from a position of those who deeply understand the machinery of it and quickly find themselves within the cognitive distortion of struggling to see how the strong adoption is justified when the technology to use it was there all along this whole time. This isn’t anything new to markets either, as someone pointed out in HN, you could say “Jev is just BERT” and then again “Dropbox is just an ftp account mounted with curlftpfs!”Note 5johnfn on Hacker News, 19 September 2026.
Okay, then what is Jev? What’s this BERTNote 6BERT (Google, 2018) reads a whole sentence at once and hands back numbers instead of writing text back at you. The name is a Sesame Street joke, and the field kept going: ELMo, ERNIE, Big Bird, Grover. So engineers do name things carefully. They just name them for people already in on the joke. thing? See, that is the point of Jev. I don’t have to tell you what BERT is, and that’s the point I am making. Actually, perhaps for a very junior engineer, both things are just weird nouns that sound like someone’s name, just think about that. Names. Imagine knowing so much about something, being so blessed with knowledge, that from a single word you can understand how to instantly build it. There’s a xkcd article I like that brings this self awareness that for each thing “everyone knows” by the time they’re adults, every day there are, on average, ten thousand people hearing about it for the first time. Nailing product market fit is kinda like cracking the delivery of solved problems to these many ten thousand people popping up each day helping shape our industry through the evolution of how society and humans choose to interact.
One way I practice being aware of this while doing anything is remembering quotes like Ecclesiastes 1:9 “What has been will be again, what has been done will be done again; there is nothing new under the sun.” or Lavoisier’s “In nature nothing is created nothing is lost, everything changes”.
Fun fact: when having Codex help me publish this, it helped me learned that actually Anaxagoras and Lucretius were arguing about the same much earlier. The first basically said that things don’t really appear or disappear, they mix and separate, while the latter said “nothing from nothing ever yet was born”. The cool twist is that it still reads to me like the same thought getting re-cut for a new audience only two and a half thousand years apart. Which is sort of my whole point.
Jev’s innovation went beyond just classical machine learning that “everyone knows” and was already using. They created a standard application interface with data contracts and core primitives crafted for the existing software paradigm in a way that landed exactly where it hurt in the market. System One Model reads much softer than BERT, even my wife, who doesn’t program or knows engineering, was able to understand “Model” referred to “AI” as in “the tool she uses” whereas BERT sounded like someone to her, whereas System One she understood as a primitive. Jev also got into using the Fast and Slow analogy to explain each,Note 7System 1 / System 2 is not Kahneman’s coinage. He adopted the labels from Stanovich & West (2000) and says so in Thinking, Fast and Slow (2011). Two things worth holding: the dual-process framing is contested in cognitive science, and in 2017 Kahneman publicly conceded that the book’s chapter on social priming rested on evidence he had over-trusted. which I find is a really great mental model when focusing purely on applications of using it. Even during onboarding, I got a modal that asked whether I was ready to use it and challenged me to check my understanding.
It’s not all packaging either, with Jev you can make up a schema and it just answers without needing labels or training, plus they have a larger context window than the clones out there. Curious to see if they will publish more around the architecture or if another competitor will solve for that yet.
Many are pointing out defensibility and moat of their tech, and while that may be valid, where are all the ftp accounts mounted with curlftpfs since Dropbox launched in 2008? Well, turns out that after shipping a product there’s the whole next chapter of it which is sustaining a relationship with your customers. In the early days of Stripe, they were famous for being a very horizontal organization internally and it was all hands on deck for Support, the same engineers building the product were there to take calls from customers, the people selling, even their CEO.Note 8Paul Graham’s Do Things That Don’t Scale (2013), which names the “Collison installation”: rather than emailing a signup link, the Collisons would ask “give me your laptop” and set the merchant up on the spot. That was just as important for growing self-awareness of how your product lands outside your bubble, as it was for managing the relationship with those behind the screen making decisions to trust your work.
Lastly, Jev onboarding says to only ask it for the kind of judgment a knowledgeable person makes in a second, given the right context. So they optimized it for the time it takes software to make a decision, and in their marketing strategy they optimized for the time it takes you to make one about them too.
Alright, I’m gonna here. Hopefully this was a fun read, I could have gone deeper into things but I’m still figuring out how I want to blog and write, and these days there’s too much noise everywhere.
Go try Jev, it’s fun, or join the party into making your own,Note 9People took this literally within a week. Kev (Jared Palmer) on Qwen3.5, Laya (ConvAI), OpenDecision, openjev-sglang, and by one HN count three separate projects calling themselves “openjev”. The trick itself is written up publicly. in any case just remember that communication is everything.