OpenAI Just Killed Jev. It Took Three Weeks.
Jev launched as an AI model that makes decisions instead of writing essays. Three weeks later, OpenAI shipped the same idea. Jev is cheaper, but OpenAI is already in your account, and that's why a head start isn't a moat.
On September 15, a startup called TypeSafe launched a new kind of AI model called Jev. Developers lost their minds. TechCrunch wrote that it was thrilling developers. Vercel said it reached more than twice as many paid teams in its first 24 hours as any previous model launch. TypeSafe got so much traffic it briefly couldn't serve its own API.
Three weeks later, OpenAI shipped the same idea inside the account every developer already has.
To be fair, Jev isn't dead. TypeSafe is still selling it, it raised a $40 million seed, and the company says it has since closed a big Series A led by a16z. But I think OpenAI killed it anyway. Here's why, and why it matters even if you've never written a line of code.
What Jev actually does
Almost every AI model you've used writes words. You ask a question, it writes you an answer. Jev doesn't write anything. You give it a question and a list of possible answers, and it tells you which one is right and how sure it is.
Is this email a sales lead or spam? Is this review angry or happy? Did the customer say yes to the appointment or no? That's it. A label and a confidence score, in a fraction of a second.
TypeSafe calls it a System One model, after the fast, gut-reaction kind of thinking from Daniel Kahneman's book. The founder, Diogo Almeida, used to work at OpenAI on the research behind ChatGPT. So this wasn't some random team. These were insiders.
It turns out that's what most businesses actually need from AI. One developer told Fortune that with a normal model, "you pay for a full written answer when all you need is a label." Jev charges $0.042 per million tokens going in, and nothing for the answer coming out. Basically free.

My favorite detail is the name. Jev is named after the economist William Stanley Jevons. The Jevons paradox says when you make something cheaper, people use way more of it, not less. Cheap labels mean you start labeling everything. (Apparently it applies to competitors too. Prove an idea works and suddenly everybody wants one.)
Then OpenAI showed up
Two weeks after Jev launched, OpenAI announced its Decisions API at DevDay. On October 6 it was in public beta. By October 22 it was generally available.
It does the same job. You send questions with a fixed set of answers, and you get back probabilities and a confidence score. No essay. The output is free here too.
And OpenAI didn't really hide where the idea came from. According to Fortune, OpenAI's Nikunj Handa credited Jev with "inspiring this whole thing" on the Latent Space podcast. Fortune reports it hadn't even been on OpenAI's roadmap four weeks earlier.
Four weeks. From not on the roadmap, to inspired by a startup, to shipped. That's how fast the biggest company in AI can copy you.
Is it actually better?
On price, no. Jev is still cheaper on paper: $0.042 per million input tokens against OpenAI's $0.10. (OpenAI drops to a penny when the input is cached, so even that lead depends on how you use it.) You'll still find developers arguing Jev is the better default for judgment questions because it's cheaper and faster. They might be right.
But price isn't what decides who wins. Look at what OpenAI has that Jev doesn't.
It's already in your account. No waitlist, no new vendor, no new contract. If you build on OpenAI, you already have a key and a credit card on file. Jev launched in early access with a waitlist.
It does more. You can send up to 400 questions in one request, and it reads images, not just text. Jev only takes text.
Legal already said yes. Decisions is eligible for HIPAA and zero data retention, with your data kept in the US or Europe. If you sell to a hospital, a bank or anyone with a compliance team, that matters a lot more than a few cents.

Nobody ever got fired for picking OpenAI. "We already use them and legal already signed off" beats "this startup is a little cheaper" almost every time. That's what I mean by better. Not a better model. A better position.
Why builders should care
Here's the part that actually gets me excited. When a decision costs almost nothing and comes back fast, you can build things that were too slow or too expensive before.
Save a ton of tokens. Put a cheap decision in front of your expensive model. Does this message even need the big model, or is it just someone saying thanks? Which three of these 40 documents actually matter for this question? Ask first, then only send the expensive model what it needs. Smaller prompts, fewer calls, smaller bill.
Routing gets way easier. Every AI system has a traffic cop somewhere deciding what happens next. Is this a booking, a complaint, a billing question or spam? Usually that's a pile of keyword rules or a giant prompt that breaks the first time a customer says something weird. Now it's one question with a fixed list of answers, and you get a confidence score back. Confident? Do it. Not sure? Hand it to a human. You can even ask a stack of questions in the same request: is it urgent, is the customer angry, what language is it, which department.

Live agents that actually feel alive. This is the one I'm most excited about. Think of an AI avatar with a moving face, or a voice agent on a phone call. Every split second it has to decide something. Did they finish talking? Should I nod, smile, look concerned or keep listening? On a phone call, a one-second pause feels like forever, and a normal model is too slow for that. A fast decision can pick the expression or the next move from a list while the big model works on what to actually say. That's the difference between talking to a robot and talking to something that feels like it's listening.
This is exactly why Jev went viral, and exactly why OpenAI wanted it. The idea is great. That was never the problem.
Why this keeps happening
There's a name for this in tech. When Apple builds your app's main feature into the iPhone, you got Sherlocked. It's been happening for more than 20 years. Now it happens to AI startups every couple of months.
I wrote about the same thing when AI receptionist startups lost their moat, when no-code automation tools lost theirs, and when Lovable hit the same wall. The pattern never changes. A small team proves an idea works. The big platform watches the numbers. Then it ships the idea as a feature to millions of people who already pay it.
Usually I'd bet on the small, fast team. That's the whole point of Darius had the money, Alexander had the speed. But this time the giant was fast too. When the big guy is also the fast guy, a head start isn't enough.

And honestly, I don't even blame OpenAI. Good artists copy, great artists steal. If someone showed me a cheaper, faster way to do something my customers need, I'd build it too.
What actually protects you
So if a head start isn't a moat, what is? Pick your poison, but there are really only two.
The first is distribution. Owning the customer. Being the one they already call. That's why I keep saying the only skill that beats systems thinking is sales, and why I spend so much time attacking the distribution bottleneck. OpenAI won this round with distribution, not with a better model.
The second is selling outcomes instead of AI. Nobody who owns a plumbing company wants to buy a "System One model." They want the phone answered and the job booked.
That's how I build at HyppoAI. We use these models every day, but I don't sell the model. I sell the answered call and the booked appointment. If OpenAI ships something ten times faster next month, I swap it in, and our customers never notice except that things got faster. When you sell the engine, a better engine kills you. When you sell the outcome, a better engine is a free upgrade.

The lesson for the rest of us
You're probably not building an AI model. Doesn't matter. Every business has some version of this risk: a bigger competitor, a platform you depend on, a new tool that does your thing for free.
So ask yourself one question. If the biggest company in your space copied you tomorrow, would your customers still pick you?
If the answer is yes, it's because of something nobody can copy in three weeks. Your relationships. Your reputation. Your judgment. The fact that you pick up the phone.
If the answer is no, you don't have a business yet. You have a head start. Use it while you still have it.
Want to talk about what you're building?
Get in touch


