In September 2026, a startup called TypeSafe AI released a model called Jev, and within 24 hours, nearly 13% of Vercel’s paid developer teams had tried it. That made Jev’s debut more than twice as fast as any previous model launch on Vercel’s AI Gateway. Three weeks later, OpenAI released a competing product.
The founder behind it? Diogo Almeida, a former OpenAI researcher who spent two full years quietly building before ever making a public move.
What This Actually Means
Jev is not another chatbot. It is a classification engine, a model specifically designed to make fast, high-volume yes/no and categorical decisions. Think: is this customer message a billing question or a technical issue? Does this document need human review? Should this email go to the priority inbox or the junk folder?
These are the thousands of micro-judgments that software already handles every day, but until now required either crude old machine learning classifiers or expensive, slow large language models. Jev threads the needle: natural language instructions (easy to set up), but output tuned for speed and cost at scale.
Niels Mouthaan, an independent developer, used Jev to build a grammar feedback prototype. Ashutosh Mathore, founder of environmental compliance startup Atlensa, said testing Jev made him reconsider “where we may be using LLMs out of habit even when the task is not really a language problem.” Paolo Rosson at bookkeeping company Dext estimated Jev’s cost at seven cents per 1,000 classification calls, a fraction of what a general-purpose model runs.
The lesson here for any founder is not about AI models. It is about the gap between what a tool is technically capable of and what it was actually optimized for. Almeida saw that gap at OpenAI and bet two years of his life on it.
The Numbers Behind It
- 13% of Vercel’s paid teams tried Jev within 24 hours of launch
- 2x faster adoption than any previous model launch on Vercel’s AI Gateway
- 3 weeks from Jev’s launch to OpenAI’s competing Decisions API rollout
- 2 years of development before TypeSafe made a single public release
- $0.07 per 1,000 classification calls, vs. significantly higher costs for LLM alternatives
TypeSafe was cofounded by Almeida alongside Erik Gafni and Sasha Sheng. The company named Jev after 19th-century economist William Stanley Jevons, whose paradox holds that greater efficiency in using a resource leads to more consumption of that resource, not less. If classification tasks get cheap enough, the argument goes, businesses will automate thousands of decisions they currently skip.
The Hustler’s Library Take
The most telling detail in this story is not the viral launch. It is that Almeida originally thought making Jev reliable would take a week. It took two years.
That gap between expected effort and actual effort is where most founders quit. Almeida did not, and he had a structural advantage: he had worked inside the beast. He helped build ChatGPT at OpenAI. He watched the platform up close, understood the workflow, and identified the specific pain point that a billion-dollar company was too distracted to fix for itself.
The contrarian angle here is worth sitting with: the product that beat OpenAI was built by someone who used to work at OpenAI. Not despite that experience, but because of it. He knew exactly what the internal roadmap would not prioritize. That is a playbook available to anyone currently working inside a large company, in any industry.
OpenAI releasing Decisions API three weeks after Jev went viral is not a coincidence. It is validation. When a company with a trillion-dollar ambition breaks from its roadmap to copy you in 21 days, that is the market confirming your thesis faster than any investor meeting ever could. As Almeida himself has noted, the whole thing echoes the pattern of AI making it easier to start but harder to sustain an advantage.
For the rest of us: the lesson is not “work at a tech giant before starting a company.” The lesson is solve the problem the dominant player is ignoring because they are too focused on the headline feature.
What You Should Do
1. Map the “out of habit” problem in your own business. Almeida’s insight about LLMs being used “out of habit” applies far beyond AI. Where are you using expensive or slow solutions for tasks that a purpose-built, simpler tool could handle? This could be software, staffing, or even your own time. Start with your highest-volume, most repetitive decisions. Those are the ones worth re-examining.
2. Build your product for the workflow, not the headline use case. TypeSafe did not try to build a better chatbot. They built for the unglamorous, high-frequency decision layer that runs underneath everything else. Whatever your industry, there are a dozen “boring” workflow problems that the market leaders overlook because they are chasing the big, visible feature. The skill of identifying underserved workflow problems is worth more than most formal business education.
3. Give yourself a realistic timeline, then double it. Almeida estimated one week. It took two years. This is not a failure, it is calibration. The founders who survive that recalibration are the ones who have thought clearly about what their time is actually worth and committed accordingly. If you are building something genuinely hard, build the expectation into your plan from day one. Underestimating the hard part and running out of runway is how most good ideas die.
TypeSafe is building on insight from inside the machine. The leanest, fastest-moving startups this year all share the same trait: they picked a problem the established players treat as overhead, not opportunity.
Source: Fortune | Additional context: Bureau of Labor Statistics
Want breakdowns like this every day? Join Hustler’s Library free and get the business stories that actually matter.
Ready to Know Where You Stand?
The Business Journey dashboard maps your exact position across all 13 stages. Track your progress, unlock resources for each step, and build with a framework used by thousands of founders at Hustler's Library.
No credit card required · Takes 3 minutes · Personalized to your stage
