Jev explained in 7min..

Jev is a new type of model that makes us question how we should really think about optimizing LLMs. When we follow the orthodox path in AI from ChatGPT that was released in 2022, models have been optimized to assist humans in chat applications. Of course, as coding agents like Claude Code and Codex became

mainstream around 2025, we also optimized the models for this as well. But, when we look at the AI stack, how AI actually gets end up being used in the application layer often puts a downward pressure on the layers below. And as a consequence, they tend to morph themselves to optimize for the best

outcome in the application layer. And models continue to optimize to be more helpful to humans and also optimize to be more helpful in coding agents. But, one area that always felt short in use cases was workflow automation. Even a highly intelligent model like Astra and Fable couldn't really cross a threshold here without a huge cost on the way. And

TypeSafe AI that made Jev is making an assertion here that the reason why this threshold is notoriously difficult to cross even for highly intelligent models is because the state-of-the-art models are optimized for different things. The way that they articulate this is a schism in the orthodox line where the optimizing the model for human

preference and further optimizing them for verifiable reward doesn't really carry over when decisions need to be quick given the uncertainty. Now, what we're talking about here isn't that one path is ultimately superior than the other, but Jev is sort of this antithesis to our current trajectory in AI and arguing that we really have been