AgentVet Weekly — September 21, 2026

From the founderThe debate over slowing AI development down keeps going, and I keep landing in the same place: it only works if it's global. If the US pumps the brakes while adversaries keep moving at full speed, that doesn't make anyone safer. It just hands away whatever lead we have, or puts us further behind if we're already not ahead. Real safety needs a global agreement, not a unilateral pause. Worth watching this week: a new model called Jev claiming a real break from the usual LLM tradeoffs. It's significantly faster, significantly cheaper, and more reliable, with no hallucinations. The company's explanation is that it never generates free text at all, which is what removes the hallucination problem in the first place. The speed comes from parallel processing: it generates the entire output in a single query instead of token by token. If that holds up, it's a genuinely different approach, not just an incremental one. More on this soon.