Jev founder Diogo Almeida on decision models instead of chatbots
Former OpenAI researcher Diogo Almeida explains in a 36-minute talk how Jev reads state, evaluates predefined options, and returns typed decisions with probabilities in about 70-500ms to replace costly LLM calls in agent loops.
Jev founder Diogo Almeida, formerly at OpenAI: “The next era is not another chatbot. LLMs generate - what comes next is AI built to decide.” this is him explaining why Claude Code and Codex are still part of the assistance era, why human approval remains the bottleneck, and how AI systems will operate when millions of decisions need to happen without generating a paragraph every time Jev doesn’t chat, write or explain itself it reads the current state, evaluates predefined options and returns a typed decision with probabilities in roughly 70–500 milliseconds that means the expensive questions hiding inside every agent loop - which tool to use, whether to retry, when to stop, what to approve - can become fast machine decisions instead of full LLM calls “generative models create the work, decision models control the system, and code executes it” in this 36-minute talk, he explains why the future of AI may not be one model doing everything - but an entire stack where each model does the job it was actually built for bookmark & watch the full talk - then read the Jev engineering breakdown below ↓
This post quotes another post: “14 practical tips for using Jev's decision model”