Stuck on "how will you build it into your criteria?"
While recording NexA Voice, the AI host asked it again and again: how will Jev fit into your decision criteria, and will it change how you decide? I got stuck, and not because I had not thought about it. Our systems were never designed around putting any single thing at the center of a decision.
Our setup was parallel from the start
Inside the company we run many AIs side by side, including Repo Deck and nexa-agent, and plug each one in as a part when it is needed. Decisions stack up side by side in a decision graph, and Jev is just one more vote added to it. A new model arriving does not mean everything gets replaced by it. Jev is not that general-purpose in the first place. No AI does everything, and there is no single right answer. Instead of settling the final call at the center, each part does what it thinks is better, and we fix things when problems show up.
A walking, talking baby still does not pass the entrance exam
Jev is not stupid. It is not especially smart either. If a newborn baby said something and started walking, you would be amazed. But talking and walking alone do not get you through a university entrance exam. Honestly, it is awkward to use. On top of that, I have local LLMs that effectively cost only electricity. For ordinary decisions there is no reason to pick Jev.
Where it fits: loose calls, many at once, fast
After thinking it through, one use case was left: very frequent, loose judgments that need to be made at high speed. For example, spotting suspicious comments at full speed during a VTuber live stream. Local LLMs cannot handle a flood of simultaneous requests, and some work really does need to be processed in bulk, at once, and fast. We are not a live-streaming company, though, so there is no place to use it right now. So Jev stays on hand as one of many, and we repeat the same check every time a new tool appears.