I just passed waitlist today, and tried shameless vibe-code-drop-in-replacement to my company's current AI workflow.

I just passed waitlist today, and tried shameless vibe-code-drop-in-replacement to my company's current AI workflow. All I did was giving official Agent Skill to my coding agent, and order it to adapt current benchmark script to Jev. Then... Boom! It just works! This benchmark tasks are categorizing 111 arbitrary user inputs to two types of categories. First task is 1 of 39, second one 1 of 62. Both are nullable (should be labeled 'none' if doesn't fit to any category) User inputs include unofficial terms, slangs, STT(=typo!) so we give keywords and their frequent aliases in the prompt. Interesting points: * My company is based in South Korea. So this benchmarks test *Korean* user inputs. Jev just... works with it, even prompt was fully in English (written by LLM). * Prompting in Korean also work, but it doesn't seem to improve performance, at this stage. * Current workflow calls LLM twice in parallel, but coding agent called Jev