Stop Dividing AI Compute Equally: Top Models for Senior Engineers Save Money, Newbie Grind No Longer Works

At the InfoQ media livestream during the 2026 World AI Conference, experts Teng Yu and Tang Feihu discussed the deep changes AI coding brings. They agree that AI prioritizes senior engineers' judgment, while newcomers who relied on repetitive coding lose out. Both use multiple models, but note that Token consumption is no longer the key metric; instead, task completion and value matter. Tang Feihu believes models have judgment, citing Kimi K3's ability to generate a 3D game from a single prompt, and a case where a model refused to upload code due to safety concerns. Teng Yu argues models lack true responsibility and can't make final decisions. For enterprises, they warn against using Token usage as KPI, as it leads to gaming. They suggest focusing on ROI, selecting models per task, and giving top models to senior engineers. Newcomers should start using AI early, avoid competing with AI on LeetCode, and consider humanities education. Both see a shift in software engineering from writing code to understanding systems and taking responsibility. Tang Feihu still recommends studying computer science, but with changed curricula, while Teng Yu suggests traditional CS degrees may lose value.