Skill issue: stop deploying vision language models, use them with Skills — Merve Noyan, Hugging Face
Developers should stop deploying vision language models directly and instead use them with specialized model tools via a toolkit.
“your coding agent is a bit of a computer vision engineer”
Hugging Face's Merve Noyan argues against using vision language models directly for everything, instead advocating agent-callable toolkits that wrap specialized, permissively-licensed models (Apache 2.0/MIT) like RFDETR and SAM 3.1. She introduced 'Vibe Vision,' a toolkit exposing favorite CV models as tools for coding agents, and warned developers about AGPL-licensed models like YOLO. It matters as a practical engineering pattern for building vision programs, though it is a developer-tooling insight rather than a major industry signal.