Skills vs MCP: why your agent needs a filesystem
Автор: What's AI by Louis-François Bouchard
Загружено: 2025-10-27
Просмотров: 1968
Skills, not protocols: the simplest way to make agents actually useful. Instead of shipping a context-hungry spec like MCP, declarative Skills live in your filesystem with YAML discovery, then execute real code when needed. Read a file, import a script, compress a GIF, extract a PDF table... all deterministic, fast, and cheap on tokens. Think of it like this: MCP is your API integration layer; Skills are a local plugin system with progressive loading and a sandbox that can really run. The catch: you need an execution-capable runtime. The win: near-zero tokens until the moment of action. If you’re building agents in 2025, design for execution first, prompts second. Curious to see this in action? Drop a comment and I’ll share examples.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AIengineering #LLM #Agents #short
Доступные форматы для скачивания:
Скачать видео mp4
-
Информация по загрузке:
