AI Your Way: MCPs vs Skills vs SubAgents 🤖🚀 - Sam Basu

Code is cheap, but software is expensive. Modern AI is a big opportunity to streamline and automate developer workflows for better productivity. There are some challenges though – AI Models often lack knowledge and AI Agents need expertise/guidance to reliably pull off complex workflows. Context is everything for modern AI and you can bring it.

Model Context Protocol (MCP) aims to provide a standardized way to connect AI Agents to different data sources, tools and non-public information - the point is to provide deeply contextual information/expertise to AI. Skills are higher-level behaviors and instructional guardrails, that define how and when AI Agents should leverage tools to accomplish something meaningful. Subagents in AI are specialized, task-focused agents designed to handle specific, well-defined tasks within a larger AI system.

In terms of the food industry:
AI Agent = Chef 👨‍🍳👩‍🍳
MCP Tools = Raw Ingredients 🥔 🥩
Skills = Recipe Cards 📝 📇
SubAgents = Sous Chef 🔪 🍳
Loops = Door Watcher 💂👀
Developer = Restaurant Owner 👑.

So, what should developers use to bring context and guardrails to make AI work their way? Well, it depends and sometimes, the answer might be whatever combination makes developers most productive. With official SDKs and well-thought-out guidance, it is a breeze to work with MCPs, Skills or SubAgents. Developers could bring their own data, APIs, services, coding patterns and structured guidance to make AI Agents work their way. And AI Agentic workflows work the same way in IDEs or Terminals, paving the way autonomous ways of getting work done with AI. With contextual expertise to light up unique coding workflows, AI Agents can make developers ultra productive – upwards and onwards.