Geeks and Suits — Understanding the Basics

At its core, AI today is the LLM — the large language model. That's the "brain." On its own, it powers simple use cases like a chatbot: you ask, it answers.

The real power comes from what you wrap around the model. Those wrappers are called harnesses — and they turn a smart chatbot into something that can actually do things. There are a few kinds.

1. Built-in harnesses. Features the model's own platform includes out of the box. Claude, for example, comes with tools like scheduling, skills, and many others. You don't build these — they're ready to use.

2. Off-the-shelf agent harnesses. Tools built by outside companies that connect to the model rather than being part of it. They give the AI extra abilities, so it can carry out multi-step tasks on your behalf. Expect many more of these over time — plug the AI in, and it goes off and gets work done.

3. AI-in-the-code harnesses. Here, a piece of software has AI built into it. The code still runs the way code does, but the AI helps direct its logic and workflow — making the program smarter and more flexible than code alone.

4. Connectors (MCP). A connector links the AI to another program or data source. The common standard is the MCP connector (Model Context Protocol). Each connector defines a specific set of tools the AI is allowed to use, so it can safely query that program and act within set limits. There are many off-the-shelf MCP connectors, and you can build your own.

The takeaway: the model is the brain, and harnesses are how you give it hands. Built-in features, outside agent tools, AI woven into software, and connectors that plug it into your other systems — each one extends what the AI can do for you. Understanding these building blocks is the first step to putting AI to real work.