I want to talk about something that's been bothering me for a while: the word "agent." 

If there's one buzzword that every single one of us has heard over the past year or two, it's this one.

Every pitch deck, whether it's from a scrappy startup or a massive enterprise, seems to lead with some version of "we're building an agentic platform to do X, Y, Z." And honestly, a lot of the time, that's the right move.

But if you actually talk to the people building these systems day in and day out, you'll quickly notice something: there's a lot of hype. 

There's overuse, abuse even, of the words "agent" and "agentic framework." Many times, what's being called an agent is really just a glorified function call that we've been writing as programmers for decades.

5 questions to ask your AI agent vendor before you buy
Most AI agent failures don’t happen during the demo. They happen when APIs fail, context windows explode, costs spiral, and nobody can explain why the agent made a decision. Here are five questions that separate production-ready platforms from expensive experiments.
AI Accelerator InstituteAndrew Lovell

The brain in a vat

If you take just one thing away from this article, let it be this: the LLM is the brain behind any agentic system. But the LLM is a brain in a vat. It doesn't make sense. It can't act. That's where the agentic architecture comes in…

Agents are basically tools in some form. 

💡
At their core, agents give LLMs access to tools that let them sense and act in the world. Think about us as humans for a second. We have a brain with some memory, and we have our senses: eyes, ears, skin, hands. We use these to perceive the world and to act on it.

LLMs work in a similar way, except their knowledge is restricted to whatever training data they've been fed.

To perceive, meaning to get access to fresh data, the LLM has to call out to something. If you ask an LLM "how's the weather today," it has no idea.

It needs to hit an API, call a function, or use some tool that brings back the latest info. Once it has that context, LLMs become genuinely impressive at reasoning, summarizing, and synthesizing on top of it.

For expert advice like this straight to your inbox every other Friday, sign up for Pro+ membership.

You'll also get access to 300+ hours of exclusive video content, a complimentary Summit ticket, and so much more.

So, what are you waiting for?

Get Pro+

The building blocks: from functions to MCPs

Let's walk through these constructs one at a time, because understanding them is the difference between building something useful and building something expensive that doesn't work.

Traditional function calls are the vanilla, simple, modular pieces of code that have been around since the very beginning of software engineering. I've actually seen people ask LLMs, "Here are two numbers, write me the best prompt to add them together."