![Understanding AI Agents](https://cdn.slatesource.com/9/0/6/906ae032-564a-4ae9-bbbf-60cad88e2ab8.webp)

# Understanding AI Agents

- [Made in Slatesource](https://slatesource.com/s/701)
- By [Steph](https://slatesource.com/u/Steph)
- Science & Technology
- Created on Mar 4, 2026

![](https://cdn.slatesource.com/6/c/5/6c5a2eb5-a284-4162-a50b-5df8f08b13e2.webp)

## Agent, take the wheel.

Agents are here to manage generic workflows from beginning to end. The more specialised the workflow is, the more expertise the agent can get.

## The 3 Core Parts of an AI Agent

1\. Brain (LLM) 2. Tools (APIs, browser, database) 3. Memory (context over time)

![](https://cdn.slatesource.com/5/5/2/5523685b-86b1-457a-b6b7-5a07b6bfde97.webp)

How it works

0%

You give a goal

It breaks goal into steps

It chooses tools

It checks results

It repeats until done

> That loop = autonomy.

## Real World Example

\- Searches web - Extracts data - Structures comparison - Outputs summary table - ...

[AutoGPT (GitHub – autonomous AI agent)](https://en.wikipedia.org/wiki/AutoGPT?utm_source=slatesource)

[martimfasantos/ai-agents-frameworks](https://github.com/martimfasantos/ai-agent-frameworks?utm_source=slatesource)

[HeyNina101](https://github.com/HeyNina101/real-world-llm-agents?utm_source=slatesource)

## What's the Catch?

Agents aren't perfect, their autonomy usually means you will have to double check the results. - Can make wrong decisions - Can misuse tools - Expensive (multiple model calls) - Hard to debug

## Why Everyone Talks About Agents

We are shifting from having a “Software you use” to having a “Software that works for you.” Having the speed of a machine and the brain of a human would allow workflows to be exponentially faster.

![](https://cdn.slatesource.com/1/7/a/17afadc2-952e-4156-a270-f4dfc9455792.webp)

## What Should You Learn Next?

Beginner path: 1. Prompting basics 2. APIs 3. Simple automation flows 4. Agent frameworks Start small. Automate one workflow.