At DAC.digital, we wanted to cut through the hype and explain how AI agents actually work under the hood. So our CTO, Krzysztof Radecki, recorded a practical walkthrough that demonstrates the mechanics of agentic AI step by step. This post explains how we designed that tutorial and the technologies used to build it.
The Goal Was to Show the Tech
A lot of AI content focuses on outcomes, but there's not enough content that explains how the system works.
Our goal for the tutorial was to demonstrate three key things:
What a large language model really is
How tokens and context windows shape model behavior
How AI agents are built by orchestrating systems around an LLM
To make this tangible, we created a small agent-based application and progressively extended it during the demo.
As said before, there’s a lot of hype around AI agents, but building reliable systems requires understanding the fundamentals. An AI agent is not just an LLM. It’s a system architecture that combines: a language model context management memory tools guardrails orchestration logic The tutorial was designed to make that architecture visible.
Technology Stack
To build the tutorial environment, we combined several tools commonly used in modern AI systems like Agno (Agent Framework), OpenAI tokenizer tool, Arize Phoenix.
The core of the demo is built using Agno, a framework designed for building AI agents. Agno makes it easier to orchestrate: LLM calls conversation history session management tool execution In the tutorial we use Agno to create several progressively more capable agents: a basic stateless agent an agent with session history an agent with persistent memory This allows viewers to see exactly how each capability changes the model’s behavior.
In the demo, the LLM behaves exactly as a raw model would and each prompt is independent, there's no memory that is stored between calls, and the model simply predicts the next tokens.
Watch the Full Walkthrough
If you'd like to see the system in action, you can watch the full tutorial here.
Top comments (0)