Agentic AI: Beyond Coding Assistants and Chatbots

This hands-on workshop, organized by LUMI AI Factory and Sweden AI Factory, teaches you to build AI agents that can act, not just answer. We build agents in stages, starting with a plain OpenAI REST API call and ending with multi-agent workflows in Pydantic AI. The models run on the LUMI supercomputer and are served by the Aitta inference service. Our running example is a flight-booking assistant.

Overview

• LLM basics and inference: how models work at a practical level, serving models with the Aitta inference server, and using prompt templates.
• Building agents from scratch: from a single REST call to structured agentic workflows with Pydantic AI.
• Components of an agent: the model, the harness, the loop, tools, MCP servers, skills, memory, and how they fit together.
• Security and evaluation: what breaks when agents leave the laptop, how to evaluate them, and how to run them reliably.

Key takeaways

• Identify the main components of an AI agent (model, tools, memory, harness) and recall standard concepts such as tool calling and MCP.
• Build working agents in stages, from a raw OpenAI REST API call to structured agents with Pydantic AI, and connect tools, MCP servers and memory to them.
• Analyze and identify potential pitfalls when moving agents from development to production.

Who is this for?

• IT professionals, software engineers and data scientists
• Programmers curious to move out of a chat interface and integrate AI in their routine tasks
• People who have used LLM chatbots or coding assistants and want to understand how they work behind the scenes

Prerequisites

• Prior programming experience in Python; no machine learning background is needed
• Basic use of a terminal and a code editor
• A LUMI account that is a member of any LUMI project, used to log in to Aitta