AI's default interface is a chat box. That's a demo, not a system. The model is stateless, so you re-explain everything, and judgement stays with you while the tool forgets. The result is impressive one-offs and no compounding leverage.

This is a working session. You leave with an AI system installed and understood — not a slide deck.

We start with why: the results this stack produces when it is wired as a system instead of a chatbot. Then the machine, in plain terms: text in, LLM, text out; the harness around it; and the real bottleneck, which is low judgement, not missing prompts.

From there we auto-setup so nobody is stuck on install, and we walk the architecture by using it:

• Entry point — how work actually gets into the system

• Routing — how it decides what to do

• Memory — markdown files for durable knowledge, SQLite for structured state

• Skills — procedures the system can load and follow

• Connectors and MCP — talking to real tools, not just generating text

• Self-improvement — the system getting better from use

• Automation — work that runs without sitting in a chat window

Objectives: replace "prompting" as the mental model; install a working system on your machine; understand each layer well enough to extend it; ship one automation the same day.

Impact: participants stop treating AI as a novelty tab and start running real work through a system they own.

Practical details: full day, 10:00–17:00; English; cap 16; €397 per person; laptop required. Hosted by André Casal (Casal Innovations, Lda).