The core idea behind all of it: use AI to extend judgement and capability, while keeping people responsible for goals, evidence, verification, and consequences. Everything below is a detail of that sentence.
What we covered
Eight topics from the two sessions. Each one has a page behind it if you want the longer version, and nothing here requires you to have remembered the workshop.
-
A language calculator
Models break inputs and outputs into tokens. Token counts are a practical measure of how much material a model processes, and often of how usage is priced.
-
Ethics
Before using a model or an agent, ask what data leaves your control, who owns the material, what requires consent, what could cause harm, and what must be verified by a person.
-
The ecosystem
No single model or provider is "AI". Models, providers, interfaces, tools, and data are separate layers, and knowing the layers makes switching tools easier and reduces lock-in.
-
Harnesses and agents
A harness gives a model access to context, files, tools, and actions. An agent uses those capabilities toward a goal.
-
Open source
You rarely need to reinvent the wheel. Open projects let you inspect, adapt, and combine existing work, while supporting digital sovereignty and autonomy.
-
Software development principles
Plan before building. Make small changes. Keep documentation current. Test as you go. Use revision control so experiments are visible and reversible.
-
Terminal and environments
The terminal is a direct interface to the computer. Development is where you experiment safely. Production is the live system that people or organizations depend on.
-
Case examples and brainstorming
Start with a real problem, not a technology. Look for what already exists, define the smallest useful outcome, and explore farm, research, and organizational workflows that agents could improve.
Five habits to remember
- Start with the problem, not the tool.
- Plan first, before anything is built.
- Build small, so something works early.
- Verify the work, facts and function alike.
- Keep a history, so experiments stay reversible.
How the two sessions fit together
The first session was mostly about understanding: what the AI ecosystem actually contains, how models handle language, what open source offers, and how a model becomes an agent that can do things.
The second session was mostly about working: planning and documentation, the terminal, the difference between a development environment and a live system, and a long list of practical things worth building. It ended with brainstorming, which is where most participants found their first project.
Both decks are on the workshop slides page if you want to go back to the framing.
If you only keep one thing
Agents can act, but people still set goals, choose evidence, approve consequences, and remain accountable.
That is the sentence the rest of this resource is built around, including the parts about checking the work.