Operating AI is also an organisational task
Sønderborg connects Copilot Credits, consumption costs, environment structure and governance. Roles and responsibilities need to develop alongside technical capabilities.
Read slide 42026-09-03Velatir · Odense
Hosted by Velatir in Odense · 3 September 2026
Presentations and documented discussions on AI operations, vibe coding, agentic engineering, KONTEKST, Agent Harness and modular AI.
From the material
Three things
to take with you.
Fast development needs lasting ownership
Gladsaxe raises the maintenance question for small AI-coded solutions. ITK describes a more systematic approach with tests, documentation and shared agent skills.
Read slide 6Bounded components clarify responsibility
Copenhagen’s modular approach and the Agent Harness proposal point to explicit boundaries between models, tools, data and control.
Read slide 13Bring it into practice
A place to begin.
- Identify the owners of cost, operations, security and development before moving beyond a prototype.
- Review a small solution from user need through maintenance, not just time to first demo.
- Use the harness architecture to structure questions about access, logging and control.
Still open
Questions to explore.
- How should credit and token consumption be allocated and tracked?
- When does a small local solution have a sustainable operating model?
- Which components and working practices could municipalities share?
Read the source material
The presentation.
