Irrespective of how code is created, whether it is generated by AI or written by hand, there are three kinds of automation:
- Deterministic Automation: Whether it is imperative or declarative code, the same input and conditions yield the same output. What we've always done. Familiar, predictable and reliable, but not necessarily flexible.
- Generative Automation: The instructions live completely in the AI model as a recurring or ad-hoc task, ideally complemented with RAG, MCP connections and skills. Results may vary, as instructions are never completely deterministic.
- Hybrid Automation: Deterministic automation that calls generative automation when flexibility is required and/or when the result may not be known in advance.
Generative automation is relatively easy to implement, but comes at the cost of variable results. Workflows for deployments and versioning of instructions are under development, but are not as mature yet as they are for traditional code.
Hybrid automation is in my opinion the most powerful of the three. It can use AI to assist the human operator with new insights when making decisions or use probabilistic models to make decisions for the human operator, given specific confidence levels.