Thoughts on AI - to be refined afterwards

There is just too many different terms and concepts around this area.

LLM Learning & Mediums: LLMs don't have eyes. They don't have the mediums we have. It's of course difficult to solve problems when you need to convert them to a mathematical equation - if we had to solve image pattern recognition problems that way, we would be stumped as well.

GenAI Usefulness: A LOT of effort goes into the orchestration layer above an LLM - in our thinking, we tend to mix the LLM and the orchestration layer capabilities. Orchestration layer does a lot, and I'm afraid what companies like Celonis needs to build on is this orchestration layer, rather than the LLMs themselves. If Celonis tries to implement an orchestration layer on its own, we would be competing with the LLM providers - and they will surpass us. So, no Process Copilots, no Annotation Builders, yes MCP, yes, integrations to orchestration layers provided by LLM providers (AWS, Azure) and yes to all sorts of making the process context and insights available.