AI & Automation
10 essays on ai & automation.
-
The model won't tell you it's guessing
Most models answer with the same tone whether they're certain or not. If you care about reliability, you have to build the signal yourself.
-
When your embeddings stop understanding your documents
Embedding model choice feels like a one-time decision. It isn't. Domain language drifts, and your model doesn't follow it.
-
What chasing the top-ranked model actually costs
Moving to the new leaderboard leader every six months slowly eats most of the savings you wanted, and nobody tracks the bill.
-
What a leaked system prompt says about yours
A leaked system prompt is a good reason to check if your own production prompts have an owner, a version history, or any review.
-
The unglamorous part of sovereign AI
National language-model projects usually stall on who owns the training text and who can allow its use, not on compute or architecture.
-
What your AI vendor can change without telling you
When you license a model, the vendor often keeps the right to change how it behaves, with little duty to tell you first.
-
Most of our LLM failures were predictable
When an LLM fails in production it is usually a known failure mode we did not plan for, not the model being mysterious.
-
Approval buttons that don't show you the decision
Most AI agent approval dialogs hide the data, scope, and reversibility you need to judge the action before you click approve.
-
What employees are really objecting to
When teams resist a new AI rollout, the real problem is usually how it was announced, not the software itself.
-
Der Nutzen von Change-Management: KI-Tools Mitarbeitenden näherbringen
Wie Change Management dabei helfen kann, Mitarbeitende für den Einsatz von KI-Tools zu befähigen – und was dabei wirklich zählt.