Let me be direct: we are in a bubble. Not because AI is useless. The opposite is true. We are in a bubble because money is chasing narratives faster than the narratives can catch up with reality. I build with AI every day, at the eksperts, at HYPOTEQ, in the classroom at CAAIE, so this is not commentary from the sidelines. It is a builder's inventory.
The hard facts
Valuations are priced for perfection while most AI companies are a thin layer over someone else's model. Every second pitch deck is 'AI-powered' without a single number behind it. Pilots get announced loudly and buried quietly. And the chase narratives repeat on schedule: agents will replace everyone, AGI is next year, every existing software company is dead. None of these survive contact with a real production environment.What is actually real
The capability curve is real, and the cost of inference keeps falling. AI that sits inside an existing workflow, doing one job with clear inputs, clear outputs and a human who stays accountable, pays for itself fast. That part is not a bubble. That part is infrastructure. The value is in boring processes, not in demos.What builders should rethink
First: do not build a feature that the next model update deletes. Build where you own the workflow, the data or the relationship. My honest test: would your business still matter if the model were free and everyone had it? Second: measure. If you cannot state what the model saves or earns, stop. Third: charge for outcomes, keep humans accountable, and treat models as replaceable suppliers, because they are.Where I place my own bets follows the same logic. Document intelligence at DOCUVISIO, because the compliance workflow is the moat. Assessment support at HYPOTEQ, because the specialist owns the decision. Teaching at CAAIE, because integrating people is the real bottleneck. I write about all of this regularly, in German, for the Swiss market, at ejaj.ch.
Bubbles pop. Infrastructure stays. The dotcom crash deleted the tourists and left Amazon standing. Build like the correction is coming, because it is.