For a while, AI conversations were mostly about possibility. 

Look what this can do. Look how fast this is. Look how impressive the demo is. Look how easily we can summarize, generate, search, classify, automate, answer, translate, rewrite, analyze, and probably make coffee if the prompt is dramatic enough. And yes, many of those demos are impressive. 

But at some point, every organization has to move from “look what AI can do” to “how do we make this work safely, reliably, and usefully in our environment?”. That is where things become interesting. And expensive. 

Not always expensive in licensing costs, although let’s not pretend those do not matter. Expensive in attention, governance, data quality, architecture, ownership, change management and in all the things that are usually invisible in a demo but very visible in production. 

This is why I think the AI conversation is entering a much more serious phase. The question is no longer “Can we build something with AI?” Of course we can. The tools are getting better, the entry barrier is lower, and many teams can produce something impressive quickly. The better question is: can this survive contact with the organization? Because organizations are messy. 

Data lives in too many places. Permissions were “temporarily” adjusted three years ago and then became business critical. Content is outdated, duplicated, badly named, or stored in someone’s personal OneDrive. Processes are only documented in the memory of two people who are both too busy. Security teams have valid concerns. Legal teams have valid concerns. Users have expectations shaped by consumer tools. Leadership wants value, preferably yesterday. 

And somewhere in the middle of all that, someone says: “Can we just add AI?” This is where many AI initiatives become uncomfortable. Not because AI is useless, but because AI does not magically fix weak foundations. It often exposes them. If your content is messy, AI will not politely ignore the mess. If your permissions are wrong, AI will not make them morally better. If your process is unclear, AI may automate the confusion faster. If nobody owns the data, the model, the output, the escalation path, or the business decision, then you do not have an AI strategy, but a beautifully modern accountability gap. 

This is also where the ESPC programme becomes relevant in a very practical way. Because the sessions are not just about “AI is exciting”. They go into the surrounding disciplines that determine whether AI creates actual business value or simply more digital clutter. 

AI is moving from experiment to infrastructure, and infrastructure needs boring things: Ownership, governance, monitoring, security. Documentation, architecture, support models, training and feedback loops. A realistic understanding of constraints. 

The boring things are not the opposite of innovation. They are what make innovation survive. I still think AI is one of the most important shifts in our industry. But I do not think the organizations that win with AI will be the ones with the loudest demos. 

They will be the ones that learn how to make AI work inside real organizations, with real data, real users, real risk, real constraints, and real accountability. Luckily, the ESPC programme reflects that with lots of sessions that skill you up so you can operationalize AI in your organization.  

Can’t wait to see you in Amsterdam. Will you join us? 

Luise Freese