Two harsh truths about AI adoption
Those who have seen little to no productivity gains with AI share the same two problems.
Problem 1:
They spent millions before understanding what they were building. The infrastructure wasn’t ready. The use cases weren’t defined. They bought technology looking for a problem instead of identifying a problem and finding the right solution.
Problem 2:
The projects are run by teams that don’t understand what generative AI actually requires.
These aren’t bad teams.
They’re experienced.
They’ve successfully deployed technology for decades.
But they’re applying old frameworks to something fundamentally new.
When you manage AI implementation like a traditional systems upgrade,
You default to re-engineering what exists.
You bolt chatbots onto legacy platforms.
You add agents to workflows that weren’t designed for them.
And you end up with expensive infrastructure that doesn’t function.
The institutions getting this right started differently. They built data governance first because no bank will use AI unless they control it.
They created data marketplaces that connect to legacy systems without requiring those systems to change.
They designed for a future where there will be billions of models and agents, not just the handful available today.
