Satya Nadella's recent warning about AI "hollowing out" industries has been sitting with me.
The line I keep coming back to is this: companies can outsource tasks, but they cannot outsource learning.
That is the bit that matters for most businesses. The question is not whether you have access to a good model. Everyone will have access to good models. The question is whether your business is getting better every time it uses them.
Most companies are still treating AI as a tool layer. Buy Copilot. Try ChatGPT. Add Claude. Run a workshop. Encourage the team to experiment. That is useful, but it is not much of an advantage on its own.
The advantage starts when the work creates a learning loop the business owns. A good sales call summary should improve the next call summary. A corrected proposal draft should make the next proposal better. A service issue that gets resolved should become part of the support system. A project debrief should change the delivery checklist. A senior person's review of an AI output should not disappear into chat history; it should become reusable judgement.
This is where the conversation gets much more practical.
In a trade business, that might mean capturing how the best salesperson qualifies a messy inbound lead, then turning those judgement calls into a reviewable process the rest of the team can use.
In a professional services firm, it might mean taking the partner's edits on proposals, scopes, and client updates and turning them into reusable rules, examples, and quality checks.
In a property business, it might mean preserving how an experienced agent reads buyer feedback, vendor risk, pricing pressure, and campaign momentum, so every vendor report gets closer to the standard of the best operator in the office.
In an operations team, it might mean every exception, correction, and "don't do it that way again" becoming part of the system rather than living in someone's head.
That is what I think people miss when AI strategy becomes a vendor conversation. Model access is rented capability. The learning loop is owned capability.
Nadella is making an ecosystem argument, and of course Microsoft has a position in that ecosystem. But the underlying point is still right. If a company's knowledge, workflow, feedback, and judgement are never captured in a form the business can reuse, then it is not building much of an AI capability. It is renting intelligence session by session.
The practical work is less glamorous than the headlines. Capture the judgement. Structure the workflow. Save the corrections. Build small evaluations around the work that matters. Make the next output better because the last one was reviewed properly.
That is where the value compounds: not in the model alone, but in the system the business builds around it.