Soundings
Our Perspectives and Insights
AI Is Not a Tool. It’s a Culture.
Walk into most companies right now and you can find AI in exactly one place: the technology department. It has a budget line, a steering committee, and a corner of the org chart down in IT, where it sits like another server to be maintained. That instinct, to treat AI as a system you install and a function you assign, is the most expensive mistake leaders are making this year. AI is not a tool you bolt onto the org chart. It is a new way of working, and a new way of working should not stay in one room.
I had my own inflection point on this a few years ago, taking questions in an ask-me-anything with the engineering department. One engineer asked if they could use AI to write code. My answer was an emphatic yes: every one of you should be using it, because it makes you more efficient and it is genuinely a new way to work. And in the same breath: if anyone ships AI-generated code out the door without a human validating it first, you and I are going to have a very different, job-related conversation. I was a vocal proponent of the tool and just as adamant about keeping a human in the loop. Both were true then. Both are more true now.
Why “Where Does AI Live?” Is the Wrong Question
Back then the question was about code. Today AI reaches far beyond it, into finance, marketing, hiring, and strategy. So when a leader asks which department should own AI, they have already made the mistake. They are treating AI like a CRM or an ERP: an application, so hand it to the people who run applications. But AI is not another piece of software with a login and an admin. It changes how decisions get made, what you trust, and how you check your work. Quarantine that inside IT and you have capped its value at whatever one department can imagine and validate on behalf of all the others. You wired the whole house to a single outlet and then wondered why the rest of the rooms stayed dark.
Electricity Didn’t Stay in the Generator Room
There is a useful piece of history here. When factories first electrified, they did the obvious thing: they ripped out the big central steam engine and dropped in one big electric motor in the same spot. The gains were small. For decades, almost nothing happened to productivity. The leap came only when companies stopped treating electric power as a part you swap in and started treating it as something you distribute. Put a small motor on every machine, redesign the whole floor around that freedom, and output exploded. The technology had been available for years. The reorganization was the actual innovation.
The computer told the same story. It started life in a glass room, tended by specialists, run as a department. Its value did not show up until it left that room: until finance got the spreadsheet, HR got the system of record, marketing got the database, and the factory floor got the controller. Nobody today has a “computing department” that owns all the computers. Computing became literacy: a basic skill every function was simply expected to have.
A general-purpose technology kept in one department is a contradiction in terms.
Andrew Ng put it plainly: AI is the new electricity. The point of the line is not that AI is powerful. It is that AI, like electricity, is a general-purpose capability whose entire promise is that it reaches everywhere. A general-purpose technology kept in one department is a contradiction in terms.
The Value Lives in the Other Departments
Here is the part the quarantine model misses entirely: the return on AI is almost never in IT. It is in the functions IT serves.
It is in a finance team that closes the books faster, forecasts with more confidence, and catches an anomaly before it becomes a write-down instead of after. It is in an HR team that screens more fairly, onboards in hours instead of weeks, and sees an attrition risk while there is still time to act on it. It is in legal reading every contract instead of a sample, in operations seeing a supply shock forming, in sales walking into the room already knowing the answer. None of that value is realized by the department that stood up the model. It is realized by the people doing the work.
If AI lives only in technology, you have built an impressive cost center. The transformation is happening in the rooms you locked it out of.
If AI lives only in technology, you have built an impressive cost center.
A Different Kind of Trust
There is a deeper reason AI cannot be managed like a tool, and it is the reason this is finally a culture question and not a technology one. Ordinary software is deterministic: you press the button, you get the answer, you trust the output because it is the same every time. AI is probabilistic. It is confident, fluent, and sometimes wrong.
This is exactly why the validation cannot live in IT. Technology can tell you the model is running. It cannot tell you the answer is right, because right is defined by the business, not the code. Only the finance lead knows the forecast smells wrong. Only the recruiter knows the shortlist is off. Only the operator knows the plan would never survive contact with the floor. The people closest to the work are the only ones who can judge whether an output has earned trust, which means the functions you quarantined AI away from are the exact ones who should have been holding it to account. Leave a confident model running unwatched in a domain no qualified person is checking, and you do not get efficiency. You get fast, fluent, well-formatted mistakes produced at scale, surfacing late, and reined in after the damage instead of before.
So the new habit has to spread to every function: take the speed, keep a human validating the direction. That habit cannot be installed. It has to be learned, modeled, and rewarded, by people who understand both the work and the limits of the machine. And the frontier keeps moving. What the tools can do this quarter is not what they could do last quarter. An organization that has quarantined AI has also quarantined its ability to keep up, because adaptation is sitting in one department while everyone else waits to be handed a solution. The companies that win will be agile everywhere, because the edge they are chasing moves everywhere.
A Framework: Getting AI Out of Quarantine
If you want AI to behave like a culture and not a tool, five moves get you there.
- Build literacy, not a department. Your goal is not a center of excellence that owns AI. It is a finance analyst, a recruiter, and an operations lead who each know how to use it in their own work. Spread the capability; do not concentrate it.
- Give ownership to the function, not to IT. Technology should provide the rails: access, security, guardrails. The use cases, and the accountability for validating them, belong to the people closest to the problem.
- Teach the new trust. Make human validation a stated expectation, not an afterthought. Take the acceleration, verify the heading. Reward the person who catches the confident mistake.
- Redesign the work, don’t bolt AI onto it. The electrified factory only paid off after the floor was rebuilt. Do not point AI at last year’s process and call it transformation. Ask what the process would look like if it had been built around the capability from the start.
- Make adaptability the standing skill. The frontier moves. Build a company that expects to relearn its tools every few months, in every function, rather than one that waits for the AI department to circulate a memo.
The Bottom Line: Wire the Whole Building
AI is the most general capability most organizations will ever adopt, and generality is the whole point. Treat it as a tool and you will get a tool: one outlet, one department, one narrow return. Treat it as a current, something you run through finance and HR and operations and sales and every decision in between, and it becomes what it actually is, which is a new way the company works.
Electricity did not change the world from the generator room.
Electricity did not change the world from the generator room. It changed the world when it reached every machine in the building. AI is no different. The only real question is whether you are going to wire the whole building, or guard a single socket and wonder why nothing downstream ever changed.