Articles

If AI Solves Everything, Who Answers at 3 A.M.?

I promised myself I wouldn't write an article about AI. The hype passed my tolerance threshold some time ago. Then I noticed that the sentence I hear today is one I've heard before: now we can do it without IT.

I heard it with self-service reporting and with Power Apps, and I hear it again with artificial intelligence. It isn't entirely wrong, which is exactly what makes it dangerous.

We've seen this before

When self-service Business Intelligence, Power BI and Power Apps arrived, the promise was clear: every function could build its own report, its own workflow, its own small application, without waiting for IT. In many cases it worked, and in some it created real value.

Then came the consequences. Duplicate dashboards defining the same number in different ways. Applications built for a local need and left without an owner. Licenses, consultants, hours of development: a series of expensive little toys, each reasonable on its own.

And while lightweight tools multiplied on top, the heart of the company stayed where it was. In many organizations it was SAP. In more traditional ones, a mainframe. The storefront had changed; the engine hadn't. The tools were never the problem. The problem was believing the facade could replace what stood behind it.

What IT strategy actually means

In a large organization I once worked in, the function that managed IT demand, and had a role in digital transformation, sat inside HR. As I've written before, where a function sits on the org chart is rarely the real issue. The issue is whether it has the competence the role requires.

The deeper misunderstanding was about what "IT strategy" means. It was treated as a list of business requests, ordered by priority. But an IT strategy is something else: architecture, risk, capacity, deciding what to build and what to buy, the cost of keeping what already exists, the evolution of the core. Reduce it to a portfolio of requests and you produce, very neatly, the sum of everyone's requests.

It is the same mistake we are repeating with AI, in new clothes: believing IT is just the one who executes, and that if someone or something executes faster, IT is no longer needed.

Where AI genuinely helps in IT operations

AI in operations is a real opportunity, and it helps with concrete things:

  • Cutting noise by correlating alerts to see which ones are the same problem.
  • Summarizing an incident for whoever joins a war room late, without making everyone repeat themselves.
  • Suggesting runbook steps to whoever is investigating.
  • Classifying and routing tickets, and finding similar past cases in existing knowledge.
  • Drafting a first post-mortem.

Think of the fifty-eight people on a call, two of whom were actually solving the problem. A good use of AI might help find those two faster.

AI removes noise, not responsibility.

Where the reasoning breaks

AI amplifies the state of your data. If the inventory of systems is incomplete, if ownership is unclear, if documentation lives in a few people's heads, AI will give confident, wrong answers. A system that isn't explicit enough stays that way; AI just makes it wrong faster.

Debt arrives faster. If anyone can have an application generated, we are back to a world where anyone can ask and no one is accountable, only at much higher speed. Every generated application is a promise of maintenance, and with AI promises are made more easily and without noticing.

Who answers when an automated action breaks production at night? That is not a technical question; it is a question of accountability. If the answer is "I don't know", the automation is not ready.

Knowledge hollows out. If nobody understands how the system works, nobody notices when AI is wrong, or can fix it. It is the mirror image of the CIO who still needs the Java: not the leader who dives too deep, but the organization where nobody knows the detail anymore.

Data and security. Tools used without rules carry information where it shouldn't go. Someone has to be responsible for governing that.

The core is still the core

As with BI, AI usually lives on top of something. Underneath there is still SAP, a mainframe, systems decades old, and they carry a cost the income statement pays every year, even if no balance sheet line carries their name. AI doesn't erase that. If anything it makes it more visible, because whoever wants to use it on their own data quickly discovers what state that data is in.

Questions before saying yes

  • What data does it rest on, and who is responsible for it?
  • Who answers for an automated action that goes wrong?
  • What happens when it is unavailable?
  • Who will maintain it in three years?
  • What real problem does it solve, and how do we measure it?

AI doesn't remove the need for an IT function: it raises the cost of doing without one.

Domenico Sannelli
Domenico Sannelli

Technology operations leader and author of Never Down. I write about running critical services while continuously changing the organization behind them. Connect on LinkedIn.