AI won't fix bad operations

There is a dangerous seduction in the idea that the right technology solves any problem. Every time a new generation of tools appears, the same promise resurfaces: install this, connect that, and the mess disappears. With artificial intelligence, that promise got even louder — and even more misleading.

The problem is simple to state and hard to accept: automation amplifies the system that already exists. If the system is good, it makes it faster. If the system is bad, it only makes the mistakes happen sooner, with more confidence and at greater scale.

What automation actually does

Before automating a system, you need to understand why it works the way it does. Processes that look irrational almost always carry a hidden logic: a team constraint, an informal agreement, a scar from an old incident. When you automate without seeing that logic, you freeze decisions no one had a chance to revisit.

The tool has no opinion about the quality of the process it runs. It just runs it — faster.

That is why mature operations treat AI as the last step, not the first. First you make the process visible. Then you make the process simple. Only then does it make sense to make it fast.

A concrete example

Imagine a ticket triage flow where priority is set by unwritten conventions. Automating it without documenting it produces something like:

function prioritize(ticket) {
  // rule nobody documented
  if (ticket.client === "strategic") return "P0";
  return classifyWithAI(ticket); // amplifies the bias above
}

The model inherits the exception hard-coded in the first line and starts applying it to thousands of cases a day. What was an occasional shortcut becomes a silent policy. No one decided this — the system decided by omission.

Where to start, then

Start with clarity, not with speed. Map the process as it is, not as you wish it were. Find the points where decisions are made and ask who makes them and based on what. It is uncomfortable, and it usually reveals that half of the “transformation” work is really organizing work.

AI is still extraordinary. But it is a multiplier, and multipliers have no signal of their own. They inherit the signal of what you have already built.