From state of the art to legacy in months, not decades

Last week left me reflecting on a phrase I have heard all my working life, but which now carries a very different meaning.

“Back in the day.”

At the start of the week, I was asked to write about the role telecoms operators could play in addressing the growing problem of ageing infrastructure. Some of that infrastructure dates back decades. It was described to me, quite rightly, as state of the art when it was built. Today, it is expensive to maintain, difficult to upgrade, and increasingly an obstacle rather than an enabler.

Poles, wires, switching systems, layers of investment built over time. Each generation solving the problem of its day, and quietly creating the constraints of the next.

A few days later, I found myself in a very different conversation with a client about AI. And the same phrase came up again.

“Back in the day.”

Except this time, we both knew that “the day” in question was not ten or twenty years ago. It was three months ago.

The models had moved on. The capabilities had shifted. What felt cutting edge at the start of the quarter now felt dated. Not broken, not obsolete in the traditional sense, but already behind the curve.

This is the shift leaders are struggling to fully grasp.

We are used to thinking about infrastructure as something physical and long lived. Telecoms networks, transport systems, energy grids. They require capital, planning, regulation, and they decay slowly. The problem is visible. The cost of maintaining legacy becomes obvious over time.

Software feels different. It feels fluid, intangible, easy to change. There is a tendency to believe that because it is not bolted to the ground, it does not create the same kind of legacy.

That is a mistake.

AI, cloud platforms, data architectures, integration layers, these are all forms of infrastructure. They may not sit on poles and wires, but they underpin how the organisation operates. They shape what is possible, how quickly you can move, and what it costs to do so.

And crucially, they age.

Not over decades, but over months.

What was state of the art in AI three months ago can quickly become sub optimal. Not because it has stopped working, but because something better, faster, more efficient, or more capable has emerged. The gap between what you have and what is now possible starts to widen almost immediately.

If you are not careful, you recreate the same problem the telecoms operators are now facing, just at a different speed.

You build layers. You integrate tools. You embed processes. You optimise around what you have. And before long, you are carrying complexity that is expensive to maintain and difficult to unwind.

The difference is that this time, you do not have decades before it becomes a constraint. You may have quarters.

So the question is not whether legacy will exist in the AI era. It will.

The question is how to avoid repeating the mistakes of the past.

The first shift is mental. Leaders need to stop thinking in terms of permanent solutions. There are no permanent solutions in a world where the underlying technology is evolving this quickly. Every decision about platforms, models, and architecture should be treated as time bound.

Not in a vague sense, but explicitly. How long do we expect this to be fit for purpose? What will trigger a rethink? What is our exit route?

Telecoms infrastructure was often built with the assumption of longevity. That made sense in a slower moving world. In AI, that assumption becomes a liability.

The second shift is architectural. The organisations that will cope best are those that design for change rather than stability. Modular approaches, clear interfaces, avoiding deep coupling between systems. This is not new thinking, but it is now non negotiable.

If your AI capability is tightly bound into core systems in ways that are hard to disentangle, you are effectively pouring concrete around something that will need to move.

The third shift is economic. Leaders need to rethink how they evaluate investment. In the past, large capital investments were justified over long time horizons. In AI, the value curve is shorter and steeper. The question is less about long term depreciation and more about speed of return and flexibility.

It may be entirely rational to invest in something that you expect to replace within a year, if it delivers immediate competitive advantage. But only if you have planned for that replacement.

Finally, there is a leadership shift. The hardest part of all.

Letting go.

One of the reasons legacy builds up is not just technical, it is human. People become attached to what they have built. Systems become embedded in processes, in incentives, in identity. The organisation resists change not because it cannot change, but because it does not want to.

In a world where “back in the day” can mean three months ago, that mindset is fatal.

Leaders need to normalise constant renewal. To make it acceptable, even expected, that what we built recently may already need to evolve or be replaced. Not as a failure, but as a consequence of operating at the edge of change.

The lesson from telecoms infrastructure is not simply that old systems become expensive and restrictive. It is that they were built for a different pace of change.

AI is forcing us into a different rhythm entirely.

Poles and wires took decades to become legacy. Software will do it in months. The organisations that understand this, and design for it, will stay ahead.

Those that do not will find themselves, sooner than they expect, maintaining a very modern version of the same old mess.