If public services are never really finished, why do we still organise so much change as though they are?

There is a distinction in government that feels increasingly important to examine: the difference between running something and changing it.

Operational activity tends to be treated as BAU. Transformation tends to be treated as a programme. The first keeps the service going; the second arrives with a defined objective, investment case, governance structure, benefits profile and an implied end point.

That makes sense when change is expensive, episodic and separable from the operation itself. But many modern services do not really work like that.

A digital service is never finished in the same way a building programme might be finished. User expectations move, security threats change, legislation changes, technology changes, data improves and the organisation learns more about how the service actually works. The natural state of a good service should therefore be continuous improvement.

We have governance for programmes of change. Do we have equally good governance for services that should never stop changing?

The problem is wider than AI, although AI makes it more obvious. If AI and better software tooling make small changes cheaper, faster and easier to test, then the economically sensible response is often not another large transformation programme. It is a permanent capability that can keep improving the service in small increments.

That can sit awkwardly with governance and funding structures designed around bounded interventions. A programme can explain what it will deliver, when it will finish and what benefits will be realised. A permanent improvement capability is harder to describe in those terms because the point is not to reach a final state. The point is to keep learning and improving.

This creates a potential incentive problem. If additional investment is easier to justify when it is packaged as a programme, organisations may repeatedly create temporary transformation structures around services that are actually permanent.

The pattern can become: transform, hand over to BAU, allow improvement capacity to fall away, then create another transformation programme when the gap becomes large enough.

That is not necessarily because anybody is making a poor decision. It may simply be what the governance system makes easiest to fund and explain.

A more useful question might be how government can retain accountability while funding enduring capability. That does not mean giving teams permanent budgets without scrutiny. Continuous improvement still needs evidence that it is creating value.

The distinction may be between governing a predetermined package of change and governing the capacity to keep identifying, testing and delivering worthwhile improvements.

AI increases the urgency because it makes incremental change cheaper. But the underlying question would exist without AI.

Perhaps government is better designed to approve change than to sustain improvement.

If that is true, the challenge is not simply to make programmes more agile. It is to ask whether the programme should exist in the first place.