In a large organisation, somewhere in the middle of its structure, a team is preparing a presentation on its AI progress.
The slides are well-constructed. There is a section on tools adopted over the past six months: a language model integrated into internal workflows, a coding assistant rolled out to engineering, a third-party platform for document analysis. There are pilot programmes underway in operations and customer support. Early metrics are included, cautiously framed but positive.
There is also a roadmap. More integrations are planned. Additional vendors are being evaluated. Budget has been allocated for expansion.
What is less clear, even to the people presenting, is what has fundamentally changed.
The same decisions still require the same approvals. Work moves through the same stages. Outputs may be generated faster, but they are reviewed, revised, and escalated in much the same way as before. The organisation is, undeniably, doing more with AI.
It is less clear that it is doing things differently.
This is not an unusual situation. It is, increasingly, the norm.
A strategy, or a response to pressure
It is worth asking why this confusion has become so widespread.
Part of the answer lies in timing. The current wave of AI capability has arrived quickly, and with a degree of public visibility that few technologies have matched. The pressure to respond is not only operational, but reputational. To appear inactive is to risk looking unprepared.
Under those conditions, action becomes a signal.
Buying tools, launching pilots, and forming partnerships are all visible, legible responses. They demonstrate engagement. They can be reported upwards and outwards. They create the impression of forward motion in a situation where the direction of travel is still unclear.
Strategy, by contrast, is slower to form and harder to communicate. It requires a view not just of what is possible, but of what matters. It demands choices that may not yet have consensus.
So organisations respond first, and define later.
The expanding layer of AI on top of unchanged systems
Inside many organisations, AI is not replacing existing systems. It is being added to them.
A new tool is introduced into a workflow that was not designed for it. Outputs are generated more quickly, but still pass through the same approval structures. Insights are surfaced earlier, but decisions are made at the same cadence. The system becomes more capable, but not fundamentally different.
Over time, this creates a layered architecture.
At the surface, there is evidence of modernisation: AI-assisted drafting, automated analysis, conversational interfaces. Beneath that surface, the underlying logic of the organisation remains largely intact. Responsibilities are unchanged. Incentives still reward the same behaviours. Risk is managed according to frameworks built for a different technological baseline.
The result is a kind of partial adaptation. Enough change to demonstrate progress, not enough to alter outcomes in a sustained way.
The accumulation problem
As tools proliferate, a second issue begins to emerge.
Each new system introduces its own interface, its own assumptions, and its own demands on attention. Teams are asked to learn, evaluate, and integrate multiple AI-enabled tools, often without a clear model of how they relate to one another.
What begins as capability expansion becomes cognitive load.
Different parts of the organisation adopt different tools for similar tasks. Outputs vary in format and reliability. Informal workarounds appear. Knowledge fragments. The organisation becomes more capable in aggregate, but less coherent in practice.
This is rarely framed as a strategic issue. It is treated as an inevitable side effect of experimentation.
But over time, it becomes a constraint.
Why procurement persists
None of this is accidental.
Procurement fits the existing shape of organisations. It aligns with budgeting cycles, governance structures, and accountability models. It allows leaders to act without immediately confronting deeper structural questions.
It also distributes responsibility.
If a tool underperforms, responsibility can be shared with the vendor. If a pilot fails, it can be contained. The risks are bounded, at least in the short term.
Structural change offers no such containment.
Redesigning workflows requires coordination across functions. Changing decision rights can create internal friction. Embedding AI into core processes exposes the organisation to new forms of error, which must be owned rather than outsourced.
Seen in this light, the persistence of procurement-led “strategy” is not surprising. It is the path of least resistance through a complex landscape.
Where leverage is actually created
And yet, the organisations that are beginning to extract real value from AI look different.
They are not defined by the number of tools they use, but by the degree to which those tools have been absorbed into how work is organised.
In these environments, AI is not an add-on. It is part of the workflow itself.
Tasks are redefined rather than accelerated. Steps are removed rather than optimised. The boundaries between roles shift, as some forms of judgement become less scarce and others more important. In some cases, entire processes are restructured around the assumption that certain types of intelligence are now continuously available.
These changes are not always visible from the outside. They do not lend themselves to simple announcements. But they alter the internal economics of the organisation in ways that tool adoption alone cannot.
The question of judgement
One of the less discussed consequences of AI adoption is the changing role of human judgement.
As systems become more capable of generating outputs — text, analysis, recommendations — the bottleneck shifts. The question is no longer whether something can be produced, but whether it should be used.
This requires a different kind of capability.
Organisations need to develop shared standards for evaluating AI outputs. They need clarity about when to trust the system, when to verify it, and when to override it. They need to decide who is accountable when AI-informed decisions lead to negative outcomes.
These are not procurement questions. They are strategic ones.
Without answers to them, the presence of AI increases ambiguity rather than reducing it.
Strategy as constraint
A genuine AI strategy, if it exists, is often defined more by what it excludes than what it includes.
It identifies a limited number of areas where AI will be applied deeply, and accepts that elsewhere, change will be slower or absent. It aligns resources, incentives, and attention around those areas. It creates coherence where there might otherwise be diffusion.
This kind of focus can be uncomfortable, particularly when the surrounding narrative emphasises speed and breadth. But without it, organisations tend to spread effort thinly, accumulating tools without building momentum.
Constraint, in this context, is not a limitation. It is a prerequisite for impact.
What happens next
The current phase of AI adoption is still shaped by access.
Having the tools, demonstrating usage, showing engagement — these are still treated as indicators of progress. They matter, but they are temporary advantages.
As AI systems become more widely available and more standardised, access will cease to differentiate. The baseline will rise. What was once notable will become expected.
At that point, the focus will shift.
The organisations that have treated AI as a layer on top of existing systems will find that their gains plateau. The organisations that have used AI as a prompt to rethink how they operate will continue to compound.
The difference between the two will not always be obvious in language. Both will continue to talk about strategy. Both will point to investments and initiatives.
But over time, the outcomes will diverge.
And the gap between procurement and strategy — easy to ignore in the early stages — will become one of the defining lines between those who adapted, and those who simply accumulated.