Interview Question: Explain why AI strategy should not start with tools.

Because adopting AI tools is not the same thing as having an AI strategy.

Over the past few weeks I’ve been asking ChatGPT to generate a series of interview-style questions designed to make me think more deeply about topics I care about. The exercise was intended as self-reflection, but I found that many of my answers naturally evolved into short essays. Rather than leave them buried in my notes, I’ve decided to publish them as well.

The Questions:

Explain why AI strategy should not start with tools.

Talk me through what “responsible AI adoption” actually means in a real organisation.

My answer:

These are really two versions of the same question.

If you ask why AI strategy should not start with tools, you quickly end up talking about what responsible AI adoption looks like in practice. And if you ask what responsible AI adoption means, you very quickly discover that it cannot begin with a scramble to buy the latest product.

The AI field is moving incredibly fast, and that creates a very particular kind of confusion.

Because the technology changes so quickly, the most visible part of the conversation is usually the tooling. One week it is a new model. The next it is a new agent framework, orchestration layer, coding assistant or enterprise platform. Organisations can very easily mistake participating in that conversation for having a strategy, but they are not the same thing.

Buying AI tools is an implementation choice. Strategy is something deeper. Strategy should explain what problems matter, what outcomes are being pursued, what constraints apply, and how a capability will be developed over time. Tools may support that strategy, but they are not a substitute for it. That distinction matters because the tools are the most visible part of the picture, but they are also the part most likely to change.

The excitement is real, but so is the distraction

I understand why organisations start with tools. Tools are tangible. You can demo them. You can compare feature lists. You can procure them. You can announce that something is happening, that creates momentum, which is not nothing. The problem is that it can also create the illusion of progress.

An organisation can license a chatbot, run a pilot, stand up a proof of concept, or buy access to a fashionable platform and still be no closer to understanding where AI genuinely fits into the business. In that case the technology becomes a kind of theatre: impressive enough to generate interest, but not grounded enough to create durable value. I think this is one of the biggest traps in AI adoption.

The newest thing always looks strategically important when viewed in isolation. But if you do not know what you are trying to improve, what decisions need to be better, what workflows need to change, or what operating risks need to be controlled, then tool selection becomes a guessing exercise with a budget attached to it.

Strategy should begin with the work, not the product

In my view, successful AI adoption starts much earlier. It starts with questions like:

• What problems are we actually trying to solve?
• What outcomes would matter if we improved them?
• Where is the organisation currently losing time, quality, insight or consistency?
• Which of those problems are genuinely suitable for AI?
• What would responsible adoption look like in this particular context?

Those questions are less glamorous than debating models and vendors, but they are far more important.

Once you start there, the conversation changes. AI stops being a generic innovation initiative and becomes part of a wider operating problem. It becomes tied to service design, decision quality, workflow fit, risk appetite, ownership, and the practical realities of the organisation. That is a much healthier place to begin.

Foundations matter more than model choice

If the underlying foundations are weak, even excellent tools tend to become expensive experiments. For most organisations, the real prerequisites are not mysterious:

• trustworthy data
• proportionate governance
• clear ownership
• realistic use-case selection
• an operating model that allows adoption to happen responsibly
• enough organisational learning capacity to adapt as the technology changes

Without those things, the usual pattern is predictable. A team buys or pilots a tool, gets some promising results in a narrow setting, then struggles to scale the work because the surrounding organisation is not prepared to absorb it. Questions about data quality, accountability, security, procurement, integration, and business ownership arrive later, when they are harder and more politically awkward to answer.

At that point, the problem is no longer “Which tool should we use?” The problem is “Why did we try to build capability on top of foundations we had not prepared?”

Responsible adoption should be phased and intentional

This is where the second question comes into view more directly.

Responsible AI adoption, at least in a real organisation, should be phased, intentional, and tied to solving genuine problems. It should not be driven by the vague belief that AI must be introduced everywhere as quickly as possible.

Every AI initiative should begin with a clearly defined problem, an intended outcome, and some way of measuring whether anything has actually improved. Otherwise, it becomes very difficult to distinguish meaningful capability-building from expensive experimentation.

For most organisations, the sensible place to start is with relatively low-risk, high-value use cases. Automating repetitive administrative work is often a good example. If AI can remove hours of manual effort each week, people can spend more of their time applying judgment, creativity and domain expertise where those things matter most.

That kind of early success is valuable for reasons beyond the immediate efficiency gain. It helps build confidence. It develops internal capability. It gives the organisation a safer place to learn what good adoption looks like before moving on to more ambitious use cases.

That is a much stronger pattern than trying to begin with the most advanced or politically exciting application first.

Stronger organisations become less tool-dependent

One of the interesting side effects of building the foundations first is that the tooling decision often becomes less dramatic. That might sound counterintuitive, but I think it is true.

If an organisation has good data discipline, sensible governance, clear ownership and a practical operating model, then it is much easier to change tools later. You are not relying on one vendor to compensate for structural weaknesses elsewhere. You are choosing products inside an organisation that already knows how to adopt them. That creates a very different kind of resilience.

If a stronger model appears next year, which it almost certainly will, or if a platform shifts direction, raises prices, or turns out to be a poor fit, you can adapt without having to rethink your entire strategy. The strategy is anchored in the capability you are trying to build, not the product you happened to choose first. That is one of the reasons I dislike tool-led AI strategy. It often produces dependency before understanding.

I am sceptical of always choosing the newest thing

This is partly a strategic argument, but it is also a temperamental one.

I am naturally sceptical of the idea that the newest thing is automatically the wisest choice. New tools can be genuinely transformative, and sometimes early adoption is the right call. But newness also comes with immature ecosystems, changing best practice, thinner operational knowledge, and fewer hard-earned lessons.

More established platforms often offer something less exciting but more useful:

• stronger documentation
• better integration paths
• larger communities
• more implementation experience
• clearer support models

In real organisations, those things matter a great deal.

There is a difference between innovation and novelty-chasing. Innovation can create genuine leverage. Novelty-chasing often just creates churn. From a strategic perspective, reliability, maintainability and organisational fit usually matter more than being first.

AI should augment people rather than displace judgment

One of the principles I hold most strongly is that AI should augment people rather than replace them.

The goal is not to remove humans from the process. It is to amplify what they are capable of. Humans remain responsible for judgment, ethics, creativity and accountability. AI is often very good at processing information, synthesising material, exploring alternatives, accelerating research, and reducing routine cognitive effort.

The greatest value comes from collaboration between the two.

That is not just a theoretical position for me. It is increasingly how I work myself. I often work alongside what I think of as a Digital TeamMate: using AI to research unfamiliar topics, challenge my assumptions, organise my thinking, and help turn ideas into concrete outputs. I am still responsible for the decisions, but AI increases the speed and breadth of my thinking quite dramatically.

In practice, I find that strategic judgment, prioritisation and final accountability are still best led by humans, while AI is exceptionally useful for gathering evidence, surfacing options, identifying patterns and producing the raw material that informs those decisions.

Responsible adoption should preserve that distinction rather than blur it away.

Trust depends on boundaries, not enthusiasm

Responsible AI adoption also means understanding where AI should and should not be trusted.

Not every task is suitable for automation. Not every output should be accepted at face value. And not every organisational decision should be pushed toward the model just because the model can produce a plausible answer quickly.

Real organisations need clear guidance on appropriate use. They need to understand the limitations of the models they are using. They need important decisions to remain appropriately governed. And they need working habits that encourage verification rather than passive acceptance.

AI should increase confidence in decisions, not replace critical thinking.

This is also where information security and privacy become part of the same conversation. Organisations should understand the services they are using, their retention policies, their training policies, and the risks attached to putting proprietary information into them. Free consumer tools may be perfectly appropriate for public information or individual experimentation, but they should not quietly become the default destination for sensitive organisational data.

Even with enterprise platforms, the most confidential information still needs particular care. And for the most sensitive workloads, locally hosted models can sometimes be the better fit because they allow much tighter control over where data resides, how it is processed, and who can access it.

AI is an organisational capability, not a technology project

This is really the centre of my view. I do not think AI should primarily be treated as a technology acquisition problem. I think it should be treated as an organisational capability-building problem.

That means the real work is often the less glamorous work:

• improving data foundations
• designing workable governance
• deciding who owns what
• clarifying the operating model
• creating enough shared understanding that adoption does not remain trapped in a specialist corner

The organisations that benefit most from AI will not necessarily be the ones with the biggest budgets or the flashiest tooling. They will more often be the ones that are able to integrate AI into real work without losing control of quality, accountability or direction.

The tools will continue to evolve. That part is almost guaranteed. The fundamentals will still matter regardless.

Ultimately, responsible AI adoption is really about building trust: trust that the technology is solving meaningful problems, trust that it is operating within appropriate governance, trust that people understand its strengths and limitations, and trust that it enhances human capability rather than diminishing it.

That is why I do not think AI strategy should start with tools. It should start with purpose, constraints, capability and organisational readiness. Only then does the tooling conversation become properly meaningful, and only then does “responsible AI adoption” become something more than a slogan.