Capability is only the beginning
Most conversations about artificial intelligence start with capability. What can the system write, summarise, predict or recognise? How fast is it, how accurate, how cheap? Demonstrations are designed to answer those questions, and they can be genuinely impressive. A demonstration shows what a tool can produce in favourable conditions.
It tells us much less about the setting in which people will use it. A classroom, a small nonprofit, a hospital ward and a public service each bring different goals, responsibilities and constraints. The same tool can be helpful in one and harmful in another. Capability is a property of the tool; usefulness is a property of the tool in a particular place, used by particular people, for a particular purpose.
Describe the work before choosing the tool
Before deciding how to adopt an AI tool, it helps to describe the work itself. What is actually difficult about it? Which parts take time without adding value, and which parts are where the real judgement happens? What should remain a human responsibility, and why? What would a better result look like to the person receiving it, not only to the person delivering it?
These questions often reveal that the most time-consuming part of a job is not the part that matters most, and that the part that matters most is the hardest to automate well. They also surface quieter forms of work, such as reassurance, relationship building and noticing when something is wrong, that rarely appear in a task list but carry much of the value.
Starting from the work, rather than from the tool, makes it easier to choose where assistance genuinely helps and where it would only move effort somewhere less visible.
The real question is agency
The deeper question about AI is therefore not only what it can do, but who gets to decide how it is used. Who defines the purpose of a system? Who chooses what counts as success? Who can challenge an output, and what happens when they do? Whose experience counts when deciding whether the system is working?
In many organisations these decisions are made by a small group, often quickly, and often by the people furthest from the work. Those who will use the tool every day, and those whose lives it will affect, may only meet it once it is already in place. That is a recipe for systems that look efficient on paper and create friction, mistrust or harm in practice.
Agency is not an abstract principle. It shows up in concrete design choices: whether a person can see why a recommendation was made, whether they can override it easily, whether there is a clear route to a human, and whether feedback from users actually changes anything.
Make room for informed choice
People need enough understanding to question an output, recognise uncertainty and decide when a tool is inappropriate. That kind of AI literacy goes well beyond learning a sequence of prompts or commands. It includes knowing that a fluent answer can still be wrong, that systems can reflect the patterns and gaps in their training data, and that confidence and accuracy are not the same thing.
In education this matters twice over. Students need to learn with new tools, and they also need to learn about them: how they work, where they fail and how to use them honestly. Teachers need time, support and a voice in decisions, not just a new platform and an instruction to adapt.
The design of a system can encourage or discourage those habits. Clear explanations, meaningful controls, visible uncertainty and an easy route to human review deserve the same attention as speed and convenience.
Questions to ask before adopting AI
A short set of questions can make any AI decision more deliberate. What problem are we trying to solve, and how will we know if it has improved? Who will use the tool, and were they involved in choosing it? What should never be delegated to it? How will people report errors, and who is responsible for fixing them? What happens to the data people provide? And how will we decide, after a fair trial, whether to continue?
None of these questions requires technical expertise. They require clarity about purpose and a willingness to include the people who know the work best.
Keep the purpose in view
Adoption is not the same thing as improvement. A useful experiment sets out what it is trying to change and how people will judge the result. It allows for the possibility that the answer is no, and treats that outcome as a success of learning rather than a failure of ambition.
Artificial intelligence will keep becoming more capable. The question that will shape whether it serves people well is not only what it can do next, but who has a say in how it is used, and whether that say is real.
