What Should AI Be Used For?
A decision framework for AI adoption — not what AI can do, but what it should do.
This page is not about what AI can do — that is a descriptive question, answered in our Uses of AI guide.
This page is about what AI should do — a normative question about human choice, value, and judgement.
§ 01
The normative question
What should AI be used for?
This page is not about what AI can do — that is a descriptive question, answered in our Uses of AI guide. This page is about what AI should be used for — a normative question about human choice, value, and judgement.
Artificial Intelligence has become one of the most accessible technologies ever built. The hardware already exists in billions of pockets. Access is no longer the constraint. The harder question — the one this page addresses — is: what should humans choose to do with it?
The answer is not "everything." Not every problem needs AI. Some are better solved with conventional software, human expertise, or structured processes. The discipline of deciding where AI genuinely belongs — and where it does not — is the subject of this guide.
§ 02
Can vs should: the distinction that matters
What is the difference between what AI can do and what AI should do?
What AI can do is a question of capability. What AI should do is a question of value. The two are not the same — and confusing them is the most common error in AI adoption.
AI can generate text, images, code, and analysis at speed. That is capability. Whether generating those things creates value for a specific function, under specific governance, with verifiable outputs — that is suitability. Capability is necessary but not sufficient.
Organisations that adopt AI because it can do something, without asking whether it should, end up automating processes that were never well-designed, generating outputs no one verifies, and spending on tools that produce activity instead of value. The normative question — should this use AI? — must precede the technical question — can AI do this?
§ 03
When AI should be used
When should you use AI?
AI should be used when all five of the following conditions hold. If any one is missing, the use case is weaker — and if two or more are missing, AI is likely the wrong tool.
The task involves pattern recognition at scale
Condition 1AI excels at finding patterns in large volumes of data — documents, images, transcripts, logs — that a human would take too long to process.
The task benefits from speed of generation
Condition 2Drafting, summarising, translating, and transforming content where speed matters and a human can review the output.
Sufficient data exists
Condition 3AI needs input — context, examples, or training data. Without it, the output is generic or wrong.
The output can be verified
Condition 4A human can check the result against a standard before it is used. If the output cannot be verified, it should not be trusted.
The decision remains human
Condition 5AI assists; the human decides. If the use case requires AI to make the final decision, it is not suitable — it is delegation without accountability.
§ 04
When AI should NOT be used
When should you not use AI?
Not every problem needs AI. Some are better solved with conventional software, human expertise, or structured processes. The following are cases where AI is the wrong answer — even when it is technically capable.
When the task is deterministic
AvoidIf the task has a single correct answer and a conventional algorithm produces it reliably, AI adds probabilistic variation where none is needed. Use conventional software.
When the stakes are high and verification is hard
AvoidIf a wrong output causes harm and a human cannot reliably check the result before it is used, AI should not be the tool.
When the cost of AI exceeds the value
AvoidIf the subscription, API, compute, and rework costs exceed the value the AI produces, the use case is not worth it. See our AI CapEx vs OpEx guide.
When the process is badly designed
AvoidAutomating a badly designed process makes it run faster — not better. Fix the process first, then consider AI.
When human judgement is the product
AvoidIf the value of the work is the human judgement itself — advisory, therapeutic, legal, strategic — outsourcing it to AI removes the value the client pays for.
§ 05
The decision framework
How do you decide whether to use AI?
A structured decision replaces guesswork. The framework has five steps, asked in order — before any AI product is considered.
Step 1 — Define the problem
Step 1What decision, task, or work is being addressed? Define the problem, not the AI. If the problem is unclear, no tool will solve it.
Step 2 — Assess whether AI adds value
Step 2Does the task meet the five conditions in §03? If not, AI is not the answer. If yes, proceed.
Step 3 — Assess whether AI is safe
Step 3What are the governance boundaries? What may AI not change? Who approves? If governance is undefined, define it first.
Step 4 — Assess whether AI is verifiable
Step 4Can a human check the output against a standard before it is used? If not, the use case is not ready.
Step 5 — Decide: adopt, do not adopt, or augment
Step 5The decision is one of three: adopt AI for this use case, do not adopt it, or augment an existing process with AI rather than replacing it. See our Build, Buy or Augment research.
§ 06
The first principle: problem first, AI second
What is the first principle of AI adoption?
The first principle of AI adoption is: define the problem before the solution. This is the Logic Centre of AI Logic Engineering applied to the adoption decision itself.
Most AI failures begin with the reverse — a tool in search of a problem. Organisations buy AI, then look for uses. The result is activity without value: tools deployed, subscriptions paid, outputs generated, and no measurable return. MIT, Gartner, McKinsey, and IBM all report the same pattern — pilots that never scale, experiments that never produce financial returns.
The discipline that prevents this is simple to state and hard to practise: understand what you are trying to do before you choose what to do it with. AI is a tool, not a strategy. The problem comes first; the AI comes second.
§ 07 — Bridge
The normative question leads to the discipline.
Deciding what AI should be used for is a human responsibility. The discipline that governs that decision — and every decision after it — is AI Logic Engineering.
§ FAQ
Frequently Asked Questions About What AI Should Be Used For
01What should AI be used for?+
AI should be used for tasks that involve pattern recognition at scale, benefit from speed of generation, have sufficient data, produce verifiable output, and keep the decision in human hands. If any of these conditions is missing, the use case is weaker.
02What is the difference between what AI can do and what AI should do?+
What AI can do is a question of capability. What AI should do is a question of value. Capability is necessary but not sufficient — AI should only be used where it creates genuine value under proper governance.
03When should you use AI?+
Use AI when the task involves pattern recognition at scale, benefits from speed, has sufficient data, produces verifiable output, and keeps the final decision with a human.
04When should you not use AI?+
Do not use AI when the task is deterministic, when stakes are high and verification is hard, when cost exceeds value, when the process is badly designed, or when human judgement is the product itself.
05How do you decide whether to use AI?+
Follow a five-step framework: define the problem, assess whether AI adds value, assess whether it is safe, assess whether it is verifiable, then decide — adopt, do not adopt, or augment.
06What is the first principle of AI adoption?+
Define the problem before the solution. Most AI failures begin with a tool in search of a problem. The problem comes first; the AI comes second.
07Should every business process use AI?+
No. Not every problem needs AI. Some are better solved with conventional software, human expertise, or structured processes. Forcing AI into every process produces activity without value.
08What problems should AI not solve?+
AI should not solve deterministic problems (conventional software is better), high-stakes problems with hard-to-verify outputs, or problems where human judgement is the value being delivered.
09Is AI always the best solution?+
No. AI is one tool among many. It is the best solution only when the five conditions in this guide are met. For many tasks, conventional software or human expertise is superior.
10How do you know if AI is worth it?+
Compare the value AI generates against the total cost — subscriptions, API, compute, integration, and rework. See our AI CapEx vs OpEx guide for the investment framework.
11What should AI not be used for?+
AI should not be used for tasks that are deterministic, high-stakes with unverifiable outputs, more cheaply solved by conventional software, or where the process itself is badly designed and needs fixing first.
12What is the AI adoption decision framework?+
A five-step framework: define the problem, assess value, assess safety, assess verifiability, then decide whether to adopt, not adopt, or augment an existing process.