We spend a lot of our time helping businesses put AI to good use, so it might seem odd for us to write a piece about where it does not belong. But this is exactly the kind of honesty that earns trust. The loudest voices right now want you to believe AI can do everything, and that is not true. Knowing where it is the wrong tool is just as valuable as knowing where it shines, and frankly it is rarer advice.
Reaching for AI on the wrong job is not clever automation. It is a quiet way to introduce risk into your business and call it progress.
This is not a warning against AI. It is a map of the places to keep human hands firmly on the wheel. Use it well in the right spots and keep it out of the wrong ones, and you get the best of the technology without the nasty surprises.
Where AI is the wrong tool
There are a few categories where we consistently advise clients to keep a person in charge of the decision, not the machine.
- Decisions needing guaranteed accuracy you cannot verify. AI can be confidently wrong. If a task demands an answer that is exactly right, and you have no quick way to check it, you are trusting a guess dressed up as a fact.
- High-stakes legal, medical and financial calls. AI can assist a professional here, gathering and drafting. But the final call, the one a person's wellbeing or money rides on, belongs to a qualified human who carries the responsibility.
- Anything needing real accountability. When something goes wrong, a person has to answer for it. A model cannot be held accountable, cannot be questioned, and cannot stand behind its choice. For decisions that require an owner, that owner must be human.
- Tiny one-off tasks. If setting up and checking the AI takes longer than just doing the thing yourself, the AI is not saving you anything. Not every job is worth automating.
- Cases where a wrong answer is costly and unverifiable. The dangerous combination is high cost of error plus no easy way to catch the error. That is precisely where you do not want a confident guess.
- Relationships built on trust. A difficult conversation with a long-standing client, a sensitive bit of feedback to a staff member, a heartfelt apology. People can tell when these are outsourced to a machine, and it damages the very thing you were trying to protect.
The problem with AI is not that it gets things wrong. It is that it gets things wrong with total confidence, and that confidence is contagious.
A simple test before you reach for it
You do not need a policy document to make good calls here. Two questions handle most cases.
- Can a human verify the output quickly? If yes, AI is a reasonable assistant, because a person can catch its mistakes before they matter. If no, be very careful, because errors will slip through unnoticed.
- Is a mistake reversible? If a wrong answer can be spotted and undone cheaply, the downside is small. If a mistake is expensive, public, or permanent, that is a job for a human to own.
When both answers point the wrong way, when nobody can easily check the work and a mistake would be hard to undo, that is your signal to keep AI out of the driver's seat.
Notice that these questions say nothing about how clever the AI is. That is deliberate. The decision is not about the model's capability, it is about your exposure if it gets things wrong. A brilliant tool on an unverifiable, irreversible task is more dangerous than a mediocre one on a job you can check in seconds. The risk lives in the task, not the technology.
Assist, do not decide
In all of these grey areas, AI still has a role. It can gather the background, draft a first version, summarise the documents, and surface the options. What it should not do is make the final call. Let it do the legwork so your people have more time and better information, and let the human make the decision and carry it. That single distinction, assist rather than decide, keeps you on the right side of almost every risk.
Confidence comes from knowing the edges
There is nothing timid about drawing a clear line around where AI belongs. It is the opposite. A business that knows exactly where the technology helps and where it does not is far more confident in using it than one that throws AI at everything and hopes. The first kind sleeps well. The second kind gets a nasty letter one day.
So by all means use AI, and use it boldly where it earns its place. Just keep the high-stakes, hard-to-verify, trust-based and accountable decisions where they have always belonged: with a person who can think, check, and answer for the call. Knowing that boundary is not a limit on what AI can do for you. It is what makes the rest of it safe.
