Section C8 · Put it down
When AI is the wrong tool
This manual is relentlessly upbeat, on purpose. But a manual that only tells you when to reach for a tool, and never when to put it down, is selling you a hammer and calling every problem a nail. There are jobs where a modern AI is genuinely the wrong tool, and reaching for it anyway looks fine right up until it doesn't.
Why the machine guesses
A language model does not look anything up and does not calculate. It predicts the next most plausible word, very fast. Most of the time plausible and correct are the same thing, which is why it feels like magic. Sometimes they come apart, and the model has no idea, because it was never checking against the truth. It is confident in exactly the same tone either way. A calculator is never wrong about arithmetic and never sounds confident about anything. When the cost of plausible-but-wrong is high, you want the boring machine.
The one question
Before handing over a task: would I accept a confident guess here? If yes (a draft, a summary, a brainstorm, a shortlist you will check anyway), crack on. If no, slow down; you may be holding the wrong tool.
Five places to put it down
When the answer must be exactly right and checkable. School-fees totals, VAT, medicine doses by weight. Arithmetic wearing a suit. Use a spreadsheet, a formula, a calculator. Let the AI set the formula up; never let it produce the final number nobody rechecks.
When someone is accountable. Medical, legal, financial, hiring, anything assessed or regulated. Use AI to prepare; keep the decision, and the accountability, human. "The chatbot said so" is not a defence you want to be running anywhere.
When there is no undo. The email to the whole client list, the transfer, the legal letter. On no-take-backs actions the review step is not optional, and it is not the AI's job. It is yours.
When something may be trying to trick it. Any input from a source you do not control can hijack an AI's instructions (the trade calls it prompt injection). Anywhere the input is hostile, keep a human between the AI and anything that matters.
When you need the same answer every time, plus an explanation. Who qualifies for the discount, how the penalty is calculated. Ask a model twice, get two confident answers. That is what written rules are for, and rules remain AI's older, more reliable cousin. You don't bring a chainsaw to prune the roses, however much you enjoy the chainsaw.
The professional stamp
A special case of accountability that this group, being full of engineers, builders and consultants, keeps meeting: when AI does work that a professional would normally sign (a design, a calculation, a rendering presented as buildable, advice a client acts on), the work gets faster but the liability does not move. The model carries no insurance and holds no licence. If your name goes on it, price the checking, not just the generating, and word what you publish honestly: an illustration is not a survey, an opinion is not advice.
The rule of thumb
Use AI to explore, draft, summarise and think out loud. Put a deterministic backstop wherever being wrong is expensive: a formula for the number, a human for the decision, a source for the fact, a rule for anything that must be consistent and explainable. Most weeks the clever guesser is the right tool. The weeks it isn't are the ones worth getting right.