Someone has read the headlines and is genuinely worried. They do not need statistics about the field and they do not need to be told it will be fine. Here is what is actually worth saying.
Start by taking it seriously
Do not open with reassurance. Telling someone not to worry when they have a legitimate concern ends the conversation and teaches them not to raise it again.
Acknowledge that the concern is reasonable. Work does change and some of it is displaced. Denying that is not honest.
Then narrow it. Move from robots and jobs in general to their job specifically, because the general question has no useful answer and the specific one usually does.
Ask what they actually do. Task by task, across a normal week. Most people have never articulated this and doing so is often clarifying on its own.
Ask what prompted the worry. Frequently something at work — an announcement, a colleague leaving, a new system — rather than the news. That is a different conversation and usually a more actionable one.
The aim is to convert an unbounded anxiety into a specific assessment, because specific assessments can be acted on and unbounded anxiety cannot.
The honest assessment
Once you know what they do, this is the substance.
Which of their tasks are repetitive, low-variation, in a stable setting, with a clear right answer. Those are the exposed ones. Say so plainly.
Which involve judgement, unfamiliar situations, people, or dexterity in varied conditions. Those are not, and they are usually a larger share than the person expected.
Give a proportion, not a verdict. Something like a third of what you do is the kind of thing that gets automated first, and two thirds is not. This is more useful and more accurate than yes or no.
Explain what usually happens. The exposed portion gets absorbed, the remaining work expands, and the role changes rather than ends. Not always — but usually, and it is the honest base case.
Do not promise. If their work is heavily in the exposed category, say that too. False reassurance is worse than none, because it delays the response.
What they can actually do
Concrete steps, in order of usefulness.
Do more of the work in their role that is not exposed. Almost every job has some. Deliberately shifting toward it is the highest-return move and requires no retraining.
Learn to use these tools. Someone who can use AI tools well is in a materially different position from someone who cannot, and this gap is opening faster than the automation itself. This is available to anyone with a phone.
Build depth in something specific. A particular business, industry, place or customer base. General capability is what these systems have; specific knowledge is what they lack.
Keep or build a physical skill involving variation. Trades requiring dexterity in irregular conditions are among the least exposed work there is.
Talk to people at their workplace. What is actually planned is usually more informative than any forecast, and it is knowable.
Do not retrain into something equally exposed. A common and expensive mistake — leaving one repetitive role for another.
What not to say
Formulations that make things worse.
It will not affect you. You do not know that, and if you are wrong they will remember.
Just learn to code. Poor advice generally, and increasingly so, since routine software work is itself exposed.
Robots cannot do what you do. Too absolute, and it invites disproof.
You should have seen this coming. Unhelpful and unkind.
It creates more jobs than it destroys. True at an aggregate level and irrelevant to an individual, since the new jobs are elsewhere and require different skills.
Nothing you can do about it. False, and it is the belief that prevents people acting while they still have time.
What works instead: specific, honest, proportionate, and ending with something they can do this month. Anxiety without an available action turns into avoidance, and avoidance is what actually costs people their options.
The longer perspective
Worth offering once the practical part is covered.
This has happened before and the pattern is known. Mechanisation, computerisation, each displaced particular work and created other work. Painful for those directly affected, absorbed over time at the level of the economy. Both halves of that sentence are true and neither cancels the other.
The pace matters more than the fact. Change spread over a decade is manageable for most people. Change over eighteen months is not. Current evidence points to the slower end, and that is genuinely the more likely case.
Being informed is protective. People who see it coming adjust. People who do not are the ones caught out — and this is the strongest argument for having the conversation at all.
Worry that produces action is useful. Worry that produces paralysis is not. The task is converting one into the other.
And most jobs are not on the edge of disappearing. They are changing, which is uncomfortable and is not the same thing, and confusing the two causes a great deal of unnecessary fear.
Frequently asked questions
How should the conversation start?
Not with reassurance. Acknowledge the concern is reasonable, then narrow from robots in general to their job specifically, task by task — the general question has no useful answer and the specific one usually does.
What does an honest assessment look like?
A proportion rather than a verdict: which share of their tasks is repetitive and low-variation, and which involves judgement, people or dexterity. Usually a larger share is unexposed than they expected.
What is the highest-return step?
Doing more of the work in their existing role that is not exposed. Almost every job has some, and shifting toward it deliberately requires no retraining.
What should not be said?
That it will not affect them, that they should just learn to code, that it creates more jobs than it destroys, or that nothing can be done. The last is false and it is what stops people acting while they still have time.
More in Cost and selection and Humanoids at work.
