A job title is a surprisingly poor description of a working day. Someone called an analyst might spend the morning cleaning a spreadsheet, the afternoon resolving an ambiguous request and the final hour explaining a decision to a colleague. Those activities do not respond to automation in the same way. That is why I find a task list more useful than a confident prediction about which profession disappears next.
.jpg){kind=link}
Exposure is a question about tasks
The International Labour Organization’s 2025 research examines occupational exposure to generative AI using task-level information. Exposure describes the potential for technology to affect work; it should not be read as a count of jobs that have already disappeared. The distinction is important because the same role can contain routine text production, relationship management and decisions that require accountability.
For a personal exercise, take one representative day and divide it into actual activities. “Prepare a report” is still too broad. Gathering the inputs, checking the definitions, drafting a paragraph and deciding whether a result is credible are separate steps. Once those steps are visible, the question becomes where assistance might help and where a person must remain responsible.
A faster draft is only part of the calculation
Consider a hypothetical task that takes 40 minutes without assistance. A tool creates a first draft in five minutes, but checking it takes 20 minutes and fixing an error takes another ten. The finished task now takes 35 minutes. That is an improvement, but it is very different from the eightfold acceleration suggested by comparing drafting time alone. These numbers are an example, not a measured productivity result.
The missing term is often review cost: the effort required to make an output usable. Review can be quick when the answer is easy to verify and much slower when an error looks plausible. I would rather save five dependable minutes than claim a large gain that simply transfers work to the next person in the process.
Choose a small experiment with a clear finish
A useful trial has a defined input, an acceptable output and a way to compare the result. For instance, ask whether a tool can turn a non-confidential outline into a readable meeting summary, then assess omitted decisions and invented details. Follow your employer’s rules on confidential information and approved services. Convenience does not establish permission to submit business data.
Keep a short record of total time, corrections and downstream questions. A draft that produces three extra clarification messages may not be an improvement. Equally, a tool that saves little time but makes a difficult document easier to understand may still be valuable. The measure should reflect the reason for doing the task.
Build judgment alongside familiarity
My view is that learning to inspect an answer is at least as valuable as learning to ask for one. In a spreadsheet, that may mean tracing a number to its source. In a customer message, it may mean recognizing a promise the business cannot keep. In either case, subject knowledge gives the user a way to judge a fluent response.
The practical next step is modest. Pick one recurring activity, define what good work looks like and run a limited comparison. Keep the steps that improve the finished result and abandon those that create hidden review work. A career will not fit into a single productivity score, but a clearer understanding of your own tasks is a useful place to begin.
What I would measure first
Measure the complete task, including checking and corrections. Exposure to AI is not the same as a demonstrated job loss.
Use Meeting overlap finder for two time zones ↗
What is the difference between AI exposure and automation?
Exposure describes tasks that technology could affect. Automation describes work actually performed by a system; adoption, reliability and human responsibilities influence that outcome.
Sources & further reading
Source material reviewed Sep 6, 2026. These links support the factual background. Worked examples and editorial interpretations are identified in the text.
Join the conversation
What would you add, question or explain differently? Please discuss the idea and respect the person.
Comments are screened for spam and abuse. Some are held for review. We store your comment and a daily security identifier; see privacy. Keep personal contact details out of your comment.