Thesis
AI will not “take jobs” in one clean sweep. It will take workflows first.
The practical unit of change in the next decade is not the occupation, the department, or even the task list. It is the workflow: the repeatable sequence of steps that turns an intention into an outcome. AI “agents” are simply software that can execute parts of those sequences, and increasingly coordinate with people and other software.
That framing is less dramatic than the headline version of automation. It is also more actionable. If you want to stay economically relevant, you do not need to “beat AI.” You need to learn how to re-design your workflows so that AI handles the predictable parts and you handle the accountable parts.
Context
For years, the debate about automation has bounced between two extremes.
One extreme is panic: a story in which machines replace humans, and the only rational move is to brace for mass unemployment.
The other extreme is complacency: a story in which technology “just augments,” nothing really changes, and the best move is to wait.
The world is rarely that binary. The evidence from major institutions and large-scale surveys is that the effects are likely to be uneven across sectors, occupations, and countries.
The World Economic Forum’s Future of Jobs Report 2025 summarizes employer expectations through 2030, projecting both large role displacement and large role creation, with a net increase in jobs overall, even as the content of work shifts materially. This “churn” picture is crucial: the same economy can experience job creation and job displacement at the same time.[1]
The International Labour Organization’s research on generative AI argues that the dominant effect is likely to be augmentation rather than full automation, but that augmentation can still be disruptive because it changes wages, job quality, and bargaining power.[2]
The ILO’s 2025 refined index emphasizes “exposure” rather than “replacement,” based on a very granular task da
AI will not “take jobs” in one clean sweep. It will take workflows first.
The practical unit of change in the next decade is not the occupation, the department, or even the task list. It is the workflow: the repeatable sequence of steps that turns an intention into an outcome. AI “agents” are simply software that can execute parts of those sequences, and increasingly coordinate with people and other software.
That framing is less dramatic than the headline version of automation. It is also more actionable. If you want to stay economically relevant, you do not need to “beat AI.” You need to learn how to re-design your workflows so that AI handles the predictable parts and you handle the accountable parts.
Context
For years, the debate about automation has bounced between two extremes.
One extreme is panic: a story in which machines replace humans, and the only rational move is to brace for mass unemployment.
The other extreme is complacency: a story in which technology “just augments,” nothing really changes, and the best move is to wait.
The world is rarely that binary. The evidence from major institutions and large-scale surveys is that the effects are likely to be uneven across sectors, occupations, and countries.
The World Economic Forum’s Future of Jobs Report 2025 summarizes employer expectations through 2030, projecting both large role displacement and large role creation, with a net increase in jobs overall, even as the content of work shifts materially. This “churn” picture is crucial: the same economy can experience job creation and job displacement at the same time.[1]
The International Labour Organization’s research on generative AI argues that the dominant effect is likely to be augmentation rather than full automation, but that augmentation can still be disruptive because it changes wages, job quality, and bargaining power.[2]
The ILO’s 2025 refined index emphasizes “exposure” rather than “replacement,” based on a very granular task da