"Will AI replace my job?" Everyone is asking it. It's an understandable question, but it's poorly framed. AI rarely eliminates an entire job in one go. It dissolves tasks, one by one, from within the job.

The real question is more precise: which part of your value is becoming automatable, which part is becoming scarcer, and what new level of expectation is emerging? As long as we think in terms of "job versus AI", we're looking at the wrong scale.

A job isn't a block, it's a stack of tasks

No job is homogeneous. It mixes repetitive, standardized tasks, preparation work, decisions that carry responsibility, and moments of judgment that are hard to reduce to an instruction. AI doesn't act on the job as a whole: it acts on each of these layers differently.

A simple framework helps you see clearly. For each activity, ask yourself: can it be automated right now? Can it be augmented, with AI preparing and a human validating? Is it hard to automate because it depends on context, trust or responsibility? Or is it an entirely new task, created by the arrival of AI?

What automation really shifts

What goes first are the repetitive, verifiable, low-stakes tasks: formatting, first drafts, sorting, standard reporting, mechanical coordination. What remains, and gains value, is framing, decision-making, validation, and responsibility for what is produced.

In other words, AI doesn't hollow out your work. It shifts where you add value: less in execution, more in asking the right question, checking, and owning the result.

The trap: automating a bad process

Many automation projects fail not for technical reasons, but because they automate a process that was already bad. Automating a bad process doesn't make it better. It only lets you produce the bad result faster, and at a larger scale.

Before automating, the first question isn't "how?" but "does this process deserve to exist at all?" A useless task performed by an AI is still a useless task.

Good automation doesn't just remove clicks. It makes a decision more reliable, or makes possible a service that couldn't be delivered manually at that scale.

The new tasks nobody had in their job description

AI doesn't just take work away. It creates work. And this new work is rarely mentioned in the talk about the "end of jobs":

  • Oversight. Checking what the machine produces, spotting the error that looks plausible but is wrong.
  • Orchestration. Chaining tools together and making information flow between them.
  • Accountability. Deciding what gets published, what gets sent, what gets signed.
  • Error handling. Planning for what happens when the automation gets it wrong, because it will.

So what should you do in practice?

Rather than asking whether your job will disappear, list your tasks and run them through the framework. Identify the ones that are repetitive and frequent: they are the first candidates for automation. Identify the ones that call for judgment, trust or responsibility: those are where to invest. And for every automation you're considering, ask who understands the system, and what happens the day it breaks down.

The real question isn't whether AI can do it. It's who still has something distinctive to ask of it, and who owns the result.