Audit your own job before someone else does

Every person who works now owes themselves one uncomfortable piece of analysis.

Take your job apart, task by task, and ask:

How much of this could AI already do?

Not in five years. Not when the technology becomes perfect. Now — with the right instructions, access, context, tools and human checks.

It is an unpleasant question because a serious answer may challenge the story we tell ourselves about why our work requires us. It is also becoming irresponsible not to ask it.

The reason is simple: other people are already doing the exercise.

The distance will grow quickly

AI does not improve everyone’s performance by the same amount.

One person uses it occasionally to rewrite an email or summarize a document. Another redesigns the entire workflow: research, analysis, drafting, checking, documentation, coordination, follow-up and learning.

The second person is not merely a little faster.

They take on more work. They test more ideas. They learn from more attempts. They build reusable instructions and systems. Each completed assignment makes the next one cheaper and better.

That advantage compounds.

Soon, the comparison is no longer between two people doing the same job at slightly different speeds. It is between one person performing a familiar role and another operating a small, continuously improving system.

The people who make this shift will not simply pass those who do not. They may move so far ahead that they are no longer doing economically comparable work.

I can see it in my own work

This is no longer theoretical for me.

In parts of my work, I can now do what would recently have required ten people — and often go further.

I can investigate a market, collect and compare evidence, develop a strategy, produce client material, build and test software, prepare commercial activity, document decisions and improve the operating method while the work is being done.

That does not mean I have suddenly become ten times more intelligent or that human expertise has stopped mattering.

It means that I no longer have to perform every step personally.

AI can execute a large share of the research, production, comparison, documentation and coordination. My attention can move towards intent, judgment, priorities, relationships, consequential decisions and accountability.

The result is not only that the same work happens faster. The possible scope of the work becomes larger.

That is the part many productivity discussions miss.

A job is only a bundle of tasks

People often defend their position by saying that AI cannot do their whole job.

That is probably true — and increasingly irrelevant.

A job is a bundle of tasks, decisions, relationships and responsibilities. AI does not need to replace the complete bundle for the economics of the role to change. It only needs to perform enough of the work that one person can carry what previously required several.

We saw this when we mapped 45 tasks across a legal-work lifecycle. Nineteen appeared capable of becoming fully orchestrated normal-flow tasks. Another nineteen could largely be performed by agents with a human gate around material judgment or action.

Only seven clearly remained tasks that a human should continue to own, even if AI could prepare much of the work.

The important number was therefore not nineteen. It was thirty-eight.

Once we stopped asking whether AI could replace “the lawyer” and started examining the actual work, the architecture changed.

The same audit should now be conducted in every profession.

The audit needs to be methodical

Start with the work as it actually happens, not with the job description.

List the recurring tasks, the decisions, the handovers, the searches, the documents, the meetings, the checks, the follow-ups and the administrative residue.

Then classify each part:

• Remove: Does this work need to exist at all?
• Delegate: Can AI perform it reliably from beginning to end?
• Gate: Can AI do the work while a human approves the consequential point?
• Keep human: Does the task depend on judgment, trust, negotiation, responsibility or legitimate authority?
• Expand: What valuable work becomes possible only because AI makes it affordable?

Then test one real workflow.

Measure the time, quality, corrections and failure modes. Turn repeated corrections into instructions. Give the system memory. Add independent checking where the consequences justify it. Move human attention towards the places where it genuinely changes the outcome.

This is not a one-time efficiency exercise. It is the beginning of a learning system.

The headcount question cannot be avoided

New work will appear. Demand may expand. Some roles will grow, and entirely new ones will be created. The effects will not be uniform across industries, companies or countries.

But it is difficult to look honestly at the capacity already available and conclude that the same volume of work will continue to require the same number of people.

If one capable person with a well-designed AI system can do the work that previously required five, ten or more people, organisations will eventually reorganize around that fact.

Some will use the capacity to grow. Some will lower prices. Some will improve quality. Some will simply need fewer people.

We should not pretend to know the final labour-market outcome. We should also not use uncertainty as an excuse to ignore the direction of travel.

Self-automation is still the better choice

There is something deeply uncomfortable about trying to automate parts of your own job.

It can feel like participating in your own replacement.

But refusing to examine the work does not preserve it. It only leaves the redesign to someone else — a colleague, a competitor, a new entrant or the person deciding what the work should cost.

Doing the audit yourself gives you agency.

You can decide where human judgment matters. You can build the controls. You can redirect the capacity towards better questions, stronger relationships, more ambitious work and responsibilities that machines should not hold.

The choice is not between changing and keeping the job exactly as it is.

The real choice is whether you help redesign your role while you still understand it better than anyone else — or wait until someone else redesigns it around you.

What changed: I stopped thinking of AI primarily as a productivity tool. In my own work, it is becoming operating leverage: one person can direct a system that performs work once spread across many roles.

What I think matters: Everyone should audit their work before their employer or competitor does. The people who methodically redesign their own workflows will not merely become more efficient; they will increase their range, speed and rate of learning at the same time.

What remains uncertain: We do not know how much new demand and new work this capacity will create, or how quickly organisations will convert technical potential into real operating change. But uncertainty about the final number of jobs does not remove the immediate responsibility to understand what is already possible.