Governments around the world are asking the same question:

Which jobs will AI change first?

A new set of research models may offer an answer.

Studies from institutions like Massachusetts Institute of Technology, alongside work examining tools from companies such as Microsoft and Anthropic, are beginning to map how closely artificial intelligence overlaps with human work.

The goal isn’t just prediction.

It’s preparation.

One of the most interesting efforts comes from MIT’s Iceberg Index, a simulation that attempts to measure how much of the labour market is exposed to AI.

Researchers built what they call a “digital twin” of the workforce.

Using the U.S. government’s O*NET occupational database, the model analysed the tasks involved in more than 900 occupations and simulated how AI tools could perform them.

The results were striking.

The study estimates that 11.7% of the labour market could already be replaced by AI using existing technology.

Not in theory.

With tools that already exist today.

The key idea behind the model is something called skill overlap.

Some jobs share many tasks with what AI systems can do.

Others share very few.

For example:

Low overlap

These jobs rely heavily on physical activity and environment-specific skills that AI struggles to replicate.

High overlap

These roles involve structured information and digital workflows — exactly the environment where AI performs best.

One of the clearest signals from the research involves software development.

AI coding assistants are already generating enormous volumes of code every day.

Systems like GitHub Copilot and AI models such as Claude can now write, debug, and explain code in seconds.

According to the study, AI systems are now generating more lines of code daily than human developers.

That doesn’t necessarily mean programmers disappear.

But it does suggest the role of developers may shift dramatically — from writing code line by line to supervising and refining machine-generated output.

For policymakers, this kind of modelling offers something valuable:

A roadmap.

If governments can identify which skills are most exposed to automation, they can target training programmes more effectively.

The Philippines, for example, has already launched a national AI upskilling initiative aimed at students and workers, focusing on helping people adapt before large-scale disruption arrives.

Instead of reacting to job losses later, the idea is to prepare the workforce early.

The most interesting insight from these models isn’t just which jobs AI can replace.

It’s how work is being broken down into tasks.

Very few professions disappear overnight.

But many jobs contain tasks that AI can already handle.

Writing emails.
Summarising documents.
Debugging code.
Answering customer questions.

As AI absorbs these tasks, the shape of work begins to change.

The future of employment may not be defined by entire jobs disappearing.

It may be defined by jobs being quietly redesigned around what humans still do best.

And in many cases, that redesign has already begun.