The headlines say AI is coming for the jobs. The usage data says something calmer.

You have read the headline a dozen times this year. This role is finished. That profession is next. A number, usually large, usually with no method behind it. You are trying to plan a workforce, and the loudest inputs are the least reliable. Somewhere under the noise is a simpler question you actually need answered. Where is AI really landing in the work your people do, right now?

There is a calmer and more useful way to read this. Not the predictions, which are cheap, but the usage: what people are really handing to these tools, in what proportion, across which tasks. That data has started to arrive, and it tells a steadier story than the headlines. It will not settle every worry. It will give you something firmer to plan on than fear.

What does the real usage data actually show?

One of the better windows is the Anthropic Economic Index. It is an ongoing analysis by the AI company Anthropic of how its Claude model is used, across millions of anonymised conversations. It sorts each use into two buckets. Augmentation means the AI helps a person do the work. Automation means it does the task on its own. In the recent data, augmentation leads: about 52% of use, against 45% for automation. The share doing the work with a person, not instead of them, went up.

That single split reframes the panic. The dominant pattern is a person plus a machine, not an empty chair. It fits what leaders keep telling me they see on the ground: the tool drafts, the person decides; the tool suggests, the person judges. It is the same picture behind the shift from doing the task to designing and running the loop.

Is AI coming for whole jobs, or for tasks?

Tasks, mostly, and that distinction changes how you plan. A job is a bundle of many tasks. The data shows the tool reaching into that bundle unevenly. About 49% of jobs now have at least a quarter of their tasks touched by Claude in some way. Very few have the whole bundle done for them. The work is being taken apart task by task, and reassembled, rather than switched off role by role.

Concentration matters too. Computer and mathematical tasks account for roughly 35% of conversations on the consumer app, far above any other category. Most work sits well below that. So the effect is real but localised, heaviest where text and code are the raw material, lighter elsewhere. Honesty demands one caveat, stated plainly. This is one platform's usage, not a full census of the economy. Treat it as a strong signal of direction, and pair it with what you can see in your own operation. It is a better basis than a headline, and it is closer to where AI actually changes the work.

So how should a leader plan from this?

Stop asking which jobs will vanish, and start asking which tasks are already moving. That is the question the data can answer, and the one your planning actually turns on. Work it in this order:

1. Map the tasks, not the titles. Break the key roles into their real tasks, and mark which ones AI already touches in your operation. The picture is always more specific than the headline.
2. Follow the augmentation first. The fastest, safest value is where a tool helps a person do more, not where you try to remove the person. Start where the evidence is strongest.
3. Redesign the role around what remains human. As tasks shift to the tool, the job becomes more about judgement, relationships and direction. Plan the role toward that, deliberately.
4. Reskill for the tasks that are moving. Aim training at the specific work changing under people's hands, not at AI in the abstract.
5. Watch your own usage data. Your logs are a better guide to your business than any global average. Read them monthly.

What does this make possible?

A calmer, better-aimed plan than fear would ever produce. When you plan from real usage, the work stops being a countdown to lost jobs. It becomes a map of tasks in motion, most of them shifting toward a person and a machine working together. You can see where to invest, where to reskill, and where the human part of the role actually grows. That is a far better footing than the guesswork that stalls so many AI efforts.

Picture your team a year on. The tool has taken the repetitive slices of a hundred jobs, and your people spend more of their day on the parts that need a human. You planned it from evidence, task by task, so it landed as a lift rather than a shock. That is what the shape of work after business-as-usual looks like when a leader reads the data and moves early.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.