Truck drivers hand more work to AI than programmers do. Hardly any of them use it.

Out of every 100 conversations with Claude about truck driving, 65 hand the task over. That is the highest rate of any of the twenty jobs we have written up, ahead of software developers at 61 and accountants at 45.

It is also built on almost nothing. Truck driving makes up under 0.01% of the conversations in the data. Nobody drives a truck with a chatbot. Some drivers, it seems, bring it the paperwork.

That one number carries most of what is useful, and most of what is misleading, about the one measure on this site that records what people actually did with AI rather than what it could do.

Where the number comes from

The Anthropic Economic Index sorts conversations with Claude by the kind of work they are about, matched to 542 US occupations. For each one it reports two shares that add up to 100: conversations where the person handed the task over to be done, and conversations where they worked through it together.

Two things to hold on to before any of it means anything. It is published by Anthropic about its own product, so it describes people who use Claude, not workers in general. And it never feeds into a score here. It sits on the job pages as observed context, labelled every time, because it answers a different question from the ILO task ratings the scores are built on.

The figures below are from the release of 26 June 2026, covering conversations from May.

The jobs that hand over most barely use it

Sixteen of our twenty written profiles have a match in the data. Sorted by how much gets handed over, the top of the list is not what you would guess:

Job	Handed over, per 100	Share of all conversations	Automation, JSA assessment
Truck driver	65.2	under 0.01%	21
Software developer	60.8	0.35%	63
Electrician	60.8	under 0.01%	16
Aged care worker	51.9	0.08%	21
Bank teller	50.8	0.02%	54
Accountant	45.0	0.12%	54
Graphic designer	29.6	1.07%	49

The last column is the one worth comparing. Jobs and Skills Australia assessed each occupation for how much of its work AI could do without a person, on its own scale. For software developers the two sources roughly agree. For electricians they are nowhere near each other: an assessed 16 for what AI could take over, against 61 in 100 conversations handed over.

Both can be true at once, because they measure different things. The assessment covers the whole job, most of which is on a ladder. The conversations cover the sliver of it that reaches a chatbot at all, and that sliver is the part most likely to be handed over whole: a quote to write up, a compliance form, a logbook. The data does not say which tasks they were. The page for every one of these thin jobs carries the same label: read this one as a warning, not a finding.

It is not just our twenty

The pattern holds across the whole release. Of the 542 occupations, 162 each make up under 0.01% of the conversations. On average they hand over 60 in every 100. Every other occupation, from the barely-used to the heaviest, averages about 49.

In other words, the less a job appears in the data, the more automated it looks. That is not a finding about those jobs. It is what a split built on a handful of conversations does: it drifts towards whatever the few people who turned up happened to be asking for, and nobody opens a chatbot to lay bricks.

The heaviest users don’t agree with each other either

At the other end, where there are enough conversations for the split to mean something, there is still no single story.

Graphic designers make up 1.07% of all conversations, more than any other job we have written up, and hand over fewer than 30 in 100. Writers and authors, at 2.22% of the data, hand over 28.5. Editors, at 2.58%, hand over 33.6. The people who use it most for words and pictures mostly keep their hands on the work.

Computer systems analysts, at 1.91%, hand over 84. Document management specialists, the single largest group in the release at 5.1% of conversations, hand over 67. Weighted by how much each job is used, the whole dataset comes out at almost exactly half and half.

So heavy use does not mean hand-it-over, and it does not mean keep-it-close. It depends on the work, which is the whole point of looking at work task by task.

How to read the panel on a job page

Every job page that has this data shows it under “When people use AI for this job, what do they hand over?” Read it in this order:

1. The share first. Below 0.1% of conversations, the page flags it. Take the flag seriously: a split built on a sliver is describing whoever turned up, not the job.
2. Then the split. If the share is healthy, the split tells you something real about how people who already use AI for this kind of work are using it.
3. Then set it beside the score, not inside it. What people hand over is behaviour. The score is what the tasks allow. When the two disagree sharply, as they do for electricians, the disagreement is usually the most informative thing on the page.

None of it says what an employer will do, and none of it predicts anything. It is a record of a month of conversations with one company’s product, which is exactly as much as it claims to be.

Find your job and look for the panel. If your job isn’t there, that is information too.