The pipeline, not the headcount

Every leadership conversation about AI eventually reduces to one question, and it is almost always the wrong one: how many people does this replace?

In June 2026 Anthropic published an Economic Index report built on something better than speculation — usage data sampled between 10 April and 10 June 2026, linked to survey responses from 9,700 people who use this at work. Not opinions about AI. Answers from people whose jobs already contain it.

Start with what those people say about their own risk.

10% rated losing their job likely or very likely within twelve months. For context, roughly 13.4% of US jobs end in an ordinary year for entirely ordinary reasons.

The people closest to the technology rate their risk below the base rate of simply having a job end.

What the people using it actually report

The rest of the survey explains the calm.

Read the last bar twice, because it is the one that contradicts the standard narrative. A majority say AI made their skills more valuable, not less.

And the pattern holds where you would least expect it: the people delegating the most work to AI are the most optimistic about their future pay, job security and sense of meaning. Heavy automation users report learning at the same rate as everyone else; 68% report learning more overall.

Asked what they want, over half said they want to collaborate with AI on meaningful work. The second most common answer was to have the tedious parts taken away so they get time back.

None of that reads like a workforce being hollowed out. It reads like a workforce that has found a better tool and would like more of it.

The finding that should change your planning

Here is the number that matters, and it is easy to skim past.

Workers with 15 or more years of experience report that AI can do roughly 10 percentage points less of their work than first-year workers report about theirs.

Sit with the direction of that. Seniority does not increase exposure. It reduces it. The expensive, experienced people are less automatable than the ones you just hired — because what they do is decide, judge, negotiate, and take responsibility, and what junior people do is draft, summarise, check, and produce a first pass.

The first pass is exactly what this technology is good at.

So the honest read of the exposure data is not "AI comes for the senior roles". It is the reverse, and it is more awkward: the automatable work is disproportionately the work your juniors are currently doing.

Where I think that leads

What follows is my inference rather than a measured finding, and I would rather label it than smuggle it in.

Junior work is not only output. It is also the mechanism by which people become senior. Nobody develops judgement about which analysis is wrong by being told; they develop it by producing four hundred analyses and being corrected. Drafting the thing badly, having it sent back, and drafting it again is the training programme, and in most companies it is the only one that exists.

Automate that layer efficiently and the arithmetic works beautifully for about three years. The output still ships. The costs fall. And the supply of people capable of judging whether the machine's output is any good stops being replenished — inside a company that now depends on that judgement far more than it used to.

That risk shows up in no quarterly number. It has no line on a P&L. It is visible only in a question nobody asks in a business case: where will the person who checks this come from in 2032?

What to actually do

Five steps, in order, and the first one costs nothing.

Find out who already uses it. In almost every company of any size, someone is quietly doing their job three times faster and has told nobody, because the incentives for admitting it are unclear. That person is your most valuable resource in this whole transition and they are usually invisible.

Measure tasks, not jobs. Job titles are bundles of tasks bought together for historical reasons. AI arrives at the task level, so the first effect is reshuffling of the bundle, not removal of it. Plan at the granularity where the change actually happens — the same discipline as choosing what to automate first.

Protect the training ground on purpose. Decide which work juniors still do by hand, write down why, and defend it when someone points out it would be cheaper otherwise. It will always be cheaper otherwise. That is the trap.

Retrain towards judgement. Prompt-writing is a weekend. The durable skill is knowing what should be built and being able to tell whether the output is right, which is the same capability the whole harness argument rests on.

Say it out loud. The survey says your AI-using people are not panicking. What generates fear is not the technology — it is a leadership team that has obviously decided something and will not say what. If the plan is no redundancies, say so. If the plan is that some roles change substantially, say that instead. Silence gets filled with the worst available guess.

The honest summary

The evidence available today does not support the version of this story where AI arrives and the workforce empties out. People who use it every day report working faster, doing more, and — a majority of them — becoming more valuable, and they put their own odds of losing their job below the ordinary rate at which jobs end.

The exposure is real, but it is pointed somewhere else than the headlines suggest. It falls hardest on the early-career work that happens to be how people learn to do the senior work, which no measurement will flag and no quarter will punish.

Which makes it a governance problem rather than a technology one. The question is not how many people this replaces. It is which capability you are quietly deciding to stop growing.

Frequently asked questions