What the AI Company Hires For Now
The bottleneck now is us. That is the good news.
Most of what we hear about AI and jobs is a forecast. Somebody guessing, loudly.
This week I read something different. Anthropic, the company that makes Claude, looked at its own workforce and measured what changed. Their report, How employment is changing at Anthropic, came out on October 16. It draws on internal data about hiring, AI use, output and calendars, plus 17 interviews with employees.
It is one company, and an unusual one. The authors call it a possible leading indicator for the rest of the economy. I think that is the right way to read it: a look a little way down the road.
What they found
Inside Anthropic, AI agents now log six hours for every one hour their people spend directing them in interactive sessions. Junior hires in technical roles fell from 27% of technical hiring to 9%. Each person produced more code, closed more sales and processed more contracts.
And then the part that surprised me. Those agents spend about 80% of their session time waiting. Waiting on a person to read, decide, approve, redirect. One engineer described it as “an army of Claudes, with every one asking for my attention.”
The bottleneck now is us.
The gap nobody talks about
In a companion survey, The AI expectations gap, Anthropic staff said they expect AI to be able to do 75% of their current work by 2030. American workers overall said 28%.
That gap is the reality I keep coming back to. The people who work closest to these tools expect the most change. Most of us, quite reasonably, expect a lot less. We are busy. We have jobs to do. Nobody has sat us down and shown us what the people inside the lab are seeing.
Fear won’t help us here. One of their own executives said it better than I could: “a workforce that is anxiously worried is not a workforce doing its best work.” Worry does not prepare anyone. Practice does.
What becomes the job
The line from the report I have reread the most came from someone outside engineering: “that 20% that was strategic becomes 90% of the job.”
When the routine parts get automated, what is left is the part that always needed a person. Deciding what matters. Talking to people. Owning the call. The report says hiring managers there “increasingly emphasize applicants’ autonomy and judgment relative to specific technical expertise.” Notes about a candidate’s creative use of AI tools tripled for technical roles and quadrupled for non-technical ones.
That applies well beyond engineering. It fits anyone whose job has a routine half and a judgment half, which is most of us.
How I’m getting ready
None of this needs a new degree. It needs a few habits, started now, while the stakes are low.
- Find your 20%. Look at last week. Which hours needed your judgment, and which were routine? Do more of the first on purpose.
- Give AI one real task a week. Pick something you would have done yourself, with real stakes. You learn more from one honest task than ten experiments.
- Get good at the brief. Say the goal, what good looks like, and why. I wrote about learning that the hard way.
- Start before anyone asks you to. Most of us do not work somewhere with an early-warning system for changing roles. Anthropic built one. The rest of us will have to be our own.
Nobody is coming to rewrite our jobs for us. That sounds heavy, but I find it freeing. It means we get to start first.
What part of your job is the 20% that needs you? Reply to the newsletter and tell me.