Anthropic's AI agents now clock six hours for every hour its staff work, and somebody still has to read what they produce
By Brett Gavaghan, Founder of Adaptd ·
Anthropic has published data on how AI changed work inside the company: more output per person, fewer junior technical hires, and human attention as the new bottleneck.
The short version
- Anthropic published research on 16 October 2026 about how AI agents changed work, hiring and output inside the company.
- By August, its staff's agents worked six times as many hours as the staff, and junior technical hires fell from 27% to 9%.
- Plan for who checks AI output, not just who produces it, and hire for judgement.
Most of us worry about getting AI to do more. Anthropic's own numbers suggest the next problem is keeping up with what it's done.
What did Anthropic's employment study find?
Anthropic's study found its staff produce much more since adopting AI agents, while hiring has shifted away from junior technical roles. The research, published on 16 October 2026, uses the company's own data on employment, AI use, output and hiring, plus 17 staff interviews.
Code commits per engineer are about five times what they were six months earlier. Output also rose outside engineering: lawyers processed more contracts and salespeople closed more deals. The share of junior hires in technical teams fell from 27% to 9%.
What does this mean for the people doing the work?
Human attention is now the bottleneck, according to the study. Even for non-technical staff, reading everything their Claude agents produce would take half their working week, so people spend more time building ways to monitor agents.
Non-technical staff are also building their own tools. One salesperson built a custom dashboard in three to four hours, and an executive assistant built an agent for scheduling and email. Hiring managers now put more weight on autonomy and judgement than on technical know-how.
What should your business take from it?
Your business should plan for who checks AI output, not only for how much more it can produce. Anthropic is an AI company selling its own model, so treat its results as an early signal rather than a forecast for a 12-person firm.
My take: if your team starts producing twice as many quotes, posts or reports with AI, someone has to read them before customers do. Decide who that is now, and look for judgement when you hire.
More output is lovely. Unread output is just a bigger inbox.
Wondering what this means for your business?
That's exactly what the discovery call is for.