Welcome, busybees
Anthropic counted the hours. In March 2026, the combined hours worked by its AI agents passed the hours worked by its human staff for the first time. By August, agents running in employee-directed sessions were logging 6 times as many hours as the people directing them.
Meanwhile, the share of junior hires in its technical teams fell from 27% to 9%. Anthropic published the data on October 16, in a report on how work is changing inside the company.
Here is what it says, and what it means for your job search.
1. The hours flipped
Agent work now outweighs human work inside Anthropic. Engineers produce about 5 times as many code commits as they did six months earlier. Among non-technical employees with no prior commits, coding-agent sessions went from 2 a month in August 2025 to 90 a month by May 2026.
The humans are now the bottleneck. About 80% of interactive Claude session time is spent idle, waiting for a person to respond. One engineer described the workload as “an army of Claudes, with every one asking for my attention.
2. The junior door narrowed
The junior share of technical employment fell from 21% to 14% since early 2025. The junior share of technical hires fell from 27% to 9%, and junior postings dropped sharply after December 2025.
About one-third of ML managers and one-fifth of software engineering managers said they were hiring less because of AI. One employee said the uplift for performance engineers “basically overnight changed what we were looking for.”
The report adds two cautions. Part of the drop may reflect normal growth in how the company is organized, and Anthropic’s own internal measures do not address the slowdown in junior hiring.
3. What hiring managers now screen for
Mentions of a candidate’s AI tool track record appeared in about 5% of technical and 7% of non-technical interview assessments in 2024. By 2026 those mentions had tripled for technical roles and quadrupled for non-technical roles.
In technical roles, the biggest rise was in comments about autonomy and judgment, while comments about domain expertise shrank as a share. In non-technical roles, customer and mission orientation rose and talk of execution skills fell.
4. What to do with this
Build a visible record of using AI on real work. One project in your portfolio beats a line on your resume.
Show your judgment. Note what the agent got wrong, what you changed, and why.
Learn the meta-work. Building agent tools and monitoring pipelines went from 10% to 22% of interactive sessions at Anthropic in about six months.
Ask interviewers how their team uses agents. The answer tells you whether the role is a growth path or a shrinking one.
My take: show hiring managers the work you directed an agent to finish, along with the judgment you used to check it.
Before you generalize
This is one company, and it builds the technology it studies. The productivity comparison covers 192 engineers who adopted agents and 74 who did not, and the authors flag selection bias and automated classifiers. Treat the numbers as a signal, not a forecast for every employer.
Subscribe to Incredible Edge and unlock the AI Career Playbook, free. It includes 100 companies hiring now, an automated job tracker, 100+ remote jobs, a Claude AI learning guide, an AI hiring agent workflow, an HR prompt stack, and the 2026 AI learning bundle.
That’s a wrap for today, busybees. Hit reply and tell me what you are applying for this week. See you tomorrow.
For me the scary number here is not 27% to 9%, it's what happens five years later. We also stopped hiring juniors, and I understand why everybody does it, an agent does the junior tasks faster. But seniors don't appear from nowhere, they are juniors who made a lot of mistakes on real projects. And the thing managers now look for, judgment, is exactly what you can't get without those years. I started to think that a junior today should learn to check the agent's work instead of writing the code, because that's where judgment grows now. Did the report say anything about how the juniors who still get hired are trained?