Anthropic says it hires fewer juniors and meets more

Anthropic's internal report says junior technical hires fell from 27% to 9% while median meeting time rose 20% as AI agents took on routine work.
What it means for founders
- Hiring plans may tilt senior. If this pattern spreads, entry-level engineering roles get scarcer and candidates who can run agents with little oversight get pricier. Budget for fewer, costlier technical hires, and decide how you will train any juniors you take on.
- Attention becomes the scarce input. Agents that wait on a person most of the time are only as fast as whoever reviews them. Spend on specs, review habits and dashboards, not just more seats on Claude Code.
- Write decisions down. The report says agents lose context when choices happen in unlogged meetings. Recording decisions in shared channels is a cheap habit to adopt while your team is small.
- Expect the revenue per head question. Anthropic claims two to four times the revenue per employee that Meta and Google had at the same age, a figure that sits alongside its prospectus numbers. Investors may test your headcount plan against it. Watch whether other labs release comparable internal data.
The story
Anthropic published a study of its own workforce on Friday, October 16, reporting that it hires far fewer junior technical staff and that employees spend more time in meetings as AI agents take over coding and admin tasks. The report, How employment is changing at Anthropic, draws on internal headcount, calendar, survey and agent telemetry data plus 17 staff interviews. TechCrunch covered the findings on Saturday.
How Anthropic's hiring mix has shifted
Anthropic says junior employees made up 21% of its technical departments in early 2025 and 14% now. Among new technical hires the drop is steeper, from 27% to 9%, with junior postings falling hardest after December 2025. Technical teams also shrank as a share of all staff, from 61% to 56% over the year, though technical headcount still grew in absolute terms while sales, finance and recruiting grew faster.
The report ties part of that shift to non-engineers writing their own software. Coding agent sessions among employees with no prior commits climbed from 2 per month (August 2025) to 90 (May 2026), the company says.
Agents do more of the work, people do more coordinating
Anthropic says agents launched by individual staff logged six hours of runtime for every human work hour by August. Code commits per engineer are about five times the level of six months earlier, and the company says lawyers, salespeople and accounts payable staff also lifted output. Comparing what one engineer wrote by hand against what their agents wrote, it found no gap in how often lines were later deleted or reverted, though it cautions that extra code is not extra impact.
The cost lands on people. The report estimates Claude sessions sit idle around 80% of the time waiting for a human, and that reading every word agents produce would fill over half the week even for non-technical staff. Work spent steering agents, such as writing specs and building monitors, grew from 10% of interactive sessions (fall 2025) to 22% (spring 2026). Median meeting time is up 20% on a year ago, and 16% for technical individual contributors.
What we don't know yet
- How the figures would look at a company that is not selling the model it uses. Anthropic concedes tool adoption was not random and its business boomed over the same period.
- The absolute headcounts behind the percentages, which the report does not give.
- Whether the early warning system Anthropic is building to flag at-risk roles will ever be published.
Sources
Primary sources
Reporting
Enki Daily
Get stories like this every weekday morning.
The day's AI stories for founders, each with what it means for your company. Free.
More in Funding & Business
- Asia's Q3 startup funding hits $38.5B, 55% to AI

For founders: Seed money is easier to find than later rounds suggest. Seed and angel totals more than doubled year on year, so a credible AI angle can open early…
Crunchbase News · 1d ago - Replit makes OpenAI default, keeps Claude as option

For founders: Pin the model you trust. If you build with Replit Agent and liked what Claude produced, open the selector in Power or Max mode and choose it explicitly.
TechCrunch · 1d ago - Anthropic offers startups free Claude Team for a year

For founders: The dollar value is real, and most of it sits in the seats. Premium Team seats list at $100 per seat per month on annual billing, or $125 monthly, per…
TechCrunch · 11d ago - Lambda to raise up to $4 billion ahead of 2027 IPO

For founders: Capacity is being reserved by labs, not startups. When one AI lab can commit $35 billion to a single provider, a large share of new GPU capacity can be spoken…
TechCrunch · 11d ago - DeepSeek nears $12 billion round from CATL, Tencent

For founders: Low prices look funded, not temporary. A war chest of $12 billion or more, plus compute it plans to own, makes it more likely that DeepSeek keeps undercutting…
The Decoder · 11d ago