Special · Developer jobs

Even the company building AI is hiring fewer juniors: inside Anthropic's drop from 27% to 9%

"Anthropic is on track to stop hiring technical juniors by 2027." That one line, posted on X on Oct 17, passed 4,400 likes and 710,000 views. Another followed: "someone still has to read all that output. reading boring code and fixing it is how most of us learned. no idea where the next seniors come from." The source is a report Anthropic's economics team published on Oct 16, 'How employment is changing at Anthropic', built from HR records, agent logs, calendars, interview scorecards and 17 employee interviews. We read it end to end and sorted what the numbers show.

1. Junior share of technical hires: 27% to 9%

   Juniors fell from 27% of new technical hires in early 2025 to 9%, and from 21% to 14% of technical staff; junior postings dropped sharply from December 2025. Talk of lower junior demand surfaced internally in January 2026, when new models gave performance engineers a big uplift and "basically overnight changed what we were looking for." In an April survey, about a third of ML managers and a fifth of SWE managers said they were hiring less because of AI.

2. Agent hours passed human hours in March, and hit 6x by August

   Summed autonomous and interactive agent hours first exceeded employee hours in March 2026. Counting only employee-directed interactive sessions, agents worked six times as many hours as their humans by August, assuming 40-hour human weeks and excluding agent waits over 30 minutes. The company tripled in a year, yet revenue per employee runs two to four times what Meta and Google managed at similar points.

3. Commits per engineer up 5x, and agent code wasn't reverted more often

   Code commits per engineer are five times what they were six months earlier. Comparing 192 engineers who adopted coding agents with 74 who didn't, added code jumped after adoption. Code written by the same person's agent in the same month and directory was no more likely to be deleted or reverted. Salespeople closed more deals and lawyers processed more contracts after adopting agents.

4. 80% of Claude session time is spent waiting on a human

   Humans are now the bottleneck: in interactive sessions Claude spends about 80% of its time idle, waiting for input, and even non-technical staff would need over half a work week to read all its output. One engineer compared it to babysitting, where the kid who needs a juice box opened sounds as urgent as the one who cut a finger. In April, 20% said their role had shifted to managing Claudes, and standing agent instructions grew from a few hundred to 8,500 in a year.

5. What interviewers write about now: autonomy and judgment

   Across thousands of interview scorecards since 2024, mentions of candidates' AI tool use rose from about 5% (technical) and 7% (non-technical) to three and four times those levels by 2026. Autonomy and judgment saw the biggest gains in technical assessments while domain expertise shrank. "I expect models to continue to be jagged, so for my team I want them to be fungible across domains," one manager said. Meeting time is up 20%.

6. Entry-level admin is already past, and internal moves are up 50%

   Administrative roles that usually scale with a company barely grew. An executive assistant who built a scheduling and email agent said "we've already moved past the point of Claude augmenting away the really entry-level admin type role." Employees with at least a year's tenure were 50% more likely each month to switch departments than in 2025. The team that wrote Claude.ai's greetings moved into model evals and prompts.

7. X says "done by 2027", but the report names no such date

   The "stop hiring juniors by 2027" post passed 710,000 views, and "The first phases of pyramid replacement are happening at Anthropic" drew 1,461 likes. The report itself names no 2027 date. Shopify CEO Tobi Lütke posted the same chart from Shopify's internal data (786 likes). One founder wrote that the grads he knew who joined Anthropic "all did CRAZY amount of open work before they got in." On YouTube, AI Copium's breakdown passed 78,000 views in a day.

8. Korea asks the same question: "report drafts were a class for new hires"

   MBC's PD Su-cheop aired 'The case of the missing new hires' on Oct 20, analyzing 10.41 million job postings and finding AI-exposed jobs favor experienced hires; its trailer quotes a big-company HR team saying "give us more AI tokens, not new hires." Per Edaily, employment among 15- to 29-year-olds averaged 43.8% in January to August, down 28 months running. Seoul National University professor Lee Chan said research and report drafts "were cheap work for the company but a class for new hires," and that if no one trains juniors, it will be hard to grow the middle managers who judge AI's output in ten years.

9. One company's data, and the limits Anthropic admits

   The report flags its limits: explosive revenue growth makes it hard to credit all productivity gains to AI, early adopters may already have been on the rise, and many of its practices can't be copied across the economy. Anthropic's response is an early warning system for roles at risk and internal redeployment. Employees expect AI to be able to do 75% of their work by 2030; other Americans said 28%.