loading
loading
Anthropic just published a report on what AI has done to its own workforce. The authors used internal headcount data, agent session telemetry, calendar data, Slack patterns, and hiring scorecards, plus 17 employee interviews. Anthropic has rebuilt more of its workflows around agents than almost any company in B2B. That makes this the most detailed data available on what happens to hiring, output, and day-to-day work when a company goes this far. We run SaaStr with 3 humans and 20+ AI agents in production, and a lot of this matches what we see. We also had Anthropic’s sales leaders on stage at SaaStr AI 2026, and what they said there lines up with the report. Here are the five learnings that matter most for founders and execs. 1. Junior Technical Hires Fell From 27% to 9% of Technical Hiring Since early 2025, the share of junior employees in Anthropic’s technical departments fell from 21% to 14%. The share of new technical hires who were junior fell from 27% to 9%. Junior job postings have dropped especially sharply since December 2025. The shift started in January 2026, inside the teams that relied most on Claude. Writing code and finding bugs moved to agents first. One employee said a new model’s lift for performance engineers changed what they were hiring for almost overnight. By April 2026, engineering and research had deemphasized lower-band technical hiring. A majority of managers said they were hiring for a different set of skills because of AI. About a third of ML managers and a fifth of software engineering managers said they were hiring less because of AI. Across jobs, the more complex the Claude usage, the more junior hiring slowed between January and July 2026. Anthropic says directly that its worker protections don’t address this. Redeployment programs help people already on payroll. They do nothing for the junior engineer who never got an offer. 2. Headcount Tripled Anyway, and Revenue per Employee Is 2-4x What Google and Meta Did at Year Five Anthropic tripled in size in the last year. At five years old, it does 2 to 4 times the revenue per employee that Google and Meta did at the same age, adjusted to 2025 dollars. The mix of hires moved. Technical departments fell from 61% to 56% of headcount over the past year. Technical hiring still grew in absolute numbers, while sales, finance, and recruiting grew faster. Anthropic’s sales leaders described this from the SaaStr AI 2026 stage. Kelly Loftus scaled startup sales from fewer than 10 people to 150+ in 18 months, while the company went from about 250 to 1,300 employees. Her team still runs without traditional quotas. In her words, “we still don’t really have quotas, we have shadow targets.” Eleanor Dorfman explained why they couldn’t simply hire reps to keep up with demand. They could not absorb 3 to 5 times the reps fast enough. So she rebuilt the sales org in January 2026, and four months later 54% of new enterprise logos were coming in self-serve. That number is an outlier driven by extreme demand, and B2B companies shouldn’t plan against it. Administrative roles barely grew. Agent assistants, automated review, and outsourced ticket handling kept Anthropic light on clerical staff while it tripled. An executive assistant who built her own agent for scheduling and email triage said entry-level admin work was already automated, and her job had moved to chief-of-staff work. 3. Agent Sessions Spend 80% of Their Time Waiting on a Human In March 2026, total agent work hours at Anthropic passed human work hours for the first time. By August, counting only the agents employees directed themselves, agents worked 6 times as many hours as their humans. Code commits per engineer are running at 5x their level from six months earlier. Lawyers processed more contracts, accounts payable cleared more invoices, and sales reps closed more deals. The authors tracked 192 engineers adopting coding agents against 74 who hadn’t, and code output rose sharply after adoption. Agent-written code was deleted or reverted at the same rate as code the same engineer wrote by hand. Interactive Claude sessions spend about 80% of their time idle, waiting for a human to respond. One engineer said people are “bottlenecking the agents.” Reading load is a big part of it. Anthropic converted agent output into reading time at normal human speed. For technical teams the volume has climbed fast over six months. Even for non-technical teams, reading everything their agents produce would take more than half the work week. We see the same thing at SaaStr. With 20+ agents running, our own review and approval time sets the pace. A new agent we can’t review just adds to the backlog. The authors flag a caveat on causality. Engineers who moved to fully autonomous agents were already on a steep upward output trajectory before switching, and reps with more deal flow may have automated first. The productivity numbers are directional. 4. Standing Instructions for Agents Went From a Few Hundred to 8,500 in a Year Anthropic calls the work of managing agents meta-work, and it’s growing fast. Written standing instructions for agents grew from a few hundred in July 2025 to 8,500 a year later. Meta-work rose from 10% of the average employee’s interactive Claude sessions in fall 2025 to 22% in spring 2026. Most of the increase was agent management: specs for multi-agent runs, one agent reviewing another’s output, and pipeline monitoring. One engineer set up an agent chief of staff to triage requests from other agents. Another wrote a performance review for their Claude team. A lawyer moved from answering legal questions to writing frameworks that colleagues’ agents could find and apply on their own. Meetings went up. The median employee spends 20% more time in meetings than a year ago, and technical individual contributors are up 16%. Decisions that span teams still get made by humans in a room. Those decisions have to be written down. Agents can’t see a decision made in an unrecorded meeting or a DM, so they keep working from stale context. One employee whose name was on an old document about company swag kept getting agent requests for t-shirts and mugs long after the role changed. Anthropic’s response is to make decisions in open Slack threads and documents where agents can read them. 5. Interviewers Now Score Autonomy and Judgment Over Domain Expertise Anthropic ran its hiring manager scorecards since 2024 through an automated coder to see what interviewers comment on. Mentions of a candidate’s track record with AI tools tripled for technical roles and quadrupled for non-technical roles between 2024 and 2026. For technical candidates, comments on autonomy and judgment increased the most, and domain expertise shrank as a share of the assessment. For non-technical candidates, customer and mission orientation rose, and comments on execution skills fell. One manager gave the reason: models will stay uneven across domains, so the team needs people who can move between them. For a 2026 hiring plan, this means screening for how a candidate directs agents and how much they get done without being managed. Deep single-stack expertise and raw execution count for less, because agents now handle much of both. A Few More Interesting Learnings - Sales reps reached more customers. One AE said Claude automations took over about 20% of their prior work. Another said automating email and CRM work let them talk to 30% more customers. - Non-technical employees started building. Coding agent sessions for non-technical employees went from 2 per month in August 2025 to 90 per month in May 2026. One sales rep built a custom pipeline dashboard in three to four hours. - Managers ship code. Commits from managers doubled, and the engineer-to-PM ratio sits around 17 to 1 and is rising. - Internal moves jumped. Employees with a year of tenure were 50% more likely to switch departments each month in 2026 than in 2025. One product writing team moved entirely into writing prompts and evals for Claude. - Each new model raises the difficulty of tasks. After a model release, the share of very hard tasks given to agents rises about 30%, from 10% to 13%, and stays there. Easy tasks barely change. Anthropic Tied Agent Adoption to Equity and an Early Warning System for Job The report closes with two things Anthropic does for the people working alongside its agents. The first is broad stock ownership. Employees are expected to manage agents, make VP-level calls, and automate their own tasks, so they hold a financial stake in the outcome. The authors compare it to the 1950s union bargain, where workers accepted reorganization in exchange for pay tied to productivity. The second is an early warning system that tracks demand and productivity by role against AI capabilities. When a role starts shifting, the company moves people into roles with a skills fit before the old one disappears. That’s part of why internal moves are up 50%. Few B2B companies can offer Anthropic’s equity upside. Tracking which roles agents are absorbing, and planning where those people move next, costs almost nothing and can start this quarter. The full report is here: How employment is changing at Anthropic
document fetched 2026-10-20 02:51 · canonical https://saastr.com/5-interesting-learnings-from-anthropics-data-on-its-own-workforce-junior-technical-hiring-down-from-27-to-9-agents-working-6x-the-hours-of-humans-and-agents-waiting-on-a-human-80-of-the-time