On October 18, details emerged from an internal Anthropic report on how using AI agents has changed work inside the company itself. The numbers are stark: by August 2026, total agent work hours (autonomous and interactive combined) had overtaken employee hours by roughly 6 to 1, with agents having first passed humans on this measure back in March. At the same time, the share of junior hires in technical departments fell from 27% to 9%, and one engineer summed up daily life in the report with the line — "an army of Claudes, each one asking for my attention." Code commits per engineer rose fivefold over six months.
Inside the company that sells agents, agents are reshaping its own hiring
There's a detail worth not missing here: this isn't an outside research report on the labor market in general — it's Anthropic's own data about itself. The company both sells AI agents to customers and runs on them internally, which gives it a direct incentive to show that agents work well. That doesn't make the numbers untrue, but it does call for a correction: Anthropic isn't a neutral observer, it's a market participant with a story it benefits from telling this way. To its credit, the report is honest about the flip side too: the share of "meta-work" — writing specs, reviewing agent output, monitoring pipelines — rose from 10% of interactive sessions in fall 2025 to 22% by spring 2026. And agent sessions themselves, per the report, spend roughly 80% of their time waiting on a human response. In other words, the bottleneck hasn't disappeared — it has simply moved from "who writes the code" to "who can respond to the agent in time and check its work."
What this means for hiring at a 15-person company, not a 15,000-person one
Anthropic — a company that tripled in size over the period in question and can afford to rebuild hiring entirely around "autonomy and judgment" instead of narrow technical expertise — is exactly what the report says hiring managers are now looking for more of, in both technical and non-technical roles. For a small business — say, a 12-person Portuguese web development agency — transplanting that logic directly is risky. That kind of company has no budget for a dedicated "agent operator" whose only job is checking agent output; instead, that checking usually falls on the same senior person who was supposed to be writing the code in the first place. If an agent spends 80% of its time waiting on a human, and it's always the same human, the hours saved don't automatically translate into payroll savings — especially in the first months of rollout.
The practical takeaway for small and mid-sized businesses here isn't "hire fewer juniors" — it's "work out who on your team is spending time checking and explaining tasks, to a person or to an agent." If, at an 8-person accounting firm, one employee already spends two hours a day re-explaining a client's context to an agent for every new task, the formal speed gain gets eaten by those two hours, and "hire one more junior for the routine work" can end up cheaper than it looks on paper. The report also shows that agent sessions among Anthropic's non-technical staff rose from two per month in August 2025 to 90 by May 2026 — meaning agents aren't just moving into engineering, but into marketing, support, and admin too, so this workload recalculation applies to any function with repetitive tasks, not just IT.
Honestly, the biggest risk in this story isn't the agents themselves — it's smaller companies starting to copy hiring decisions from firms like Anthropic without adjusting for scale. 6 to 1 is a ratio calculated for an organization with a massive agent infrastructure and roles dedicated to supporting it; applying it literally to a ten-person team is a methodological mistake that can cost real money in hiring or headcount decisions.
At Dayava, we roll out agent-based automation with exactly this kind of workload recalculation in mind — not to produce a nice ratio for a report, but so the hours a team actually frees up go toward work an agent genuinely can't close on its own.
If a similar solution interests you for your business, leave a request at dayava.pt/contactos/.
Source: Dealroom, "Anthropic's AI agents outwork staff 6-to-1 as junior tech hiring collapses"