Anthropic's Shift towards AI-Driven Workflows

Anthropic, a company at the forefront of artificial intelligence adoption, is undergoing significant changes in its employment and work organization. The company has been hiring for fewer junior technical positions as hours worked by AI agents relative to employees have rapidly expanded. By August 2023, interactive AI agent hours outnumbered employee hours six to one.

The shift towards more extensive use of AI agents has led to increased productivity across various job functions at Anthropic. Engineers and non-technical employees alike have seen a significant leap in output when they adopted coding agents. However, this acceleration also creates bottlenecks, making human attention the limiting factor for agent output.

Employees are now turning to ‘meta-work,’ which involves building visibility tools and monitoring agent pipelines, rather than focusing on core tasks. This change is reflected in shifting priorities among hiring managers, who increasingly emphasize autonomy and judgment over technical domain expertise and implementation skills.

The company’s internal data reveals that even as Anthropic grows, relative hiring for junior technical positions has slowed down. Productivity gains are evident across both engineering and non-technical roles when employees adopted coding agents. However, the increased use of AI also raises questions about what constitutes valuable work in an organization driven by automation.

Hiring managers at Anthropic now prioritize applicants’ autonomy and judgment over specific technical expertise. This shift is a defining feature of working in an AI-heavy organization, where constant pressure exists to streamline oneself out of bottleneck positions when possible. Employees remain critical but are adapting their contributions accordingly.

Anthropic has responded to these changes by setting up an early warning system to track jobs at risk of disruption and prioritizing internal redeployments as roles evolve. These actions demonstrate how specific organizational commitments can shape the impact of AI on employment, raising difficult questions about what those commitments should be.

The company’s labor demand is expanding rapidly, but Anthropic has hired less per dollar of revenue than earlier generations of large tech companies. At five years old, Anthropic takes in between 2-4 times as much revenue per employee as Meta and Google did during their own rapid growth phases.

Within technical departments at Anthropic, the share of junior-level employees slipped from 21% to 14% since early 2025. The proportion of junior technical hires fell even faster, from 27% to 9%. Postings for juniors have fallen particularly sharply since December 2025.

The overall employment share of technical departments within Anthropic has also decreased, from 61% to 56%, in the last year. However, absolute hiring in sales, finance, and recruiting has outpaced it. This shift is partly due to natural expansion as the company scales but may also reflect coding tools allowing non-technical employees to make software themselves.

Coding-focused AI agent sessions for non-technical employees leapt from 2 per month in August 2025 to 90 in May 2026. Yet this shift in hiring composition is not solely about automated coding diffusing to other domain experts. Within technical departments, the share of employees not actively coding fell by almost half over the year through April 2026.

The number of commits from managers has doubled, and supervising roles increasingly involve building and shipping code. Non-technical hiring has accelerated in areas like sales but remains slow for administrative functions that typically scale with a company’s size. A combination of agent assistants and automated review services has left Anthropic clerical-light even as it grows rapidly.

Even clerical work tightly connected to in-person interaction has grown slowly, with employees focusing on higher-order tasks rather than entry-level admin roles. Reallocation is evident, with employees working at the company for at least a year 50% more likely to move across departments each month in 2026 compared to 2025.

Whole teams have been reallocated as they handed off their work to agents. One product design team developed a context graph that guided agents to write compelling product content, raising questions about what was left for human writers. The team then rotated into agent prompting and began writing model evals and prompts for Claude.

Humans are now a thin layer atop a growing contingent of AI agents at Anthropic. Based on agent session telemetry, in March 2026, the sum of autonomous and interactive agent work hours leapt past human work hours for the first time. Part of this is fleets of agents conducting industrial-scale software engineering.

Even narrowing in on the agents deployed by individual employees, as shown in Figure 2, by August the agents worked six times as many hours as their human handlers. Anthropic’s vast bulk of tokens flowing through these agents is spent producing software code, with dashboards monitoring fleets of agents operating across the company.

The share of tasks above the 90th percentile of difficulty for old models rises sharply by around 30% (from 10% to 13% of all tasks) and stays elevated in the weeks following model releases. Spiraling agent productivity translates into output pitched across humans’ screens and attention, with one engineer comparing this to babysitting.

In interactive sessions, agents spend a vast majority of their time waiting for human input rather than working actively. We track idle time by subtracting each agent’s active working time from each session’s total wall-clock time. Claude sessions amount to around 80% idly waiting for human input.

To read all of Claude’s output from interactive sessions, employees would need half a workweek dedicated solely to reading it. Technical functions have seen the volume of agent output tokens increase significantly since early 2025.