🔍 Read the full analysis: What Changes When You Ask AI Agents For Decisions on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Anthropic’s October 16, 2026 study of its own workforce found that employee-directed agents logged six times as many hours as the people directing them by August, while spending about 80% of session time waiting for human input. The study points to a growing management and review burden; its figures come from one company and should not be treated as an economy-wide measure.
Anthropic’s study of its own workforce found that by August 2026, employee-directed AI agents logged six times as many hours as the employees directing them, yet spent about 80% of session time waiting for human input. The October 16 report highlights a practical limit on agent use: producing more automated work can also create more decisions, reviews and messages for people to handle.
The figures describe activity at Anthropic, the AI company that makes Claude, not a representative sample of workplaces. The study’s authors cautioned that employees adopted the tools by choice and that the company’s working practices may not transfer to other organizations. The findings are best read as evidence of a possible management burden, not a universal measure of how agents perform.
According to the report, about 80% of the time in employee-directed sessions was spent with agents waiting for a person. The study also estimated that reading all agent output would take non-technical employees about half their working time. The share of sessions devoted to “meta-work,” described as mostly managing agents, rose from 10% in fall 2025 to 22% in spring 2026. Anthropic also reported that written standing instructions for agents grew from a few hundred to 8,500 over a year.
The article drawing on the study argues that the apparent low cost of generating reports, proposals or analyses can conceal the time required to read, check and act on them. It recommends setting a clear purpose and endpoint for recurring tasks, testing ideas without automatically turning every promising result into an ongoing project, and limiting how often a coordinating agent interrupts a person. These are the author’s recommendations, rather than rules tested by the Anthropic study.
The new bottleneck
Agents don't get tired. Your attention does.
Anthropic measured how its own staff work with AI agents. The agents now do far more hours. The limit is the person who has to read, check and decide.
An agent drafts the report, the code or the summary in minutes.
- Reading
- Checking
- Deciding
- Fixing
- Explaining it to others
Illustration of the idea, not a measurement.
“An army of Claudes, with every one asking for my attention.”An Anthropic engineer, quoted in the study
Source: Anthropic Economics team, How employment is changing at Anthropic, 16 October 2026. Employee-directed interactive sessions; human time counted as 40 hours a week per head. One company studying its own staff, who chose to adopt the tools.
Powered by Thorsten Meyer AI
The Human Cost of Agent Output
The findings matter because agent productivity is not just a question of how quickly software completes a task. If each result requires a person to evaluate it, correct it or decide what to do next, the overall workload shifts rather than disappears. A fast-generated report can still consume hours across several reviewers.
For workers and managers, that changes how to judge whether an agent is useful. The relevant cost includes the follow-up work created by its output: who reads the result, who checks its accuracy, and who takes responsibility for acting on it. If those steps are not planned, a collection of individually reasonable updates can become a steady stream of interruptions.
The same issue affects the value of automation. A small business may benefit when an agent drafts a proposal that leads to paid work, while a recurring summary that no one uses may add overhead without a clear return. The distinction is not simply whether an agent can do a task, but whether the human work that follows is worth the outcome. Anthropic’s figures suggest this question deserves attention as organizations expand agent use.
Products Worth Considering
DJI Osmo Pocket 3 Filter Bundle – Vlogging Camera with 1 CMOS Sensor and 4K/120fps
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
How Agent Workloads Grew
Anthropic’s report follows a period of rapid uptake inside the company. Between fall 2025 and spring 2026, it recorded an increase in sessions categorized as meta-work, from 10% to 22%. It also reported that the number of written standing instructions for agents climbed to 8,500 over a year, compared with a few hundred earlier. These are measures of the company’s own operations and do not establish that every workplace will see the same pattern.
The source article pairs the workforce findings with examples from Anthropic customer stories. A private chef, Dan Ninerell, said Claude helped him prepare catering proposals and attributed $20,000 in business over a month and a half to that use; the figure is his account, published by the vendor, not an independently verified result. A contractor, Tony Severino, described finding an area-calculation error in an early blueprint analysis and said he now asks the system to show its work. These examples illustrate both potential benefits and the need for human checks, but do not establish typical results.
The article’s central recommendation is to treat human attention as a limited resource. It proposes three practical questions before assigning work: who will read the output, who will act on it, and whether the resulting work is wanted. It also suggests giving a coordinating agent a limited attention budget, with silence as its default and no more than three decisions to present in a day. Those suggestions are the author’s framework, not findings validated by Anthropic’s study.
“I made $20,000 in the last month and a half using Claude to do these professional proposals.”
— Dan Ninerell, owner of Modern Classical Chefs, in a customer story published by Anthropic
What the Study Cannot Establish
The reported figures come from one company’s workforce, and the authors themselves cautioned that its employees chose to adopt the tools and that Anthropic’s practices may not generalize. The source material does not provide enough detail to establish how the company calculated every measure, how agent sessions were defined, or whether the same trends hold across industries and job types.
The study also does not establish that agents caused a net increase in workload for every employee, or that the waiting time represents wasted time in every case. Some pauses may reflect necessary review or coordination. Nor do the customer anecdotes establish typical financial outcomes or the error rate of agent-generated work. The scale and distribution of the burden beyond Anthropic remain unclear, as does how much better-designed systems might reduce it.
How Organizations Measure Follow-Up
The source material does not identify a scheduled follow-up study or a specific next milestone from Anthropic. For organizations adopting agents, the immediate question is whether they track the human time spent reviewing, correcting and acting on outputs, not only the time agents spend generating them.
The article recommends setting stopping conditions before making a task recurring: an end date, a rule to stop when there is nothing new to report, and a pause when nobody reads the output. It also proposes asking a coordinating agent to bring forward a small number of decisions, with a recommendation and supporting evidence. Whether those practices reduce the burden will depend on the task and workplace; their effectiveness is not established by the study.
Key Questions
What did Anthropic’s study find about its AI agents?
Anthropic reported that by August 2026, employee-directed agents logged six times the hours of the employees directing them. It also said agents spent about 80% of session time waiting for human input.
Does this prove AI agents increase workload everywhere?
No. The study examined Anthropic’s own workforce, and its authors cautioned that the company’s employees and practices may not represent other workplaces. The findings indicate a possible attention burden, not a universal effect.
Why might agents create more work for people?
People may still need to read, verify and act on agent output. The source article argues that generating a result quickly does not remove the human effort required to decide whether it is accurate or useful.
What does the article recommend before assigning work to an agent?
It suggests deciding who will read the output, who will act on it, and whether that follow-up work is worthwhile. For recurring tasks, it also recommends setting an endpoint and a rule for stopping when reports are no longer useful.
Source: ThorstenMeyerAI.com
