🔍 Read the full analysis: How To Make Your AI Agent Check-Ins Worthwhile on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Anthropic’s study of its own workforce found that employee-directed agents spent about 80% of session time waiting for human input, while staff faced a large volume of output to review. Thorsten Meyer’s October 27 report recommends budgeting for the human work agents create, testing ideas without adopting all of them, setting stop rules for recurring tasks and limiting a coordinating agent’s alerts.
In an October 27 report, Thorsten Meyer set out four rules for making AI agent check-ins useful, drawing on an Anthropic study of its own workforce that found agents spent about 80% of session time waiting for human input. The findings point to a practical problem for people using fleets of agents: producing more output does not save time if people cannot review, verify and act on it.
Meyer’s report cites Anthropic’s figures for employee-directed agent sessions. By August 2026, those agents were logging six times as many hours as the employees directing them. The study also estimated that reading all agent output would take non-technical employees about half their working time. These are findings from one company, not a measure of all workplaces or agent systems.
The report’s first rule is to count the human work that follows an agent’s output: who will read it, who will check it and who will act on it, a concern also explored in managing AI agent access. A draft that is quick for an agent to produce may still require substantial review from several people. Meyer argues that the relevant cost is not only generation time but the attention required afterward.
The next two rules are to run small experiments without turning every promising result into a standing project, and to decide how recurring tasks end before automating them. Meyer recommends an end date, a stop condition for periods when there is nothing new, and a pause if nobody is reading the results. His fourth rule is to give a coordinating agent an attention budget: default to silence and surface no more than three daily decisions, each with a recommendation, evidence and the consequence of taking no action—an approach relevant to coordinating AI agent swarms.
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 Cost of Agent Output
The report’s central concern is that human attention can become the constraint even when agents make tasks faster to produce. Every notification or digest creates a possible review obligation. If a person cannot check the output or decide what to do with it, automation may shift work from production to monitoring rather than remove it.
That trade-off matters to teams adopting agents for research, drafting or recurring checks. A large volume of low-cost output can still consume staff time and create risk if polished results go unverified. Meyer’s suggested limits are practical management rules, not proof that they will work in every organization; they make the expected review and decision work visible before a task becomes routine.
Products Worth Considering
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
Dell UltraSharp 27 Thunderbolt Hub Monitor (U2726DER)
Dell UltraSharp U2725QE 27-Inch 4K UHD IPS Black Monitor
As an affiliate, we earn on qualifying purchases.
What Anthropic Measured
Anthropic published its workforce study on October 16, 2026, according to Meyer’s report. It tracked employee-directed agent sessions and described rising coordination work: the share of sessions spent on “meta-work,” mostly managing agents, increased from 10% in fall 2025 to 22% in spring 2026. The report also says written standing instructions for agents grew from a few hundred to 8,500 over a year.
Anthropic’s study has limits. The workforce chose to adopt the tools, and the company’s work practices may not reflect those of other sectors. Meyer presents the figures as a signal rather than a universal rule. His examples also show why verification matters: he describes a contractor checking an agent’s blueprint calculations and a business owner asking an agent to distinguish assumptions from known information.
“an army of Claudes, with every one asking for my attention.”
— An engineer quoted in Meyer’s report
Limits of the Workforce Findings
The study covers Anthropic employees, and the material does not establish that the same proportions apply elsewhere. The report notes voluntary tool adoption and differences in how the company works. It also does not provide enough detail here to determine how waiting time varies by task, agent configuration or team.
Meyer’s four rules are recommendations based on the reported findings and work examples, not results from a controlled test of those rules. The claimed business earnings in one customer story are attributed to the business owner and vendor material. The report does not independently verify that figure or establish that agent use caused it.
Set Limits Before Automating
Meyer’s recommendations point users toward setting review responsibilities and stop conditions before launching recurring agents. Teams can specify who owns an output, when an agent should stay quiet, what counts as “nothing new,” and when a recurring task should be paused or ended.
The next measure of whether an agent is helping is not simply how much it produces. Users will need to see whether the output is read, checked and acted on without creating more coordination than the task warrants. The report does not name a follow-up study or a formal rollout of these rules; broader evidence on how they perform across workplaces remains open.
Key Questions
What did Anthropic’s study find about agent waiting time?
As summarized by Meyer, agents in employee-directed sessions spent about 80% of session time waiting for human input. The finding comes from Anthropic’s own workforce and should not be treated as a universal estimate.
What is an attention budget for an AI agent?
In Meyer’s recommendation, a coordinating agent defaults to silence and raises at most three decisions a day. Each should include a recommendation, supporting evidence and what happens if the user does nothing.
How should teams handle recurring agent tasks?
Meyer recommends deciding on an end date and stop conditions in advance. A task can pause when there is nothing new to report or when no one is reading its output.
Does the report prove that AI agents save time?
No. It describes Anthropic’s employee-use findings and Meyer’s practical recommendations; it does not establish that agents save time across organizations. The human time needed to review and use output remains part of the workload.
Source: ThorstenMeyerAI.com
