🔍 Read the full analysis: Give AI Agents Clear Rules And Get Decisions Back on ThorstenMeyerAI.com
Get smart everyday buys 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
A report published October 27 uses Anthropic’s study of its own workforce to argue that human attention, not agent output, can become the limiting factor. It proposes four practices: count the review work, test many ideas but keep few, set clear stopping rules for recurring tasks, and limit an agent coordinator’s requests for decisions.
Thorsten Meyer published a set of four rules for managing AI agents on October 27, arguing that the time people spend reviewing agent output can outweigh the time saved by automation. His report cites Anthropic’s study of its own workforce, which found that employee-directed agents logged six times as many hours as the employees directing them by August 2026, while spending about 80 percent of session time waiting for human input.
Meyer’s central recommendation is to treat every agent result as a claim on human attention. Before assigning work, he says, users should identify who will read the output, who will act on it and whether the resulting work is worth doing. A task that takes an agent minutes to produce may still require substantial review by one or more people.
The other rules are to run small experiments without committing to every result, set end dates and stop conditions before making a task recurring, and give any coordinating agent a limited attention budget. Meyer recommends defaulting to silence and allowing a coordinator to bring forward no more than three decisions a day, each with a recommendation, supporting evidence and the consequence of taking no action.
Anthropic’s figures are observations from its own employees, not a measurement of all workplaces. The report says that reading all agent output would take non-technical staff about half their work week, according to its estimate. It also reports that the share of sessions devoted to meta-work, mostly agent management, rose from 10 percent in fall 2025 to 22 percent in spring 2026.
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
Why Agent Output Creates More Work
The advice addresses a practical problem for teams adopting agents: lower production costs do not remove the cost of review. When agents generate reports, proposals or alerts quickly, people still need to decide whether the information is accurate, useful and actionable. If those steps are not counted, automation can shift work rather than reduce it.
That distinction matters for small businesses and larger organizations alike. A useful proposal may lead directly to paid work, while a frequent report that nobody reads can consume time without helping anyone decide. Meyer’s rules offer a way to judge agent use by the human work that follows, rather than by output volume or speed alone. The recommendations are practical guidance, not evidence that the four rules have been tested in a controlled trial.
Popular right now
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Samsung 42-Inch S90H OLED 4K Smart TV Bundle
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
As an affiliate, we earn on qualifying purchases.
Anthropic’s Workplace Findings
Anthropic published the workforce study on October 16, 2026. Meyer’s October 27 report draws on its findings and emphasizes the gap between the hours agents are active in employee-directed sessions and the time they spend waiting for people. He notes that the study’s authors caution against treating the results as representative of every workplace: Anthropic employees adopted the tools by choice, and the company’s work may not match practices elsewhere.
The report also describes how agents can make exploration feel inexpensive. Users can test several ideas quickly, but each ongoing project may bring follow-up updates, reviews and maintenance. Meyer’s second rule is to keep experimentation separate from commitment: test broadly, then retain only what proves useful. His third rule applies the same discipline to recurring work, with an end date, a stop when there is nothing new, and a pause if nobody reads the output.
For checking consequential results, Meyer points to examples in an Anthropic customer story. Contractor Tony Severino said he found an error in an early AI-assisted bid and now asks the system to “show me your work.” The example illustrates a verification method, but it is a customer account published by the vendor, not an independent evaluation of the tool’s accuracy.
“An army of Claudes, with every one asking for my attention.”
— An engineer quoted in Meyer’s report
How Far the Findings Apply
The available evidence does not establish whether the same patterns hold across other companies, industries or kinds of agent use. Anthropic’s study looked at its own staff, and its authors said that voluntary adoption and company-specific work limit how widely the findings can be applied. The reported figures should be read as observations from one organization, not universal estimates.
Meyer’s four rules are recommendations drawn from those findings and his own work with agents. The source does not provide controlled evidence that adopting the rules reduces workload or improves business results. It also does not specify how Anthropic measured every reported category or establish whether the effects continued after the study period. The customer examples and the reported revenue figure are attributed accounts, not independently verified outcomes.
Testing the Rules in Practice
Meyer’s report does not announce a formal trial or a scheduled follow-up. Its proposed next step for users is to apply the rules to the agents they already run: identify who must review each output, decide when experiments become projects, define stopping conditions for recurring tasks, and cap the number of decisions a coordinator sends.
For organizations, the next useful evidence would be whether these practices reduce unread output and management time without preventing valuable work. Until broader findings are available, teams will need to track those costs locally and treat Anthropic’s measurements as a company-specific signal, not a forecast for every workplace.
Key Questions
What is the main point of Meyer’s report?
Meyer argues that human review and decision-making can become the bottleneck when AI agents produce work quickly. He recommends counting that follow-up work and limiting unnecessary agent updates.
What did Anthropic report about its own employees’ agents?
The study, published October 16, said employee-directed agents logged six times as many hours as their human directors by August 2026 and spent about 80 percent of session time waiting for human input. Those figures describe Anthropic’s workforce.
What are the four rules?
Count the human work that follows; run small tests but commit to few; set an end date and stop conditions for recurring jobs; and limit a coordinating agent’s requests for attention, with silence as its default.
Do the Anthropic findings apply to every workplace?
That is not established. The study examined Anthropic employees, and its authors cautioned that the company’s voluntary adoption and work practices may not represent other organizations.
Has Meyer shown that these rules reduce workload?
No controlled test is described in the report. The rules are recommendations; the source does not establish that they produce a measured reduction in workload or improved business outcomes.
Source: ThorstenMeyerAI.com
