🔍 Read the full analysis: Four Rules For Getting Better Decisions From AI Agents on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A October 27 article by Thorsten Meyer argues that AI agents can create more work for the people directing them when their output requires constant review. Drawing on Anthropic’s study of its own workforce, Meyer proposes four practices: count the human work after an agent’s output, test ideas without adopting them all, set endpoints for recurring tasks, and limit how many decisions a coordinating agent brings forward.
Thorsten Meyer published four rules on October 27 for getting useful decisions from AI agents without letting their output overwhelm the people supervising them. The recommendations follow Anthropic’s October 16 study of its own workforce, which Meyer says found that employee-directed agents logged six times as many hours as their human directors by August 2026, while spending about 80 percent of session time waiting for human input.
Meyer’s central argument is that an agent’s low cost or quick output does not make the work free: people still have to read, check and act on results. He advises teams to ask, before starting an agent, who will review its work, who will use it and whether the resulting work is worth doing. A polished answer can still contain errors, he warns, and recommends asking agents to show their reasoning or distinguish assumptions from known information when the task requires checking.
The second rule is to test many ideas but adopt only a few. Agents can explore options quickly, Meyer writes, but each experiment that becomes an ongoing project adds review and maintenance work. His third rule is to set a task’s endpoint before making it recurring: include an end date, a stop condition for periods when there is nothing new, and a pause if people are not reading the output.
For a coordinating agent, Meyer’s fourth rule is to set an attention budget and make silence the default. He suggests bringing a person no more than three decisions a day, with each item including a recommendation, supporting evidence and the consequence of taking no action. The article also points readers to six copyable prompts, though the supplied source material does not include their full text.
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
Human Attention Sets the Limit
The advice addresses a gap between producing information and making a decision from it. When agents generate reports, alerts and proposals faster than people can review them, additional automation may shift the workload rather than remove it. Review capacity becomes a practical limit: an organization can have more agent output without gaining more useful decisions.
That matters for teams considering agent deployments because the cost of a task includes the downstream human time, not just the time or computing required to generate an answer. Meyer’s rules offer a way to assess that cost before a new workflow becomes routine. They also emphasize selective attention: agents should surface work that merits a decision, rather than treating every update as worthy of a person’s time.
The recommendations are guidance, not evidence that the four rules have been tested across organizations. Their value is as a practical framework for deciding which tasks to automate, which results to verify and when to stop a process that is no longer helping.
Products Worth Considering
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
Apple iPad Air 11-inch (M2), 512GB, Wi-Fi, Space Gray
Sonos Era 100 SL Compact Microphone-Free Speaker, Black
As an affiliate, we earn on qualifying purchases.
What Anthropic’s Workplace Study Reports
The figures in Meyer’s article come from Anthropic’s study of its own employees, published October 16, 2026. According to Meyer’s account of the study, employee-directed agents logged six times the hours of their directors by August 2026, and agents spent around 80 percent of session time waiting for a human. The study also estimated that reading all agent output would take non-technical employees about half their work week.
Meyer reports that sessions devoted to “meta-work,” described as mostly managing agents, rose from 10 percent in fall 2025 to 22 percent in spring 2026. He also says written standing instructions for agents grew from a few hundred to 8,500 over a year. These are figures from one company’s workforce, not a measurement of every workplace. The article notes that the study’s authors cautioned that employees adopted the tools by choice and that Anthropic’s work may not reflect practices elsewhere.
“The bottleneck is your attention, not the agent.”
— Thorsten Meyer
How Far the Findings Apply
The study covers Anthropic employees, so it does not establish how much time agents spend waiting or how much review they require in other workplaces. The article reports that the study’s authors themselves cautioned against treating their findings as representative of the wider economy. The supplied material does not include the full study methodology or detailed definitions for every reported measure.
It is also unclear whether Meyer’s four rules have been evaluated in a controlled comparison or how much time they save in practice. The article presents them as operating principles based on his experience and the cited workplace data, not as proven results. Further detail on the six prompts mentioned in the article is not provided in the supplied source text.
Testing the Rules in Daily Work
The immediate next step for readers is to apply the rules to a specific agent workflow: identify who owns review, decide what evidence is needed, set a stopping point and limit how many requests reach a person. Teams can then monitor whether the output is being used and whether the process reduces or adds human work.
No formal follow-up study or evaluation date is given in the source material. The broader question is whether these practices can keep agent output manageable as organizations increase their use of automated assistants; the available Anthropic figures alone do not answer that.
Key Questions
What are the four rules for managing AI agents?
Meyer recommends counting the human work required after an agent responds, testing ideas without adopting every experiment, setting an endpoint before making a task recurring, and limiting a coordinating agent’s requests for attention.
What did Anthropic’s study report?
As summarized by Meyer, the study reported that employee-directed agents logged six times as many hours as their directors by August 2026 and spent about 80 percent of session time waiting for human input. The data are from Anthropic’s own workforce.
Does the study prove that AI agents create more work everywhere?
No. The figures describe one company, and the article says the study’s authors cautioned that Anthropic’s employees and work may not represent the wider economy.
How many decisions should a coordinating agent send to a person?
Meyer suggests a maximum of three decisions a day. Each should include a recommendation, the evidence behind it and what may happen if the person does nothing.
What remains unproven about Meyer’s recommendations?
The supplied material does not show a controlled evaluation of the four rules or quantify the time they save. They are presented as practical guidance, not as independently tested results.
Source: ThorstenMeyerAI.com
