🔍 Read the full analysis: The 4 Rules I Use To Keep AI Agents Moving Without Me on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Thorsten Meyer’s October 27 essay offers four practices for using AI agents without letting their updates and outputs create a second job: count follow-up work, test ideas before committing, set stop conditions and limit what coordinating agents bring to people. The recommendations draw in part on Anthropic’s study of its own workforce, which Meyer says found agents spent about 80% of employee-directed session time waiting for human input.
In an essay published October 27, Thorsten Meyer outlined four rules for keeping AI agents from adding more work than they save: count the human follow-up, test ideas without automatically adopting them, set an end point for recurring jobs and limit the decisions a coordinating agent can bring to a person. The recommendations respond to a finding Meyer attributes to Anthropic’s workplace study: agents in employee-directed sessions spent about 80% of their time waiting for human input.
Meyer’s central point is that faster generation does not make the work around an output disappear. A person may still need to read, verify and act on what an agent produces. His first rule is to ask, before starting a task, who will review the result, who will use it and whether the work it creates is worth doing. He argues that the human cost can outweigh the apparent savings when an output requires substantial review.
The second rule is to use agents for small, low-cost experiments but keep only the ideas that prove useful. Meyer warns that a test can become an ongoing commitment if it is adopted without deciding who will maintain it. His third rule is to define when a recurring task ends, including a stop when it finds nothing new and a pause if nobody is reading its results.
Under the fourth rule, a coordinating agent should default to silence and use a limited attention budget. Meyer proposes that it bring no more than three decisions per day, each with a recommendation, supporting evidence and an explanation of what happens if the person does nothing. These are Meyer’s suggested practices, not reported results from a controlled test of the four rules.
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
How Agent Output Uses Human Time
The essay addresses a practical problem for teams adopting agents: assigning more work to software can increase the volume of material people must supervise. Meyer’s argument is that the scarce resource is attention, not the ability to generate another report, draft or update. If employees cannot review outputs or make decisions on them, automation may shift work rather than remove it.
That distinction matters for organizations evaluating whether agents are saving time. Counting only how quickly an agent finishes a task misses the hours needed to check its work, manage the process and respond to its messages. Meyer’s proposed limits aim to make those human costs visible before a task becomes routine. The rules may help readers frame that problem, but the essay does not establish that they will produce a particular productivity gain.
Popular right now
Dyson Car+Boat Cordless Handheld Vacuum
Dyson Ball Animal 3 Extra Upright Vacuum
Breville Bambino Espresso Machine, Brushed Stainless Steel
As an affiliate, we earn on qualifying purchases.
What Anthropic’s Workplace Study Reports
Meyer cites an Anthropic study published October 16 that examined the company’s own employees’ use of agents. According to the figures in his essay, by August 2026 employee-directed agents logged six times as many hours as the employees directing them. He also reports that even non-technical staff would need about half their working time to read all the agents’ output, according to the study’s estimate.
The essay says the share of sessions spent on “meta-work,” mostly managing agents, rose from 10% in fall 2025 to 22% in spring 2026. It also says written standing instructions for agents grew from a few hundred to 8,500 over a year. These are figures from a single company’s workforce, as relayed by Meyer; they should not be read as measurements of all workplaces.
Meyer acknowledges that limitation, noting that Anthropic employees adopted the tools by choice and that the company’s working practices may not transfer to other organizations. The study offers evidence about one setting, while the four rules are Meyer’s own recommendations for his publishing and software work.
“an army of Claudes, with every one asking for my attention.”
— An engineer quoted in Meyer’s essay
Limits of the Evidence and Advice
The figures Meyer cites come from Anthropic’s own workforce, and the essay does not provide enough detail to determine how the findings would apply across industries, company sizes or different agent systems. The study’s reported measures also do not, by themselves, show whether the agents increased overall productivity or reduced total work.
Meyer presents the four rules as practices he uses, not as a tested framework with measured outcomes. The essay does not quantify how much time his approach saves, how the suggested three-decision limit was chosen or whether that limit suits different roles. The customer examples are attributed accounts, and Ninerell’s revenue claim is not independently verified in the source material. It remains unclear how widely these practices are being adopted or what results they produce outside Meyer’s own work.
Applying the Rules to Agent Work
Meyer’s immediate recommendation is to apply the rules before adding an agent or making a task recurring: identify who will review the output, decide what evidence would justify continuing an experiment, set a stop condition and agree on a limit for interruptions. Those steps can be used as a trial checklist, but the essay does not announce a formal rollout, product or follow-up study.
For organizations, the next useful measure would be to track not only agent activity but also the time people spend reviewing, correcting and managing results. Whether Meyer’s attention limits reduce that burden remains an open question. More evidence across workplaces would be needed to show which practices work best and under what conditions.
Key Questions
What are Meyer’s four rules for managing AI agents?
He recommends counting the human work that follows an agent’s output, running small tests without committing to every idea, setting end conditions for recurring jobs and limiting the interruptions made by coordinating agents.
What did the Anthropic study find about agents waiting for people?
Meyer reports that Anthropic’s study found agents spent about 80% of employee-directed session time waiting for human input. The finding concerns Anthropic’s workforce, not all workplaces.
Does the essay prove that these rules improve productivity?
No. Meyer describes them as practices he uses. The source material does not report a controlled evaluation or measured productivity gains from applying the four rules.
Why does Meyer recommend stopping recurring agent tasks?
He argues that repeated tasks can keep producing updates even when they find nothing new or nobody reads the results. An end date and a “nothing new” stop can prevent a low-value task from becoming ongoing work.
Source: ThorstenMeyerAI.com
