🔍 Read the full analysis: From Status Updates To Decisions: My Rules For AI Agents on ThorstenMeyerAI.com
Get hardware and tech essentials 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
Thorsten Meyer argues that AI agents can create substantial follow-up work for the people using them, and proposes four rules to manage that burden. His October 27 article cites Anthropic’s study of its own workforce, which reported agents often waiting for human input and estimated that reviewing all agent output could take non-technical staff about half their work week.
Thorsten Meyer has published four rules for using AI agents, arguing that the time saved generating output can be offset by the human work needed to review it, check it and act on it. His October 27 article cites an Anthropic study of the company’s own workforce, which reported that employee-directed agents spent about 80% of session time waiting for human input.
Meyer’s rules are to count the work that follows an agent’s output, use agents for low-cost tests but commit only to ideas that prove useful, set an end point before making tasks recurring, and give a coordinating agent a limited attention budget. He proposes that the coordinator default to silence and raise no more than three decisions a day, each with a recommendation, supporting evidence and the consequences of taking no action.
The Anthropic figures are presented as evidence of a possible workload problem, not a universal measure. According to Meyer’s account of the study, by August 2026 employee-directed agents logged six times as many hours as the employees directing them. The study estimated that reading all agent output would take non-technical employees about half their work week. It also found that the share of sessions spent on “meta-work,” mostly managing agents, rose from 10% in fall 2025 to 22% in spring 2026.
Meyer says the practical cost of an agent is not just the time it takes to produce a report or proposal. It also includes the time required to verify and use that work. He recommends asking in advance who will read the output, who will act on it and whether the resulting work is worth taking on. For tasks involving calculations or other consequential details, his examples include asking an agent to show its work or distinguish what it knows from what it is assuming.
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 Needs a Human Budget
Meyer’s argument is that automation can make producing information cheap without making human review cheap. If agents generate frequent summaries, alerts and drafts, people still have to decide which deserve attention. That can shift the bottleneck from producing work to checking and prioritizing it.
The suggested rules are aimed at limiting that hidden workload. Requiring a stop condition for recurring tasks can prevent low-value updates from continuing indefinitely. A cap on escalations can also force an agent system to distinguish decisions that need a person from routine status messages. These are Meyer’s recommendations; the source material does not report a controlled test showing that the rules reduce workload.
The stakes vary by use. A draft that supports a task a business already wants to do may be useful, while a polished result that contains an error can create extra checking or lead to a poor decision. Meyer’s emphasis on assigning review and accountability before work begins offers readers a way to assess whether an agent is saving time overall, rather than simply producing more output.
Products Worth Considering
LG UltraGear 45GX900A-B 45-Inch Curved OLED Gaming Monitor
LG UltraGear 32GX870A-B 32-inch 4K OLED Gaming Monitor
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
What Anthropic’s Workforce Study Found
The Anthropic research discussed by Meyer was published on October 16, 2026 and examined the company’s own employees’ use of agents. Meyer notes that the study’s authors cautioned against treating their results as representative of all workplaces: Anthropic employees adopted the tools by choice, and the company’s working practices may not transfer to other organizations.
The article also gives examples from Anthropic customer stories. Private chef Dan Ninerell told Anthropic that he made $20,000 in the previous month and a half using Claude to draft catering proposals. That is his account, published by the vendor, rather than an independently verified result. Contractor Tony Severino described finding an error in Claude’s early blueprint measurements and said he now asks the system to “show me your work.” Meyer uses the examples to illustrate how review can be directed toward the parts of an output where errors matter.
The article’s focus is not a new agent product or a formal workplace policy. It is Meyer’s personal operating advice for publishing and software projects. He presents agents as useful for exploring possibilities, while warning that each successful experiment can become a continuing commitment with review and maintenance costs.
““an army of Claudes, with every one asking for my attention.””
— An engineer quoted in Meyer’s article
How Far the Findings Apply
The Anthropic results concern one company’s workforce. Meyer reports that the study’s authors said employees adopted agents by choice and that aspects of Anthropic’s work cannot be copied across the economy. The figures therefore do not establish how much time agent oversight takes in other workplaces, or whether the same pattern holds across different tools and tasks.
The article does not provide a measured evaluation of Meyer’s four rules, so it remains unclear how much they would reduce review time or improve outcomes. The $20,000 business result is Ninerell’s own account as published by Anthropic, and the article does not describe independent verification. More broadly, the balance between time saved and time spent checking will depend on the quality of an agent’s output and the consequences of errors.
Testing the Rules in Daily Work
Meyer presents the four rules as practices people can use with agents they already have, rather than as a change that depends on a new product. His recommended next step is to apply a three-question check before starting a task: who will read the result, who will act on it, and whether that follow-up work is wanted.
For recurring tasks, his advice is to set an end date, stop when there is nothing new to report, and pause updates when nobody is reading them. For a coordinating agent, he proposes limiting interruptions to three decision requests per day. The source does not identify a formal rollout, further study or specific publication date for results from testing these practices.
Key Questions
What are Meyer’s four rules for using AI agents?
He recommends counting the human work that follows an agent’s output, running small tests without committing to every idea, setting an end point for recurring tasks, and limiting how many decisions a coordinating agent sends to a person.
What did Anthropic report about agents waiting for people?
According to Meyer’s account of Anthropic’s study, agents spent about 80% of session time waiting for human input. The study examined Anthropic employees, so that figure should not be assumed to describe every workplace.
Did the Anthropic study show that AI agents save no time?
No. The findings cited in the article describe agent use and the time involved in managing and reviewing output. They do not establish that agents provide no time savings; the overall effect may depend on the task and the follow-up work it creates.
How does Meyer suggest limiting agent interruptions?
He proposes that a coordinating agent default to silence and raise at most three decisions a day. Each request should include a recommendation, evidence and what may happen if the person takes no action.
Source: ThorstenMeyerAI.com
