🔍 Read the full analysis: AI Agents That Bring Decisions: Four Rules I Rely On 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
An Oct. 27 article by Thorsten Meyer argues that AI agents can shift work onto the people who must read, check and act on their output. It offers four operating rules, drawing on an Anthropic study that reported substantial agent waiting time and rising agent-management work among its employees.
Thorsten Meyer published four rules on October 27 for managing AI agents without letting their output create more work for the people directing them. His article draws on an Anthropic study of its own workforce, which reported that employee-directed agents logged six times as many hours as their human directors by August 2026, while spending about 80% of session time waiting for human input.
Meyer’s central point is that generating an answer is only part of the workload: people may still need to read, verify and act on the result. He cites the Anthropic study’s estimate that reading all agent output would take non-technical employees about half their working time. These are findings from one company’s workforce, not a general measure of how agents perform across workplaces.
The article’s four rules are to count the human work that follows an output; run small experiments but keep only those that prove useful; set an end date and stopping conditions before making a task recurring; and give a coordinating agent a limited attention budget. Under Meyer’s proposed approach, that agent should default to silence and bring forward no more than three decisions a day, each with a recommendation, supporting evidence and the consequence of taking no action.
Meyer illustrates the first rule with examples from Anthropic customer stories. Chef-business owner Dan Ninerell told Anthropic that Claude helped him prepare catering proposals and said he made $20,000 in a month and a half using them. That is Ninerell’s account, published by the AI vendor. The article also describes contractor Tony Severino checking an agent’s blueprint measurements after catching an error in an early bid, and cites another business owner, Rick Smith, asking an agent to distinguish assumptions from known information.
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 Human Cost of Agent Output
The article focuses on a practical problem for organizations adopting agents: more automated output does not automatically mean less human work. If each agent creates reports, alerts or suggested follow-up tasks, someone still has to decide which items matter. That workload can reduce the time available for the underlying job and can make a supposedly efficient system difficult to sustain.
The figures Meyer cites offer a reason to measure agent use by more than output volume or operating time. Anthropic’s reported rise in agent-management work—from 10% of sessions in fall 2025 to 22% in spring 2026—suggests that supervising agents can itself become a larger part of work. The result is relevant to teams deciding whether to automate recurring tasks, how many agents to run and who will review their work.
Meyer’s proposed limits are operational advice, not a tested guarantee of improved productivity. Their value is in making the human obligations visible: who reads the result, who checks it and who acts. Those questions can expose when a cheap-to-generate report carries a costly review burden.
Products Worth Considering
DJI Osmo Mobile 8 3-Axis Smartphone Gimbal Stabilizer
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
As an affiliate, we earn on qualifying purchases.
What Anthropic’s Study Measured
The research cited by Meyer was published by Anthropic on October 16, 2026, and examined how the company’s employees worked with agents. Meyer reports that by August, employee-directed agents were logging six times the hours of the people directing them. About 80% of session time was spent waiting for a person, according to the account of the study in his article.
The study also tracked the growth of written standing instructions for agents, from a few hundred to 8,500 in a year, and the share of sessions devoted to meta-work, described mainly as managing agents. Anthropic’s work is a case study of its own staff. Meyer notes that employees adopted the tools by choice and that aspects of the company’s work may not apply elsewhere, so the results should not be treated as an economy-wide estimate.
The article extends its four-rule framework beyond the study’s measurements. It recommends small trials, clear stopping conditions and constrained notifications as ways to reduce unnecessary review and follow-up. Those are Meyer’s suggested practices; the supplied material does not report a controlled test showing that the rules produce a specific productivity gain.
“human attention is now the bottleneck for agent output.”
— Anthropic’s study, as described by Meyer
How Far the Findings Apply
The figures describe Anthropic employees and their work patterns. The material does not establish whether similar agent-to-human time ratios or reading burdens apply in other companies, sectors or types of work. Anthropic’s employees chose to adopt the tools, and the company’s work may differ from that of other organizations.
It is also not clear from the supplied material how the study calculated the estimated reading time, how much agent output employees ultimately ignored, or whether the increased share of meta-work reduced overall productivity. Ninerell’s revenue statement is attributed to him through an Anthropic customer story and should be understood as an individual account, not a verified causal estimate of AI’s effect on revenue.
Meyer’s rules are presented as practices he uses, but the article does not provide comparative trial results showing how much they reduce workload or improve decisions. The effects will depend on the task, the cost of errors and whether people actually act on the agent’s output.
Set Limits Before Tasks Repeat
Meyer recommends that teams apply the four rules before expanding agent use: identify who owns review and follow-up, test narrowly, define an end date and a “nothing new” stopping condition, and limit how often a coordinating agent interrupts people. He also proposes pausing recurring work when nobody reads its output. These are suggested operating steps, not a timetable announced by Anthropic.
The supplied article does not identify a follow-up study or a new product release tied to the rules. The next practical test for readers is whether a task’s useful decisions justify the human time spent checking its results. The wider question—whether the patterns in Anthropic’s workforce appear elsewhere—remains open.
Key Questions
What are the four rules for managing AI agents?
Meyer says to count the human work after generation, run small tests but keep only those that prove useful, define an end date and stopping conditions for recurring tasks, and limit a coordinating agent’s interruptions and daily decisions.
What did Anthropic report about agent use?
As described by Meyer, Anthropic’s study of its own employees found that by August 2026, employee-directed agents logged six times the hours of their human directors and spent about 80% of session time waiting for human input. The findings concern Anthropic’s workforce, not all workplaces.
Does the study prove that AI agents reduce productivity?
No. The supplied material reports agent waiting time, reading estimates and a rise in agent-management work, but does not establish a general productivity loss. It also does not show that Meyer’s rules produce a measured productivity gain.
Why does Meyer recommend limiting agent updates?
Each update may require someone to read, check or act on it. Meyer argues that those human tasks carry a cost even when the agent creates its output quickly, so recurring reports and alerts should stop when they no longer provide useful decisions.
Source: ThorstenMeyerAI.com
