🔍 Read the full analysis: Can AI Agents Work With Less Hand-Holding? My Four Rules on ThorstenMeyerAI.com
Get the latest gadgets 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’s October 27 article proposes four rules for using AI agents without letting their outputs become a second job: count the review and follow-up work, test widely but commit selectively, set stopping conditions for recurring tasks, and limit what a coordinating agent sends to a person. The piece cites Anthropic’s study of its workforce as a warning signal, while noting that its results come from one company and may not apply broadly.
In an October 27 article, Thorsten Meyer outlined four rules for using AI agents with less day-to-day oversight, arguing that human attention is the limiting resource when agents produce work faster than people can review it. The guidance follows an Anthropic study of its own workforce, which Meyer says found that employee-directed agents spent about 80 percent of session time waiting for human input.
Meyer’s rules are to count the human work that follows an agent’s output, run small experiments without turning every promising result into a permanent project, set an end point before making a task recurring, and give a coordinating agent a strict attention budget. He suggests that such an agent 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.
The figures Meyer cites come from Anthropic’s study of its own employees, published October 16. According to his account, by August 2026 employee-directed agents were logging 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 working time. Meyer also reports that the share of sessions spent on agent-management “meta-work” rose from 10 percent in fall 2025 to 22 percent in spring 2026.
The article pairs those figures with practical examples. Meyer says a private chef used Claude to prepare catering proposals and reported making $20,000 over a month and a half using them; that is the business owner’s account, published by Anthropic, not an independently verified result. Meyer also describes a painting contractor who found an error in an early blueprint analysis and now asks the system to show its work. The examples illustrate his broader point: an agent’s output still needs a reader, a check and sometimes follow-up action.
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 Can Cost More
The article focuses on a practical gap between the low cost of generating work and the time required to use it. An agent can produce a report, proposal or alert quickly, but someone may still have to read it, verify its assumptions and decide what to do. If those steps are not counted, automation can shift work rather than remove it.
That distinction matters for teams deciding whether to expand agent use. Meyer’s attention-budget rule treats notifications and summaries as demands on a person, not as free by-products of software. Limiting routine updates could leave people more time for decisions that require judgment, while stopping recurring tasks that nobody uses could reduce ongoing review and maintenance.
The recommendations are the author’s operating rules, not a controlled test showing that each one improves productivity. Their value is as a checklist for identifying hidden human effort before a team builds more agent workflows.
Popular right now
Dyson Ball Animal 2 Plus Corded Vacuum
Breville Bambino Espresso Machine, Brushed Stainless Steel
Samsung 42-Inch Class OLED S90F 4K Smart TV (2025)
As an affiliate, we earn on qualifying purchases.
The Anthropic Study Behind the Rules
Meyer presents Anthropic’s internal workforce research as a reason to pay attention to the human side of agent use. The figures he reports describe the company’s employees and their working sessions, rather than a representative survey of organizations across different industries. He notes that Anthropic employees chose to adopt the tools and that the company’s working methods may not transfer to other workplaces.
The article also says written standing instructions for agents grew from a few hundred to 8,500 over a year. Meyer interprets the study as showing that people were spending more effort managing agents as their use expanded. The reported six-to-one hours figure and the estimates of waiting and reading time are study findings as presented in the source article; they should not be treated as universal benchmarks.
Meyer’s four rules address the workflow surrounding agents rather than a particular model or product. They apply to tasks that produce documents, recommendations or repeated updates: decide who will review the result, treat experiments as experiments, specify when automation should stop, and reserve a person’s attention for a limited number of decisions.
“The bottleneck is your attention, not the agent.”
— Thorsten Meyer
How Far the Findings Apply
The source material describes Anthropic’s study but does not provide its full methodology, sample size or detailed definitions for each measure. It is therefore unclear how the reported figures were calculated or how well they generalize beyond Anthropic. Meyer explicitly cautions that the company’s employees adopted the tools by choice and that its work practices may differ from those of other organizations.
The article also does not report a controlled evaluation of the four rules or quantify how much time they save. The business examples are attributed accounts, including one financial figure shared through Anthropic; the material does not independently verify the outcomes. It remains unclear whether limiting an agent to three decisions a day, or using the other suggested practices, would suit every team or type of work.
Testing the Rules in Daily Work
Meyer’s article offers the rules as practices readers can apply to their existing agents, rather than announcing a new product or formal trial. Readers can begin by identifying who reviews each output, setting a clear end date for recurring tasks, and stopping updates that are not being read or acted on.
The next useful evidence would come from teams tracking the time spent reviewing agent output before and after adopting such limits. The source material gives no follow-up study or scheduled milestone, so whether these practices reduce oversight work remains to be tested.
Key Questions
What are the four rules for using AI agents?
Meyer recommends counting the human work after an agent produces an output, running small tests without committing to every idea, setting an end point and stop conditions for recurring tasks, and limiting how many decisions a coordinating agent sends to a person.
What did Anthropic’s study report about agent waiting time?
As summarized by Meyer, the study found that agents in employee-directed sessions spent about 80 percent of the time waiting for human input. The finding concerns Anthropic’s own workforce and is not presented as an industry-wide estimate.
Did Anthropic prove that AI agents save employees time?
The supplied material does not establish that. It reports measures of agent use and waiting time, and Meyer argues that reading, checking and acting on outputs can create substantial human work. The article does not provide a controlled comparison of net time saved.
Why does Meyer recommend setting an end date for recurring agent tasks?
He argues that automated tasks can continue producing updates after they stop being useful. An end date, a rule for stopping when there is nothing new, and a pause when nobody reads the output can prevent unused tasks from adding ongoing review work.
Are the article’s business examples independently verified?
No independent verification is described in the source material. The $20,000 figure is a business owner’s account published by Anthropic, and the blueprint example is presented as an anecdote about checking an agent’s work.
Source: ThorstenMeyerAI.com
