🔍 Read the full analysis: Four Practical Rules For Working With AI Agents on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A report published by ThorstenMeyerAI.com on October 27 describes four rules for using AI agents without letting their output overwhelm the people supervising them. It cites Anthropic’s study of its own workforce, which found that agents often waited for human input and that reviewing their output could take substantial staff time. The findings are company-specific, not a universal measure of agent use.
ThorstenMeyerAI.com published four practical rules on October 27 for working with AI agents, drawing on Anthropic’s study of its own employees to argue that human attention—not the speed of generating output—can be the limiting factor. The recommendations are to count the work required to review and use an agent’s output, treat experiments as tests rather than commitments, set end conditions for recurring tasks, and limit how often a coordinating agent interrupts people.
The article reports that by August 2026, agents directed by Anthropic employees logged six times as many hours as the employees directing them. It also says agents spent about 80 percent of session time waiting for human input. These figures describe Anthropic’s internal work, rather than a representative survey of businesses or workers.
The report says Anthropic estimated that reading all agent output would take non-technical employees about half their working time. It also cites a rise in the share of sessions spent on “meta-work,” mostly agent management, from 10 percent in fall 2025 to 22 percent in spring 2026. These measures suggest that the work of supervising agents can absorb time even when the agents themselves produce results quickly.
Meyer’s first rule is to decide who will read, check and act on an output before assigning the task. His other rules advise teams to test multiple ideas but adopt only those that prove useful; give repeated jobs a defined end date and stop conditions; and set an attention budget for a coordinating agent. In the proposed approach, that agent stays quiet by default and brings forward no more than three decisions a day, each with a recommendation, supporting evidence and the consequence of taking no 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 Costs Staff Time
The recommendations address a gap between the low cost of generating an answer and the often higher cost of deciding whether it is reliable or useful. An agent can produce a report, proposal or alert quickly, but a person may still need to read it, verify its claims and decide what to do. If those steps are not assigned or budgeted, the task has not disappeared; some of its work has shifted to review.
That distinction matters for organizations adopting agents across several teams. A stream of individually reasonable updates can collectively compete with core work, particularly when agents send information that has no clear owner or action. Meyer’s rules focus on controlling that downstream workload: make the human work visible, keep experiments reversible, end recurring tasks when they stop helping, and reserve notifications for decisions that need attention.
The article also makes a practical case for checking high-risk details rather than treating fluent output as proof of accuracy. It recounts a contractor spotting an incorrect floor-area calculation in an agent-assisted bid and describes asking the system to show its work. Those examples are presented as user experiences, not independent tests of a product’s accuracy.
Popular right now
Ninja CREAMi 7-in-1 Ice Cream Maker, NC301
De'Longhi Linea Classic Espresso Machine with Adjustable Milk Frother
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
As an affiliate, we earn on qualifying purchases.
Anthropic’s Internal Agent Findings
The advice is based in part on an Anthropic study published October 16, as described in Meyer’s article. The report says the study examined the company’s own workforce and found that employees voluntarily adopted the tools. Anthropic’s researchers cautioned that the company’s work and adoption patterns may not carry over to the wider economy. The figures should be read as a case study, not a general benchmark for every workplace.
The article also cites customer stories published by Anthropic. A private chef, Dan Ninerell, said he used Claude to draft catering proposals and attributed $20,000 in revenue over a month and a half to that work. That is the customer’s account, published by the vendor, rather than independently verified revenue data. Contractor Tony Severino said he found an incorrect area calculation in an early blueprint-related bid and now asks the system to show its work. Meyer uses the examples to distinguish output that leads to desired paid work from output that creates unwanted review or correction.
The four rules are Meyer’s recommendations, not reported findings from a controlled trial of management practices. The article also refers to an estimate that people can hold about four items in mind at once, citing Cowan (2001), but its central workplace figures come from Anthropic’s own study.
“I made $20,000 in the last month and a half using Claude to do these professional proposals,”
— Dan Ninerell, owner of Modern Classical Chefs, as quoted by Anthropic
How Far the Findings Apply
The figures reported for Anthropic’s workforce do not establish how agents are used across other companies, sectors or job types. The study’s authors, according to Meyer’s account, said employees chose to adopt the tools and that aspects of Anthropic’s work may not be transferable. The source material does not provide enough methodological detail to assess the study’s sampling, measurement choices or how waiting time was defined.
It is also unclear whether the four rules reduce total staff time or improve business outcomes when applied in practice. The article offers recommendations and examples, but no controlled comparison showing what happened to teams that adopted them. The customer revenue and accuracy anecdotes are attributed accounts, not independently audited results. Readers should treat the rules as proposed operating practices rather than proven guarantees.
Apply the Rules to Active Agents
Meyer’s proposed next step is an operational check before starting or continuing each agent: identify who will read its output, who will act on it, and whether the work that follows is worth creating. For recurring tasks, teams can set an end date, define when the agent should stop because it has found nothing new, and pause the task if nobody reads its reports.
For organizations applying the approach, the practical test is whether these limits reduce unwanted reviews and interruptions without suppressing useful work. The source article does not announce a formal rollout, later study or scheduled measurement. Further evidence about whether the rules change agent use or staff workload is not provided.
Key Questions
What are the four rules for working with AI agents?
Count the human work that follows an agent’s output; run small tests but commit only to ideas that prove useful; set end dates and stop conditions for recurring tasks; and limit a coordinating agent’s interruptions with an attention budget.
What did Anthropic’s study report?
As summarized by ThorstenMeyerAI.com, by August 2026 employee-directed agents logged six times the hours of their directors and spent about 80 percent of session time waiting for human input. The reported figures concern Anthropic’s own workforce.
Do Anthropic’s findings represent all workplaces?
No. The article says Anthropic’s study examined its own employees, who adopted the tools by choice, and that the company’s work may not be representative of the wider economy.
What does an attention budget mean?
In Meyer’s recommendation, a coordinating agent stays quiet by default and raises no more than three decisions a day. Each should include a recommendation, evidence and what may happen if the person takes no action.
Are the four rules proven to save time?
The source presents them as practical recommendations, not results from a controlled trial. It does not provide evidence that applying the rules reduces total workload across organizations.
Source: ThorstenMeyerAI.com
