🔍 Read the full analysis: Four Guidelines For Getting More Than Updates From AI Agents on ThorstenMeyerAI.com
TL;DR
Thorsten Meyer’s October 27 article argues that AI agents can create more follow-up work than they save, pointing to Anthropic research that found agents often waited for human input. His four guidelines are to count the human work after generation, test ideas without adopting every one, set end conditions for recurring tasks and limit how many decisions a coordinating agent sends.
Thorsten Meyer published four guidelines for managing AI agents on October 27, arguing that their output can create a substantial burden of reading, checking and decision-making for the people using them. His recommendations draw on an Anthropic study of its own workforce, which reported that employee-directed agents spent about 80% of session time waiting for human input.
Meyer’s first rule is to count the human work that follows an agent’s output. Before assigning a task, he recommends identifying who will read the result, who will act on it and whether the resulting work is worth doing. His point is that generating a document quickly does not make the review or implementation free.
His second rule is to let agents test multiple ideas while committing to only a few. Experiments may be inexpensive to run, Meyer writes, but every idea kept can bring ongoing review, updates and maintenance. The third rule is to set an end condition before making a task recurring: specify when it stops, what counts as “nothing new,” and when to pause if nobody is reading the results.
The fourth rule is to give a coordinating agent an attention budget. Meyer recommends silence as the default and suggests limiting the agent to three decisions a day, each accompanied by a recommendation, supporting evidence and the consequence of taking no action. The article also recommends asking agents to show their work or distinguish between known facts and assumptions when accuracy matters.
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
The recommendations address a practical gap between cheap, fast generation and the human capacity to evaluate what agents produce. An agent can make it easier to create reports, proposals or alerts, but those outputs still require someone to determine whether they are accurate, useful and worth acting on.
That distinction matters as organizations add agents to existing workflows. If every output is treated as something that deserves attention, more automation may increase the number of decisions people must make rather than reduce their workload. Meyer’s rules offer a way to manage that risk: define the follow-up work, stop low-value recurring tasks and make agents surface only items that need a decision.
The article’s suggested limit of three decisions per day is Meyer’s proposed practice, not a result established by the Anthropic study. Its usefulness will depend on the task, the stakes and how much information a person needs to make a sound decision.
Popular right now
iRobot Roomba Max 715 Pet Robot Vacuum with AutoEmpty Dock
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
LG C2 Series 65-Inch OLED evo 4K Smart TV (OLED65C2PUA, 2022)
As an affiliate, we earn on qualifying purchases.
What Anthropic’s Workforce Study Found
Meyer cites Anthropic research published October 16, 2026, about how the company’s own employees used agents. According to the figures he reports, by August 2026 employee-directed agents logged six times as many hours as their human directors. The agents spent about 80% of session time waiting for a person, and Anthropic estimated that reading all agent output would take non-technical employees about half their work week.
The study also found that the share of sessions involving “meta-work,” mostly managing agents, rose from 10% in fall 2025 to 22% in spring 2026. Written standing instructions for agents increased from a few hundred to 8,500 over a year, according to Meyer’s account. These numbers describe Anthropic’s workforce and should not be treated as representative of every company or job.
Meyer notes that Anthropic’s researchers cautioned that employees adopted the tools by choice and that the company’s work practices may not transfer to the broader economy. The figures therefore provide a case study, not a universal benchmark. Meyer also describes examples from Anthropic customer stories, including business owners using Claude to draft proposals or inspect blueprints; those anecdotes are accounts published by the vendor, not independent evaluations of performance.
„An army of Claudes, with every one asking for my attention.“
— An engineer quoted in Meyer’s article
How Far the Findings Apply
The available figures come from Anthropic employees using the company’s tools. Meyer says the researchers themselves cautioned that voluntary adoption and company-specific work practices limit how far the findings can be generalized. The article does not establish whether other organizations would see the same agent waiting times, reading burden or increase in management work.
It also does not report controlled evidence that Meyer’s four rules reduce workload or improve outcomes. The proposed attention limit and other practices are recommendations based on his experience and interpretation of the research. Their effects have not been quantified in the material provided, and the article does not specify a common measure for deciding whether an agent’s output is worth its review cost.
Testing the Rules in Daily Work
Meyer’s next step for readers is to apply the guidelines to the agents they already use: identify the people who must review and act on outputs, keep experiments separate from ongoing commitments, and give recurring tasks clear stop conditions. For coordinating agents, his proposal is to test silence as the default and reserve notifications for a small number of decisions that include evidence and a recommendation.
The article does not announce a new product, policy or scheduled follow-up. Whether these practices help will depend on how teams use agents and whether they track the time spent reviewing results. The broader question raised by the Anthropic figures—how much human attention agent workflows consume—remains a matter for further measurement across different workplaces.
Key Questions
What are Meyer’s four guidelines for using AI agents?
He recommends counting the human work after an output, testing many ideas but adopting only a few, setting end conditions for recurring tasks, and giving coordinating agents a limit on the decisions and updates they send.
What did Anthropic report about agents waiting for people?
As summarized by Meyer, Anthropic’s study found that agents in employee-directed sessions spent about 80% of session time waiting for human input. The result concerns Anthropic’s own workforce, not all workplaces.
Did the study prove that AI agents increase everyone’s workload?
No. The reported results describe one company’s workforce, and the researchers cautioned that its practices may not transfer to the wider economy. Meyer uses the figures to frame a concern about human attention and follow-up work, not as proof of a universal effect.
What does an agent attention budget mean?
It means limiting how often a coordinating agent interrupts people. Meyer suggests that it default to silence and raise no more than three decisions per day, with a recommendation, evidence and the consequence of doing nothing. This is his proposed practice, not a tested standard from the study.
Source: ThorstenMeyerAI.com
