🔍 Read the full analysis: The Rules I Use To Get Useful Decisions From AI Agents on ThorstenMeyerAI.com
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
TL;DR
Thorsten Meyer’s October 27 column proposes four rules for using AI agents without turning their output into extra work: count the human effort that follows, experiment without committing to every idea, set stopping rules for recurring tasks, and limit what coordinating agents send to people. The article cites Anthropic’s October 16 study of its own workforce, which found substantial agent waiting time and a growing burden of agent management; those figures describe Anthropic, not the wider economy.
Thorsten Meyer published four rules on October 27 for getting useful decisions from AI agents, arguing that the time needed to read, check and act on their work can outweigh the effort saved by generating it. His column draws on Anthropic’s study of its own workforce, which reported that employee-directed agents logged six times as many hours as the employees directing them by August 2026, while spending about 80 percent 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 asking who will read the result, who will act on it and whether the resulting work is worth doing. A report that takes an agent minutes to produce may still require substantial staff time to review. His point is that low production cost does not mean low total cost.
His second rule is to run small experiments but commit to only a few. Agents can test ideas quickly, but each successful experiment can create a continuing obligation to monitor, review and maintain it. Meyer warns that treating every promising result as a project can leave people managing a growing list of tasks they never deliberately chose.
The remaining rules address ongoing work and notifications. Meyer says recurring tasks should have an end date, a stop condition for “nothing new,” and a pause if nobody reads the results. For an agent that coordinates other agents, he recommends silence by default and no more than three decisions a day, each presented with a recommendation, supporting evidence and the consequence of taking no action. These are the author’s operating rules, not findings tested by the Anthropic study.
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 Consumes Attention
The advice addresses a practical mismatch: agents can generate more material than people have time to evaluate. Meyer cites Anthropic’s estimate that reading all agent output would take even non-technical employees about half their working time. If that estimate applies to a person’s workload, adding more agents may increase the review queue rather than free time.
The column also puts a useful distinction between automation and delegation. A task is not fully handed off when a person must repeatedly answer questions, verify results and decide what to do next. Meyer’s proposed limits aim to reserve human attention for decisions that matter, while allowing agents to handle low-cost exploration. Readers should treat the rules as practical guidance from the author, not as a proven productivity method.
Popular right now
Philips 3300 Series Fully Automatic Espresso Machine with LatteGo Milk System
DeLonghi Rivelia Automatic Espresso Machine Bundle with Grinder and Frother, Beige
Vitamix Propel Series 750 Professional-Grade Blender, 4 Programs, Black
As an affiliate, we earn on qualifying purchases.
The Anthropic Study Behind the Rules
Anthropic published its workforce study on October 16, 2026. As summarized by Meyer, employee-directed agents logged six times the hours of their human directors by August 2026, and agents spent about 80 percent of session time waiting for people. The study also estimated that reading agent output would take non-technical employees roughly half their work week.
Meyer reports that the share of sessions spent on “meta-work,” mostly managing agents, increased from 10 percent in fall 2025 to 22 percent in spring 2026. He also says written standing instructions for agents grew from a few hundred to 8,500 over a year. Anthropic’s results come from its own staff, who adopted the tools by choice; the study’s authors cautioned that the company’s working practices may not represent other workplaces.
To illustrate his first rule, Meyer cites customer stories published by Anthropic. Chef Dan Ninerell said Claude helped him draft catering proposals and attributed $20,000 in earnings over a month and a half to that work. Contractor Tony Severino described catching an error in an early blueprint estimate and then asking the model to “show me your work.” These are individual accounts published by the vendor, not independently verified measurements of typical results.
“an army of Claudes, with every one asking for my attention.”
— An engineer quoted in Meyer’s column
How Far the Findings Apply
The Anthropic figures describe one company, and the study’s authors said its workforce and voluntary adoption may not reflect other organizations. The source material does not provide enough detail to independently assess the study’s methods or determine how its estimates would change across different industries, roles or agent systems.
It is also unclear whether Meyer’s four rules have been evaluated in a controlled test or how much time they save in practice. The customer revenue and error-checking examples are attributed accounts, not a representative sample. The column offers a framework for managing agent work, but does not establish that every team will benefit from the same limits or notification thresholds.
Testing the Rules in Daily Work
Meyer’s article does not announce a formal trial or provide a date for follow-up findings. The immediate next step is for readers applying the approach to track review time, useful decisions and recurring-task workload alongside the time agents spend producing results. That would help distinguish tasks that genuinely reduce effort from those that shift it to oversight.
For now, Meyer recommends setting an end point before making a task recurring and giving coordinating agents a narrow channel for decisions. Whether those practices reduce the attention burden beyond his own work remains unreported. Anthropic’s study provides a company-specific signal; broader evidence would be needed to show how commonly the same pattern occurs.
Key Questions
What are Meyer’s four rules for using AI agents?
He recommends counting the human work after an agent responds, running small tests without committing to every idea, setting end dates and stop conditions for recurring jobs, and limiting a coordinating agent’s notifications.
What did Anthropic’s study report?
As summarized by Meyer, employee-directed agents logged six times the hours of their human directors by August 2026 and spent about 80 percent of session time waiting for human input. The findings concern Anthropic’s own workforce.
Does the study prove AI agents create more work than they save?
No. The figures point to time spent waiting and managing agents at Anthropic, but the material does not establish a net productivity effect across the company or other workplaces.
What does Meyer mean by an attention budget?
He proposes that a coordinating agent stay silent by default and bring people no more than three decisions a day, each with a recommendation, evidence and the consequence of doing nothing.
Source: ThorstenMeyerAI.com
