🔍 Read the full analysis: A Better Way To Work With AI Agents: Ask For Decisions on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Thorsten Meyer’s October 27 article draws on Anthropic’s study of its own workforce to argue that human attention, not agent output, can become the bottleneck. He recommends counting review work, treating agent experiments as tests, setting stop rules for recurring tasks and limiting what coordinating agents bring to people.
Thorsten Meyer published a set of rules for managing AI agents on October 27, arguing that users should ask agents to bring forward decisions rather than a stream of updates. His advice follows an Anthropic study of its own employees, which reported that by August 2026, employee-directed agents logged six times as many hours as the people directing them and spent about 80% of session time waiting for human input.
Meyer’s central point is that generating an agent’s output may be quick, but reading, checking and acting on it still takes human time. He describes a pattern in which progress notifications, completed work and summaries accumulate into a separate management task. The article presents this as a practical concern for people using agents in publishing, software and other work, rather than as evidence that agents are universally unproductive.
The four rules Meyer proposes are to count the human work that follows an agent’s output; run small experiments but avoid turning every promising result into an ongoing project; set an end date and a “nothing new” stop condition for recurring tasks; and give coordinating agents an attention budget. Under that last rule, the default should be silence, with no more than three decisions a day, each accompanied by a recommendation, supporting evidence and the consequences of taking no action.
The article also points to Anthropic’s reported estimate that reading all agent output would take non-technical employees about half their working time. Meyer says the figures come from one company’s workforce and should be treated as a signal, not a general rule. He advises users to decide in advance who will read an agent’s work, who will check it and whether they want to carry out what it recommends.
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 Time
The guidance addresses a mismatch between the low apparent cost of asking an agent to produce more work and the continuing human cost of reviewing that work. If people cannot keep up with the results, faster generation may create queues rather than save time. The issue matters to teams adopting agents because the workload can shift from producing material to supervising tools, deciding which outputs deserve attention and checking whether conclusions are sound.
Meyer’s recommendations focus on reducing that downstream burden. A limit on daily decisions can make an agent’s requests more selective; a stop condition can prevent an automated task from running after it stops producing useful results. Requiring evidence and a recommendation also gives a person a clearer basis for review than a status update alone. These are proposals from Meyer, not outcomes tested or established by the Anthropic study.
Popular right now
Vitamix 5200 Professional-Grade Blender, 64 oz. Container, Black
KitchenAid Classic Series 4.5 Quart Tilt-Head Stand Mixer K45SS, Onyx Black
De'Longhi Linea Classic Espresso Machine with Adjustable Milk Frother
As an affiliate, we earn on qualifying purchases.
What Anthropic Measured
Anthropic published its workforce study on October 16. According to the figures summarized by Meyer, employee-directed agents logged six times the hours of the employees directing them by August 2026, while agents spent about 80% of session time waiting for people. The study also estimated that reading all agent output would take non-technical employees roughly half their work week.
Meyer reports that the study found the share of sessions spent on “meta-work,” mainly agent management, rose from 10% in fall 2025 to 22% in spring 2026. Written standing instructions for agents grew from a few hundred to 8,500 over a year. These numbers describe Anthropic’s own staff and the periods identified in the study; they do not establish the same pattern across other companies. Meyer notes that employees adopted the tools by choice and that parts of Anthropic’s work may not apply elsewhere.
The article also includes examples from businesses described in Anthropic customer material. A private chef said Claude helped draft proposals, while a painting contractor described checking the tool’s calculations against blueprint measurements. These are attributed customer accounts published by the vendor, not independent evidence of typical results. Their relevance to Meyer’s argument is that useful AI output still needs a person to assess it, particularly when errors could affect a bid or a business decision.
“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, in an account published by Anthropic
How Far the Findings Apply
The study described here covers Anthropic employees, not a representative sample of workers or organizations. Meyer says the authors themselves caution that employees chose to adopt the tools and that aspects of the company’s work may not generalize. The reported figures do not establish that agents in other workplaces spend the same share of time waiting or impose the same review burden.
The article also offers recommendations rather than results from a controlled test of the four rules. It does not provide evidence that a three-decision daily limit, silence by default or a particular stop condition will work for every team. Details on how Anthropic calculated the estimates are not set out in the supplied article text, so readers cannot assess the underlying method from this material alone.
How Teams Can Test These Rules
Meyer’s proposed next step is for users to apply the rules to the agents they already run: identify who must review each output, test ideas on a small scale, define when repeated tasks should stop and restrict which matters reach a human. Teams can then check whether fewer updates still surface the decisions that need attention. The article provides guidance, but it does not announce a formal rollout, product change or follow-up study.
For readers, the immediate question is whether the agents they use are reducing work or adding review and coordination tasks. The Anthropic figures offer one company’s account of that challenge; whether the same pattern holds elsewhere remains open.
Key Questions
What is Meyer’s main recommendation for working with AI agents?
He recommends making agents selective about what they bring to people: prioritize decisions with a recommendation, evidence and the consequence of doing nothing, rather than sending frequent status updates.
What did Anthropic report about agent waiting time?
The study, as summarized by Meyer, found that agents spent about 80% of employee-directed session time waiting for human input. The figure relates to Anthropic’s workforce and should not be assumed to describe every workplace.
Why does Meyer say more agent output can create more work?
People still have to read, check and act on outputs. Meyer argues that producing material cheaply does not make the human review and follow-up work disappear.
Are the four rules proven to improve productivity?
The article presents them as Meyer’s practical recommendations, not as findings from a controlled trial. It does not establish that they will produce the same results across teams or types of work.
Do Anthropic’s findings apply to all companies?
That is unclear. The study covered Anthropic employees, and Meyer notes that both the workforce and the company’s working practices may differ from those elsewhere.
Source: ThorstenMeyerAI.com
