🔍 Read the full analysis: How To Get AI Agents To Flag Choices Instead Of Progress 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
An October 27 article argues that AI agents can create more work when they send frequent status updates and outputs that people must review. It draws on Anthropic’s study of its own workforce and proposes four practices: count the human work an agent creates, test ideas without committing to all of them, set stop conditions for recurring tasks, and limit agent notifications to a few decision-ready items.
In an article published October 27, Thorsten Meyer recommends that people using AI agents ask for decisions rather than routine progress updates, after Anthropic’s study of its own workforce found that agents spent about 80% of session time waiting for human input. The guidance matters as agent use expands: producing work may be quick, but reading, checking and acting on it still requires people’s time.
Meyer’s four rules are to count the human work that follows an agent’s output; run small tests without turning every promising result into an ongoing commitment; define an end date and stop conditions for recurring tasks; and give a coordinating agent an attention budget. Under the last rule, the default is silence, with no more than three daily notifications, each presenting a decision, a recommendation, supporting evidence and the consequence of taking no action.
The article reports figures from Anthropic’s research into its own employees’ use of agents. By August 2026, employee-directed agents were logging six times as many hours as the employees directing them, according to the article’s account of the study. It also says Anthropic estimated that reading all agent output would take non-technical employees about half their working time. These are findings from one company’s workforce, not measurements of all workplaces.
The article also points to changes in how employees managed agents: the share of sessions spent on “meta-work,” described as mostly agent management, rose from 10% in fall 2025 to 22% in spring 2026. Written standing instructions grew from a few hundred to 8,500 over a year, it says. Meyer’s central argument is that the cost of an agent’s work includes the time people spend reviewing and using its results, not just the time it takes to generate them.
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 Updates Need Limits
The advice addresses a practical problem for companies and individuals adopting agents: more automated output does not automatically mean less work. If every agent sends a digest, status note or list of possible next steps, users can spend their time processing updates instead of completing the work they delegated. Setting a notification limit and asking for evidence and a recommendation could help reserve human attention for decisions that need it.
Meyer’s first rule also shifts the measure of efficiency from how cheaply an agent produces a result to whether the result is useful enough to justify its follow-up. A fast report may still be costly if several people must review it. The approach does not establish that the four rules will work in every organization; it offers a set of practices for limiting review burdens while preserving human oversight.
Products Worth Considering
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
As an affiliate, we earn on qualifying purchases.
The Work Behind Agent Output
The article contrasts the low apparent cost of generating work with the continuing effort needed to read, verify and act on it. It gives examples from Anthropic customer stories: chef-business owner Dan Ninerell said Claude helped him prepare catering proposals and attributed $20,000 in revenue over a month and a half to that work. That is his account, published by the vendor, rather than an independently verified result.
It also describes contractor Tony Severino checking AI-assisted blueprint measurements after spotting an error in an early bid, and Rick Smith of Broadcast Blinds asking the system to distinguish assumptions from known information. These examples illustrate Meyer’s recommendation to ask agents to show their reasoning or flag uncertainty when a result will be checked by a person. The article notes that Anthropic itself cautioned that its employees adopted the tools by choice and that the company’s working practices may not generalize across the economy.
“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 Extend
Anthropic’s figures describe one company’s employees, and the source article says the study’s authors cautioned against treating the results as representative of the wider economy. The material does not provide enough detail here to assess how the study defined or measured every reported category, including agent hours and time spent reading output. It is also unclear whether Meyer’s four practices have been tested systematically or how much they reduce notification volume in different workplaces.
The customer revenue figure is attributed to the business owner through Anthropic, and the article provides no independent verification. More broadly, the evidence cited does not establish that all agent systems produce the same waiting patterns or review demands. The findings are best read as a signal about one organization’s experience, alongside practical recommendations whose effects may vary.
Testing the Four Working Rules
The article sets out recommendations rather than announcing a product, policy change or formal trial. Readers applying them can start by identifying who will read and verify an agent’s output, then set a clear end date and a “nothing new” stop for repeated tasks. For agent coordinators, Meyer’s suggested next step is to make silence the default and request only a small number of decision-ready alerts.
Further evidence would be needed to determine whether these practices measurably reduce review time without causing important information to be missed. Anthropic’s study offers a company-specific account of the attention costs; the supplied material does not describe a follow-up study or a planned next milestone.
Key Questions
What is the main recommendation?
Ask agents to surface decisions that need attention, rather than sending routine status updates. Meyer suggests a default of silence and no more than three decision-oriented notifications a day.
What did Anthropic’s study report?
As described in the article, employee-directed agents logged six times the hours of their human directors by August 2026, and spent about 80% of session time waiting for human input. The findings concern Anthropic’s own workforce.
What are the four rules?
Count the human work created by an output; run small tests without committing to every result; set end dates and stop conditions for recurring jobs; and limit agent notifications to a few items that include a recommendation and evidence.
Do the findings apply to every workplace?
That has not been established. The article says Anthropic’s study covered its own employees, who adopted the tools by choice, and cautions that the company’s working practices may not represent other workplaces.
Source: ThorstenMeyerAI.com
