🔍 Read the full analysis: My Approach To AI Agents: Fewer Updates, Clearer Decisions 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
Thorsten Meyer’s October 27 article argues that AI agents can shift work from producing outputs to reviewing and managing them. He points to an Anthropic study reporting long agent wait times and substantial reading demands, then recommends four practices to limit unnecessary updates and decisions.
In an article published October 27, Thorsten Meyer argued that people using AI agents should limit routine updates and focus on decisions that require human attention, citing Anthropic research that found agents spent about 80 percent of session time waiting for people. Meyer’s advice matters as organizations add agents to work: producing more automated output does not remove the time needed to read, check and act on it.
Meyer’s four proposed rules are to count the human work an agent’s output creates, run small experiments without turning every promising result into a permanent project, set an end point before making a task recurring, and give a coordinating agent a limited attention budget. He recommends making silence the default and having the coordinator surface no more than three decisions a day, each with a recommendation, evidence and the consequence of taking no action.
The argument draws on an Anthropic study of its own workforce, published October 16. Meyer reports that by August 2026, employee-directed agents were logging six times the hours of the employees directing them. The study also estimated that reading all agent output would take non-technical employees about half their working time. These figures describe Anthropic’s workforce and the study’s estimates; they are not measurements of every company or workplace.
Meyer says the practical test for an agent is not just how quickly it can produce something, but whether anyone will use the result and what follow-up work it creates. He advises users to identify who will read an output, who will act on it and whether the resulting work is worth doing. For outputs with consequential details, he recommends asking an agent to show its calculations or distinguish assumptions from known information.
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 Consume Attention
The article focuses on a mismatch between the low cost of generating automated work and the continuing human cost of handling it. An agent can produce a report or proposal quickly, but a person may still need to verify its facts, decide whether it matters and coordinate the next steps. If several agents send updates at once, the review work can offset the time saved on production.
That concern becomes more pressing as businesses make agent tasks recurring. A stream of low-value reports can occupy attention even when each message appears useful on its own. Meyer’s suggested limits aim to make agents bring forward decisions rather than simply activity, while deadlines and “nothing new” stopping rules can prevent unattended tasks from continuing indefinitely.
The recommendations are an individual framework, not evidence that a particular set of rules will improve productivity across organizations. Their relevance is that they put human review and follow-up into the cost of using an agent, rather than treating generated output as a completed task.
Products Worth Considering
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung 32-Inch Odyssey OLED G8 (G80SH) 4K Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Anthropic’s Workplace Findings
Anthropic’s research examined the company’s own employees and their use of agents. As summarized by Meyer, the study reported that the share of sessions spent on “meta-work,” largely managing agents, rose from 10 percent in fall 2025 to 22 percent in spring 2026. The number of written standing instructions for agents also grew from a few hundred to 8,500 over a year, according to the article.
The study’s setting limits how broadly its results can be applied. Meyer notes that Anthropic employees chose to adopt the tools and that the company’s work may not reflect practices across the economy. The findings offer a case study of one organization, rather than a general estimate of agent use or productivity in all workplaces.
Meyer’s examples of small-business owners using AI to draft proposals or inspect blueprints are also accounts published by Anthropic, which sells AI tools. One owner’s reported revenue from proposal work is his own account, not an independently verified result. The examples illustrate possible uses and the need to check outputs, but do not establish typical business outcomes.
“an army of Claudes, with every one asking for my attention.”
— An engineer quoted in Meyer’s article
How Far the Findings Apply
The supplied material does not establish whether the reported patterns at Anthropic hold across other companies, job types or agent systems. The study’s findings are specific to Anthropic’s workforce, and the article itself cautions that employees adopted the tools by choice and that the company’s work may not be representative.
It is also unclear whether Meyer’s four rules have been tested in a controlled comparison or how much time, if any, they save in his publishing and software work. The article presents them as practical guidance based on his experience, not as a measured productivity intervention. Anthropic’s customer stories are vendor-published accounts, and the business outcomes described in them should be treated as claims by the people quoted.
Putting Limits on Agent Work
Meyer’s proposed next step for agent users is to apply the four rules to each new task: decide who will handle its output, test ideas on a small scale, define when recurring work should stop and limit which decisions reach a person. For teams, the approach also implies reviewing whether existing agent updates are being read and acted on, rather than assuming that more generated material is useful.
The source material does not announce a new product, policy or follow-up study. Whether these practices reduce management overhead remains to be shown; for now, the recommendations are a framework for teams to test against their own workloads.
Key Questions
What is Meyer’s main advice for using AI agents?
He recommends counting the human work created by agent output, testing ideas without adopting every one permanently, setting stop conditions for recurring tasks and limiting the number of decisions a coordinating agent brings to a person.
What did Anthropic’s study report?
As summarized by Meyer, the study found that by August 2026 employee-directed agents logged six times as many hours as the employees directing them, and that agents spent about 80 percent of session time waiting for human input. The results concern Anthropic’s own workforce.
Does the study show that agents reduce productivity?
No. The material reports agent activity, waiting time, management work and estimated reading demands; it does not establish a general productivity effect across companies. The study’s findings should not be treated as representative of all workplaces.
Why does Meyer recommend limiting agent updates?
Because people still need to read, verify and act on many outputs. Meyer argues that updates without a decision or useful new information can add review work, even when generating them is quick.
Source: ThorstenMeyerAI.com
