🔍 Read the full analysis: How To Stop Babysitting AI Agents And Get Decisions Instead on ThorstenMeyerAI.com
Get movie nights 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
A guide published by ThorstenMeyerAI.com on October 27 argues that the main cost of AI agents is often the human time needed to review and act on their output. It draws on an Anthropic study reporting that employee-directed agents spent about 80% of session time waiting for people, and offers four practices for limiting unnecessary agent work.
ThorstenMeyerAI.com published a guide on October 27 advising people to limit the human work generated by AI agents, citing an Anthropic study that found agents in employee-directed sessions spent about 80% of their time waiting for human input. The piece argues that agent output can save production time while adding work for people who must read, check and act on it.
The guide, by Thorsten Meyer, recommends four practices: count the review and follow-up work before assigning a task; use agents to test many ideas but commit to only a few; set an end date and stopping conditions for recurring tasks; and give a coordinating agent a limited attention budget. Under the proposed approach, that agent should default to silence and bring no more than three decisions a day, each with a recommendation, supporting evidence and the consequence of taking no action.
Meyer says users should ask who will read an agent’s output, who will act on it and whether they want the work the output creates. The guide also recommends asking agents to show their work or distinguish assumptions from known information when accuracy matters. It presents these steps as practical advice, not as a measured intervention shown to reduce workload in a controlled trial.
The article cites Anthropic’s study of its own workforce, published October 16. According to the figures reported in Meyer’s piece, employee-directed agents logged six times as many hours as the employees directing them by August 2026. The article also says the study estimated that reading all agent output would take non-technical employees about half their working time. These figures describe Anthropic’s workforce and study estimates; they are not measurements of all workplaces.
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
Human Review Sets the Limit
The advice addresses a practical trade-off in workplace automation: agents can produce drafts, research and analysis quickly, but their output still consumes attention if people need to verify or use it. Fast generation does not remove review work, and a stream of low-priority updates can compete with decisions that require human judgment.
For readers deciding whether to deploy agents, the guide shifts the measure of cost from the time an agent takes to the full task lifecycle. A proposal that takes minutes to draft may still require extensive checking, while a recurring summary may be of little use if no one reads it. The article’s recommendations are intended to make those downstream costs visible before teams add more agents or automated tasks.
Popular right now
Ninja Flip Toaster Oven and Air Fryer, 8-in-1, Stainless Steel, SP151
KitchenAid Classic Series 4.5 Quart Tilt-Head Stand Mixer K45SS, Onyx Black
Roborock Robot Vacuum and Mop with LiDAR Navigation and 10,000 Pa Suction
As an affiliate, we earn on qualifying purchases.
What Anthropic Measured
Anthropic’s research, as summarized by Meyer, examined how the company’s own employees used agents. The reported findings include a rise in the share of sessions devoted to “meta-work,” described as mostly managing agents, from 10% in fall 2025 to 22% in spring 2026. Meyer also reports that written standing instructions for agents grew from a few hundred to 8,500 over a year.
The source cautions that the study reflects one company. Its employees chose to adopt the tools, and the company’s working practices may not translate to other organizations. Meyer presents the findings as a signal that human attention can constrain agent use, rather than as a universal measure of how agents perform.
The guide includes examples from Anthropic customer stories, including a chef who said Claude helped him prepare professional catering proposals and a contractor who checks how the system calculates blueprint measurements. Those are customer accounts reported by the vendor, not independently verified results. They illustrate the article’s distinction between useful output and output that still needs a knowledgeable person to check it.
“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 and relayed in Meyer’s article
How Far the Findings Apply
The cited study concerns Anthropic employees, and the source material does not establish that the same figures apply across industries, company sizes or agent products. The article does not provide enough methodological detail to assess how the study counted waiting time, estimated the time needed to read output or selected the employees included.
It is also unclear whether adopting Meyer’s four practices reduces total workload or improves decision quality. The guide offers operating advice and examples, but does not report a formal test comparing teams that use these rules with teams that do not. The reported customer outcomes are attributed claims, not independently confirmed findings.
Set Limits Before Automating
Meyer’s practical next step is for users to apply a three-question check before starting an agent: Who will read the result? Who will act on it? And do they want the work it creates? For recurring tasks, the guide recommends setting an end date, stopping when there is nothing new to report, and pausing a task if nobody uses its output.
The article does not announce a new product, policy or research milestone. Further evidence about the approach would require testing whether these limits change review time, agent use or the quality of decisions in different workplaces. Until then, the Anthropic findings should be read as company-specific observations, and the four rules as recommendations rather than proven results.
Key Questions
What is the article’s main recommendation?
Measure the human work after an agent produces an answer, including who will review it and who will act on it. Meyer also recommends limiting recurring tasks and setting a daily cap on decisions escalated by a coordinating agent.
What did the Anthropic study report about agent waiting time?
As summarized by Thorsten Meyer, agents in employee-directed sessions spent about 80% of their time waiting for human input. The finding concerns Anthropic’s workforce, not workplaces generally.
Does the study prove AI agents create more work?
No. The reported figures point to time spent waiting and managing agents, but the source does not describe a controlled test proving that agents increase total workload across organizations. Meyer’s article argues that review and follow-up work can reduce the time saved by fast generation.
What is an attention budget for an agent?
In Meyer’s proposal, a coordinating agent defaults to silence and brings a person no more than three decisions a day. Each decision should include a recommendation, evidence and what may happen if no action is taken.
Source: ThorstenMeyerAI.com
