🔍 Read the full analysis: AI Agent Management: 4 Ways To Ask For Decisions, Not Updates on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A ThorstenMeyerAI.com article published Oct. 27 sets out four practices for managing AI agents: count the human work their output creates, run small tests without adopting every idea, define stop conditions for recurring tasks, and make a coordinating agent surface only a few decisions. It cites Anthropic’s study of its own workforce, which reported extensive agent waiting time and a rising share of sessions spent managing agents; the findings are company-specific, not a general measure of all workplaces.
Thorsten Meyer’s Oct. 27 article argues that people managing AI agents should ask for decisions rather than routine progress reports, and offers four rules for limiting the human work agents create. The guide points to Anthropic’s study of its own staff, which reported that agents spent about 80 percent of session time waiting for human input—a finding Meyer uses to frame attention, rather than the speed of generating output, as the management constraint.
Meyer’s first rule is to count the work that follows an agent’s output: who will read it, who will check it and who will act on it. An agent may produce a report quickly, but reviewing a long document or correcting an error still takes people’s time. The article advises asking whether the work created by a result is work the user actually wants to take on. It gives examples from Anthropic customer stories, including a chef who said Claude helped produce catering proposals and a contractor who began asking the model to show its calculations after spotting an error in a bid estimate. Those are customer accounts published by the vendor, not independently verified results.
The second rule is to use agents for low-cost exploration without turning every promising test into a permanent project. Meyer says teams should keep ideas that demonstrate value and drop the rest, rather than accumulating pilots that generate reviews and maintenance. The third rule is to set an end point before making a task recurring. He recommends defining a date to stop, a condition for halting when there is nothing new to report, and a pause if nobody is reading the output.
The fourth rule is to give a coordinating agent an explicit attention budget. Meyer recommends making silence the default and limiting the agent to at most three decisions a day. Each request should include a recommendation, supporting evidence and the consequence of taking no action. The article presents this as a way to focus human attention on choices requiring judgment, rather than asking managers to process every status message.
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 Oversight Can Become Work
The guide addresses a practical tension in workplace automation: agents can make it cheaper and faster to produce drafts, summaries and analyses, but they do not remove the need to inspect or use those materials. If teams measure only how quickly an agent completes a task, they may miss the downstream cost of reading, checking and acting on its output. That cost can shift time away from the work the automation was meant to support.
Meyer’s recommendations focus on controlling that flow rather than simply adding more agents. Clear stopping rules can limit unattended recurring reports, while a decision-focused summary can reduce low-value interruptions. The proposed three-decision limit is the author’s management advice, not a tested universal standard. Its usefulness will depend on the task, the risk of missed information and how quickly a human needs to respond.
Products Worth Considering
Samsung 27-inch Odyssey OLED G6 QD-OLED QHD Gaming Monitor
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
What Anthropic’s Study Reports
The article cites research Anthropic published on Oct. 16, 2026, based on how its own employees used AI agents. According to the figures described by Meyer, by August 2026 employee-directed agents logged six times as many hours as the employees directing them. About 80 percent of session time was spent waiting for a person. The study also estimated that reading all agent output would take non-technical employees about half their working time.
Meyer reports that the study found the share of sessions spent on meta-work, mostly agent management, rose from 10 percent in fall 2025 to 22 percent in spring 2026. Written standing instructions grew from a few hundred to 8,500 over a year. These figures describe Anthropic’s workforce and study methods, not a representative survey of employers. The article itself cautions that employees chose to adopt the tools and that the company’s practices may not transfer to other workplaces.
“Human attention is now the bottleneck for agent output.”
— Anthropic’s study, as summarized by Meyer
Limits of the Workforce Evidence
The reported study figures are specific to Anthropic’s own workforce. The material provided does not establish how employees were selected, how session time was measured, or whether the same patterns appear in other organizations. Anthropic’s study, as summarized in the article, also notes that its employees adopted the tools by choice and that its working practices may not be transferable.
It is also unclear whether the four rules have been tested against other management approaches or whether the suggested cap of three decisions a day works across different jobs. The customer examples and reported revenue are attributed accounts rather than independently audited outcomes. Readers should treat the rules as recommendations, not proven benchmarks for productivity.
How Teams Can Test the Rules
The article does not announce a formal rollout, product change or follow-up study. Its next step for readers is practical: before adding or repeating an agent task, identify who will review the result and what action it should trigger. Teams can then set a stop condition and check whether the output is being read and used. A coordinating agent can be configured to send fewer, decision-oriented messages, with the frequency adjusted to the risk and urgency of the work.
For organizations weighing broader adoption, the open question is whether these practices reduce oversight time without hiding important developments. Tracking review effort, interruptions and useful decisions over a defined period could help answer that locally. The Anthropic findings offer a reason to measure those costs, but do not by themselves establish what will happen in other workplaces.
Key Questions
What are the four rules for managing AI agents?
Meyer recommends counting the human work that follows an agent’s output, running small tests without committing to every idea, setting an end point before making a task recurring, and limiting a coordinating agent to a small number of decision-focused requests.
What did Anthropic’s study report about agent waiting time?
As summarized in the article, agents in employee-directed sessions spent about 80 percent of their time waiting for human input. The figure comes from Anthropic’s study of its own workforce, not a broad survey of workplaces.
Does the article show that agents save employees time?
It does not establish a general productivity effect. It argues that fast output can create follow-up work, and cites Anthropic’s workforce findings and vendor-published customer accounts. Results may differ by task and organization.
What should a decision-focused agent message include?
Meyer recommends a clear decision request with the agent’s recommendation, the evidence behind it and what may happen if no action is taken. The article suggests limiting such requests to at most three a day as a management rule of thumb, not a universal limit.
Source: ThorstenMeyerAI.com
