🔍 Read the full analysis: Want AI Agents To Make Progress? Ask For Decisions on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Anthropic’s October 16 study of its own workforce found employee-directed agents logged six times the hours of the employees directing them by August 2026, while spending about 80% of session time waiting for human input. The figures come from one company and do not establish what happens across the wider economy. A commentary published October 27 argues that agent systems should limit notifications and present people with a small number of evidence-backed decisions.
A commentary published October 27 argues that teams using AI agent fleets should ask systems to bring people a few specific decisions, rather than send a steady stream of progress updates. The proposal follows an October 16 Anthropic study of its own workforce, which found that employee-directed agents logged six times as many hours as their human directors by August 2026 and spent about 80% of session time waiting for human input.
The study is evidence about Anthropic’s employees, not a measurement of agent use across all workplaces. Its authors cautioned that workers adopted the tools by choice and that many features of the company’s work may not generalize elsewhere. The figures nonetheless point to a practical constraint: people must answer questions, review results and decide what to do with them, even when agents can produce work quickly.
The study estimated that reading all agent output would take non-technical employees about half their working time. It also found that the share of sessions spent on “meta-work,” mostly managing agents, 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, according to the source material.
Thorsten Meyer’s commentary recommends four practices: count the human work required after an agent finishes; use small experiments but keep only those that prove useful; set an end date and stop conditions for recurring jobs; and give a coordinating agent an attention budget. Under the last rule, the 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.
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 proposal addresses a mismatch between the low cost of generating an agent response and the human time needed to evaluate it. A report that takes an agent minutes may still require a person to read, verify and act on it. If several systems produce frequent outputs, their combined review burden can become work of its own.
That matters to organizations adopting agents because output volume is not the same as useful progress. A system may complete tasks quickly while leaving people with more messages, approvals and follow-up work. Asking for decisions instead of routine status could focus attention on cases where a person’s judgment changes what happens next. It does not remove the need for oversight; it aims to make that oversight more selective.
The recommendations are practical guidance from the commentary, not a controlled test showing that a three-decision limit or silence-by-default policy improves productivity. Teams would need to judge whether those rules suit their risks, deadlines and responsibilities.
Products Worth Considering
Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
DJI Osmo Mobile 8 Standard Combo Phone Gimbal Stabilizer
As an affiliate, we earn on qualifying purchases.
What Anthropic Measured
Anthropic published its workforce study on October 16, 2026. The figures described in the commentary concern employee-directed agent sessions and the company’s own staff. By August 2026, agents in those sessions logged six times the hours of their human directors; around 80% of session time was spent waiting for human input.
The study also tracked the work involved in managing agents. The share of sessions categorized as meta-work increased between fall 2025 and spring 2026, while written standing instructions rose to 8,500 over a year. The commentary links these figures to a broader operational concern: making agent output is only one part of a task, and people still have to assess what the output means.
Meyer also points to examples from Anthropic customer stories. A private chef said Claude helped draft proposals associated with $20,000 in revenue over a month and a half; that is the owner’s account, published by the vendor, rather than independently verified financial data. A painting contractor said he checks the system’s calculations against blueprint measurements and asks it to “show me your work.” These anecdotes illustrate possible uses and checking practices, but they do not establish typical results.
“an army of Claudes, with every one asking for my attention.”
— An engineer quoted in the commentary
How Far the Findings Apply
The study covers Anthropic employees and their agent sessions; the supplied material does not provide a representative sample of workers at other organizations. Anthropic’s authors cautioned that staff adopted the tools by choice and that the company’s working practices may not transfer to the wider economy. The measurements should not be treated as a general rate for all agent use.
The commentary’s recommendations are not reported as outcomes from a controlled comparison. It is not clear from the source whether limiting an agent to three decisions daily, setting specific stop conditions or defaulting to silence has been tested against other approaches. Nor does the material specify how the suggested rules should be adapted for urgent tasks or high-risk decisions.
The customer examples are also limited: one business owner’s reported revenue and one contractor’s checking method do not show how commonly agents deliver similar results. The source does not establish the full costs of review, verification or correcting errors across those businesses.
Testing Attention Budgets at Work
The commentary does not announce a new product, policy or formal trial. Its next step for readers is operational: before assigning an agent a recurring task, decide who will review its output, what counts as a useful result and when the task should stop. For coordinating agents, it recommends a default of silence and a small daily limit on requests for human decisions.
Organizations applying the advice could track whether fewer updates reduce review time without causing missed issues, and whether the agent’s recommendations include enough evidence for people to act safely. The supplied material gives no timetable for a broader Anthropic study or an independent assessment of these practices. For now, the workforce figures are a company-specific signal, while the attention-budget rules remain recommendations to test.
Key Questions
What did Anthropic’s study find?
In Anthropic’s study of its own workforce, employee-directed agents logged six times the hours of the people directing them by August 2026. The agents spent about 80% of session time waiting for human input. The findings describe Anthropic’s staff, not the workforce as a whole.
What does “ask for decisions” mean?
It means asking an agent to avoid routine progress messages and bring forward cases where a person must choose what happens next. Meyer recommends that a coordinating agent present a recommendation, its evidence and the consequence of taking no action, with no more than three such decisions a day.
Did Anthropic test the four rules?
The source presents the four practices as Meyer’s recommendations. It does not report a controlled test showing that they improve productivity or reduce the time people spend managing agents.
Can the study’s figures be applied to other workplaces?
Not directly. The study examined Anthropic employees, and its authors cautioned that the staff adopted the tools by choice and that the company’s work may not generalize. The results offer a company-specific measurement, not a universal estimate.
What should a team decide before starting a recurring agent task?
The commentary recommends naming who will read the result, who will act on it, and whether the resulting work is worth doing. For recurring tasks, it also advises setting an end date, defining a “nothing new” stop and pausing the task if nobody reads its output.
Source: ThorstenMeyerAI.com
