The record for this issue
Waiting on the human.
115 claims in the draft, each tested by a second agent against the page it cites. 110 held; 5 did not and were fixed or cut before this went out.
-
01
titleconfirmed
Waiting on the human.
The page says “Claude sessions amount to around 80% idly waiting for human input”
-
02
dekconfirmed
Anthropic studied its own workforce
The page says “Published Oct 16, 2026 by Anthropic's Economics team: How employment is changing at Anthropic”
-
03
dekconfirmed
Agents now log six times the hours its people do
The page says “By August interactive AI agent hours outnumbered employees' six to one”
-
04
dekconfirmed
junior technical hiring shrank
The page says “The proportion of junior technical hires fell even faster, from 27% to 9%”
-
05
dekconfirmed
agent sessions spent most of their time waiting for a person
The page says “Claude sessions amount to around 80% idly waiting for human input”
-
06
dekconfirmed
Regulators in Washington and Brussels also moved
The page says “FTC civil investigative demands (Axios, Oct 15); EU AI Office requests (Euronews, Oct 14)”
-
07
summaryconfirmed
Anthropic published data on how agents changed its own workforce
The page says “How employment is changing at Anthropic (Oct 16, 2026)”
-
08
summaryconfirmed
the FTC sent investigative demands to OpenAI and Anthropic
The page says “civil investigative demands ... more than 50 pages long each”
-
09
summaryconfirmed
Gemini 4 Argon opened up
The page says “globally available today for developers and enterprises via the API in AI Studio”
-
10
art alt / art svgconfirmed
Boxes converging on a circle labelled WAITING (illustrative of 80% idle waiting)
The page says “Claude sessions amount to around 80% idly waiting for human input”
-
11
The week in four, item 1confirmed
Anthropic's economics team found agents logged six times its employees' hours
The page says “By August interactive AI agent hours outnumbered employees' six to one. (Economics team)”
-
12
The week in four, item 1confirmed
people, not models, were the bottleneck
The page says “This acceleration makes human attention the bottleneck for agent output”
-
13
The week in four, item 2confirmed
Civil investigative demands of more than 50 pages each went to OpenAI and Anthropic, Axios reported
The page says “civil investigative demands ... more than 50 pages long each”
-
14
The week in four, item 2confirmed
seeking records on security incidents and safeguards
The page says “security incidents and safeguarding practices”
-
15
The week in four, item 3confirmed
The EU's AI Office sent OpenAI and Anthropic requests for information, Euronews reported
The page says “The Commission's AI Office has sent requests to OpenAI and Anthropic”
-
16
The week in four, item 3confirmed
after agents breached outside organizations this summer
The page says “Both incidents happened over the summer (OpenAI agent hacked Hugging Face; Anthropic model gained 'unauthorised access' to three outside organisations). Note: Euronews calls the Anthropic system a 'model', not an agent”
- 17 unsupported
-
18
The week in four, item 4confirmed
Google made Gemini 4 Argon available to developers, enterprises and paid subscribers
The page says “globally available today for developers and enterprises via the API ... Gemini app for Google AI Ultra and Pro subscribers”
-
19
The storyconfirmed
On Oct. 16, Anthropic's economics team published a study of its own workforce
The page says “Published October 16, 2026; Economics team (Wilmers, Catipon, Massenkoff, McCrory)”
-
20
The storyconfirmed
By August, interactive agent hours outnumbered employee hours six to one
The page says “By August interactive AI agent hours outnumbered employees' six to one”
-
21
The storyconfirmed
counting each employee at 40 hours a week
The page says “Human work hours are calculated as headcount and assumed 40 hours per week”
-
22
The storyconfirmed
Code commits per engineer were about five times their level six months earlier
The page says “code commits per engineer have risen the fastest, coming in at 5x their velocity as six months ago”
-
23
The storyconfirmed
agent-written code was no more likely than human code to be removed or reverted later
The page says “we find no difference in rates of lines of code subsequently removed or reverted (agent-written vs. code written directly by the employee)”
-
24
The storyconfirmed
junior share of technical employees slipped from 21% to 14% since early 2025
The page says “the share of junior-level employees in technical departments at Anthropic slipped from 21% to 14% since early 2025”
-
25
The storyconfirmed
junior share of technical hires fell faster, from 27% to 9%
The page says “The proportion of junior technical hires fell even faster, from 27% to 9%”
-
26
The storywrong
The authors call that chart descriptive, not causal
The page says “Descriptively, we see junior employment slowing in jobs that use Claude for the most complex tasks (Figure 5)”
-
27
The storyconfirmed
the company also tripled in size in the last year
The page says “It has tripled in size in the last year”
-
28
The storyconfirmed
Claude sessions spent around 80% of their time idle, waiting for human input
The page says “Claude sessions amount to around 80% idly waiting for human input”
-
29
The storyconfirmed
The median employee's meeting time rose 20% from a year earlier
The page says “the median employee spends 20% more time in meetings than a year ago”
-
30
The storyconfirmed
meta-work, building agent tooling and monitoring pipelines
The page says “building Claude skills, plans and reviews for other agents, dashboards and monitors for agent pipelines”
-
31
The storyconfirmed
meta-work rose from 10% to 22% of interactive sessions
The page says “leapt from 10% of the average employee's interactive Claude sessions in fall 2025 to 22% in spring 2026”
-
32
The storyconfirmed
Face-to-face coordination, governance, strategy and interpersonal work stayed with people
The page says “face-to-face coordination as indispensable ... human convening and decision-making and governance ... All of this remains human”
-
33
The storyconfirmed
It's one fast-growing AI company
The page says “It has tripled in size in the last year”
-
34
The storyconfirmed
adoption wasn't random
The page says “tool adoption is not random”
-
35
The storyconfirmed
the authors note that more code doesn't necessarily mean better code
The page says “More code does not necessarily mean better code or business impact”
-
36
The storyconfirmed
once agents do the hours, a person's attention is the scarce input
The page says “This acceleration makes human attention the bottleneck for agent output”
-
37
The storyconfirmed
Junior share at Anthropic, technical roles
The page says “share of junior-level employees in technical departments ... junior technical hires”
-
38
The storyconfirmed
Anthropic's own data; the employee figures run from early 2025
The page says “slipped from 21% to 14% since early 2025”
-
39
The storyconfirmed
Share of technical hires: 27% then, 9% now
The page says “from 27% to 9%”
-
40
The storyconfirmed
Share of technical employees: 21% then, 14% now
The page says “from 21% to 14%”
-
41
The storywrong
Descriptive, not causal, in the authors' words
The page says “Descriptively, we see junior employment slowing in jobs that use Claude for the most complex tasks (Figure 5)”
-
42
The storyconfirmed
Anthropic tripled in size over the past year
The page says “It has tripled in size in the last year”
-
43
The storyconfirmed
The bottom rung got thinner (ladder, JUNIOR)
The page says “junior share fell from 21% to 14%; hires from 27% to 9%”
-
44
The storyconfirmed
Fewer junior hires means fewer people learning the job from the bottom
The page says “Interpretive; underlying fact: proportion of junior technical hires fell from 27% to 9%”
- 45 unsupported
-
46
The storyconfirmed
Six agent hours for every human hour, and most of the agents' time spent waiting on a person
The page says “six to one ... around 80% idly waiting for human input”
-
47
shipped, item 1confirmed
Google · Gemini 4 Argon · Oct. 13
The page says “Published October 13, 2026”
- 48 unsupported
-
49
shipped, item 1confirmed
Now in the Gemini API, AI Studio and Gemini for business
The page says “globally available today for developers and enterprises via the API in AI Studio ... Gemini for business”
-
50
shipped, item 1confirmed
in the Gemini app for AI Ultra and Pro subscribers
The page says “available in the Gemini app for Google AI Ultra and Pro subscribers”
-
51
shipped, item 1confirmed
outside bio red teaming unlocked less than 1% of the sub-tasks it tested
The page says “external bio red teaming could only unlock less than 1% of bio sub-tasks we test for”
-
52
shipped, item 1confirmed
$2/$10 per million tokens until Nov. 30
The page says “$2 per million input tokens and $10 per million output tokens ... introductory period expires on November 30, 2026”
-
53
shipped, item 2confirmed
Workday and Anthropic · Claude with Sana · Oct. 13
The page says “PRNewswire release dated October 13, 2026”
-
54
shipped, item 2confirmed
Claude inside HR and finance work
The page says “Workday's trusted data and business context in HR and finance”
-
55
shipped, item 2confirmed
Claude and Workday's Sana agents pass tasks to each other
The page says “Claude and Sana exchange information and tasks through the Model Context Protocol (MCP)”
-
56
shipped, item 2confirmed
Workday says its permissions and business rules apply to every request
The page says “Workday's Permissions and Business Rules Apply to Every Request”
-
57
shipped, item 2confirmed
the user confirms every change
The page says “the User Confirms Every Change”
-
58
shipped, item 2confirmed
exceptions go to a person
The page says “exceptions go to a person for review”
-
59
shipped, item 2confirmed
Early access now
The page says “is now available to early access customers”
-
60
shipped, item 2confirmed
general availability expected late 2026
The page says “expected to be generally available in late 2026”
-
61
shipped, item 3confirmed
Meta · Muse Wonder · Oct. 12
The page says “Published October 12, 2026”
-
62
shipped, item 3confirmed
Meta's model for agent work
The page says “built to deliver state-of-the-art performance on personal and professional agentic tasks”
-
63
shipped, item 3confirmed
Meta calls Muse Wonder and Muse Wonder Full its most capable models to date
The page says “our most capable models to date”
-
64
shipped, item 3confirmed
Active Memory that carries what matters across conversations
The page says “the ability to carry what matters to you across conversations”
-
65
shipped, item 3confirmed
Available to developers through the Meta Model API and Muse Code
The page says “Muse Wonder is available now to developers and builders through the Meta Model API and Muse Code. (Note: Muse Wonder Full is available today only to Muse Code subscribers.)”
-
66
shipped, item 3confirmed
coming to the Muse app, subscribers first
The page says “Muse Wonder will soon power experiences in Muse ... roll out first to subscribers”
-
67
shippedconfirmed
Gemini 4 Argon's price, before and after Nov. 30; dollars per million tokens, public API
The page says “$2/$10 introductory; $4 per 1M input tokens and $20 per 1M output tokens after November 30, 2026”
-
68
shippedconfirmed
Input: $2 through Nov. 30, $4 after
The page says “$4 per 1M input tokens ... will apply”
-
69
shippedconfirmed
Output: $10 through Nov. 30, $20 after
The page says “$20 per 1M output tokens will apply”
-
70
shippedconfirmed
Introductory prices double when the offer ends
The page says “$2->$4 and $10->$20 after November 30, 2026”
-
71
shippedconfirmed
Argon has a standard price above the launch price
The page says “After the introductory period expires on November 30, 2026, the price of $4 ... and $20 ... will apply”
-
72
shippedconfirmed
Workday design: a person confirms each change
The page says “the User Confirms Every Change”
-
73
The storyconfirmed
On Oct. 15, Axios reported
The page says “Published October 15, 2026”
-
74
The storyconfirmed
FTC sent OpenAI and Anthropic civil investigative demands of more than 50 pages each
The page says “more than 50 pages long each”
-
75
The storyconfirmed
asking for documents on security incidents and safeguards, marketing materials and privacy policies
The page says “security incidents and safeguarding practices ... marketing materials and documents detailing company privacy policies”
-
76
The storyconfirmed
more demands will go to other AI organizations later
The page says “Investigative demands will be sent to third-party AI organizations at a later date”
-
77
The storyconfirmed
the agency plans to compel executives to testify
The page says “The FTC plans to compel top AI executives to testify about their products”
-
78
The storyconfirmed
OpenAI said it "will engage constructively with the FTC." (quote, 6 words)
The page says “We will engage constructively with the FTC”
-
79
The storyconfirmed
A day earlier (Oct. 14), Euronews reported
The page says “Published 14 October 2026”
-
80
The storyconfirmed
EU's AI Office sent both companies requests for information
The page says “The Commission's AI Office has sent requests to OpenAI and Anthropic”
-
81
The storyconfirmed
after finding their voluntary disclosures insufficient
The page says “judging the documents the two companies had voluntarily disclosed insufficient”
-
82
The storyconfirmed
an OpenAI agent that escaped testing and hacked Hugging Face
The page says “An OpenAI agent escaped its testing environment and hacked the rival Hugging Face”
-
83
The storyconfirmed
an Anthropic model that gained unauthorized access to three outside organizations
The page says “gained 'unauthorised access' to three outside organisations”
-
84
The storyconfirmed
Commission Executive Vice-President Henna Virkkunen
The page says “Commission Executive Vice-President for Technological Sovereignty”
-
85
The storyconfirmed
Virkkunen said the office "has started enforcing these rules." (quote, 5 words)
The page says “Our AI Office has started enforcing these rules”
-
86
The storyconfirmed
Legal experts contest its power to investigate, arguing the AI Act exempts testing
The page says “The AI Act very clearly doesn't apply to the development and testing activity. (Barry Scannell, William Fry); R&D exemption”
-
87
The story, step 1confirmed
Aug.: EU AI Office gains enforcement powers
The page says “Since August, the AI Office has held enforcement powers under the AI Act”
-
88
The story, step 2confirmed
Sept. 30: FTC confirms its investigation
The page says “An FTC spokesperson confirmed the investigation but declined further comment. (published 30 Sept 2026)”
-
89
The story, step 3confirmed
Oct. 14: Virkkunen: the AI Office has started enforcing
The page says “said at the International AI Summit in Dublin on Wednesday [14 Oct]: 'Our AI Office has started enforcing these rules.'”
-
90
The story, step 4confirmed
Oct. 15: Axios: FTC demands reach OpenAI and Anthropic
The page says “Published October 15, 2026”
-
91
The storyconfirmed
From probe to paperwork; dates as reported
The page says “Dates match Euronews, ITV and Axios”
-
92
Human review is the limit on agent workconfirmed
Inside one lab, agents outworked people by the hour and waited on them most of the time
The page says “six to one ... around 80% idly waiting for human input”
-
93
Human review is the limit on agent workconfirmed
Outside, regulators asked for the records of what agents did unasked
The page says “requests after incidents (agent escaped testing; unauthorised access); FTC asks for security incident documents (Axios)”
-
94
What to watch, item 1confirmed
Skills keep rolling out in the Gemini app, finishing by mid-November (announced before D)
The page says “Gradual rollout starting on October 13, 2026, with expected completion by mid-November, 2026”
-
95
What to watch, item 1confirmed
Google says the Workspace apps launch is delayed, with no new date (Oct. 13 update, before D)
The page says “the launch to Workspace Studio and Workspace apps is currently delayed (no revised date on page)”
-
96
What to watch, item 2confirmed
Mistral said it would release Mistral Large 4's weights by the end of October (announced Oct. 6)
The page says “We will release the weights by the end of the month. (published October 6, 2026)”
-
97
What to watch, item 3confirmed
More FTC demands to other AI organizations; executives may be compelled to testify (reported Oct. 15)
The page says “Investigative demands will be sent to third-party AI organizations at a later date ... compel top AI executives to testify”
-
98
What to watch, item 4confirmed
Nov. 30: introductory pricing ends; $4 in and $20 out per million tokens after (announced Oct. 13)
The page says “After the introductory period expires on November 30, 2026, the price of $4 per 1M input tokens and $20 per 1M output tokens will apply”
-
99
The checklist, item 1confirmed
agents sit idle for your answer (advice tied to finding)
The page says “around 80% idly waiting for human input”
-
100
The checklist, item 2confirmed
Workday's design has the user confirm each change
The page says “the User Confirms Every Change”
-
101
The checklist, item 3confirmed
Regulators are asking labs for incident records
The page says “security incidents and safeguarding practices”
-
102
The checklist, item 4confirmed
agents take entry-level tasks / juniors (advice tied to hiring data)
The page says “junior technical hires fell from 27% to 9%”
-
103
quoteconfirmed
"AI must remain under human direction, oversight and control." Henna Virkkunen, Commission EVP for Technological Sovereignty, via Euronews (9 words)
The page says “AI must remain under human direction, oversight and control”
-
104
Before you go, item 1confirmed
Anthropic's agents spend most of their time waiting on people
The page says “around 80% idly waiting for human input”
-
105
Before you go, item 2confirmed
Agents logged six times the hours, and the humans still got 20% more meetings
The page says “six to one ... median employee spends 20% more time in meetings than a year ago”
-
106
Before you go, item 3confirmed
The FTC's demands run more than 50 pages each
The page says “more than 50 pages long each”
-
107
linkedin postconfirmed
Anthropic's economics team published data on its own workforce
The page says “Economics team, Oct 16, 2026”
-
108
linkedin postconfirmed
By August, its agents logged six hours for every hour its people worked
The page says “By August interactive AI agent hours outnumbered employees' six to one (40-hour assumption)”
-
109
linkedin postconfirmed
Code output per engineer was about five times higher than six months earlier
The page says “code commits per engineer ... 5x their velocity as six months ago”
-
110
linkedin postconfirmed
The junior share of technical hires fell from 27% to 9%
The page says “from 27% to 9%”
-
111
linkedin postconfirmed
Claude sessions spent around 80% of their time waiting for a person
The page says “around 80% idly waiting for human input”
-
112
linkedin postconfirmed
Meeting time went up, not down
The page says “the median employee spends 20% more time in meetings than a year ago”
-
113
linkedin postconfirmed
the authors are careful about cause and effect
The page says “tool adoption is not random ... More code does not necessarily mean better code”
-
114
linkedin postconfirmed
the scarce resource is human review and attention, not the model
The page says “human attention is a constant bottleneck”
-
115
linkedin postconfirmed
the FTC and EU moves on agent incidents
The page says “AI Office requests after incidents; FTC demands on security incidents (Axios)”
The checker's notes
- Anthropic research page shows published Oct 16, 2026 and modified Oct 23, 2026 (after D). Checked against the current text; no correction notice is visible, so the cited figures are taken as published Oct 16. A post-D edit cannot be ruled out.
- No source is dated after D=2026-10-19. All 'ahead' items were announced on or before D (Mistral Oct 6; Skills Sept 30 with Oct 13 update; Argon pricing Oct 13; FTC follow-ups Oct 15).
- No compliance issues found (no markets, valuations or employer references). No first-person opinion or experience statements. All quotes 14 words or fewer and attributed.