The record for this issue
Capable, but rarely cheaper.
119 claims in the draft, each tested by a second agent against the page it cites. 119 held. Nothing had to be cut.
-
01
titleconfirmed
Able isn't the same as affordable (capability vs cost framing)
The page says “robots can already perform 74% of physical tasks ... Robots are cost-competitive for just 0.3% of job tasks”
-
02
dekconfirmed
Published Sept. 30
The page says “Page dated September 30, 2026”
-
03
dekconfirmed
Two Anthropic researchers
The page says “Contributors: Russell Legate-Yang and Maxim Massenkoff”
-
04
dekconfirmed
Robots can already do 74% of physical tasks in U.S. jobs
The page says “We find that robots can already perform 74% of physical tasks in the US”
-
05
dekconfirmed
Cheaper than people for 0.3% of tasks
The page says “Robots are cost-competitive for just 0.3% of job tasks”
-
06
summaryconfirmed
Anthropic measured which physical work robots can do; ability vs worth paying for differ
The page says “can a robot today perform this task, and if so under what circumstances? ... cost-competitive for just 0.3% today”
-
07
art svg textconfirmed
0.3%
The page says “cost-competitive for just 0.3% of job tasks”
-
08
art altconfirmed
Robot arm with price tag labelled 0.3%
The page says “Matches SVG content (text element '0.3%'); figure value per report”
-
09
The essay, section 1confirmed
Published Sept. 30 by two Anthropic researchers
The page says “Sept 30, 2026; contributors Russell Legate-Yang and Maxim Massenkoff”
-
10
The essay, section 1confirmed
Robot defined as autonomous physical machine that senses and acts
The page says “We take robots to mean autonomous physical machines that sense and act”
-
11
The essay, section 1confirmed
74% of physical tasks in American jobs
The page says “robots can already perform 74% of physical tasks in the US”
-
12
The essay, section 1confirmed
Those tasks make up 34% of all working hours
The page says “making up 34% of working hours”
-
13
The essay, section 1confirmed
Cheaper than paying people for 0.3% of job tasks
The page says “Robots are cost-competitive for just 0.3% of job tasks”
-
14
The essay, section 1confirmed
At past price-decline pace, 40 years to reach 10%
The page says “If robot price declines follow past trends, it will take 40 years for that share to reach 10%”
-
15
The essay, section 1confirmed
Quote: headline says robots "can replace 74 percent of manual labor" (7 words, attributed to a headline/DigitalToday)
The page says “Headline: "Robots can replace 74 percent of manual labor, but are not yet cheaper than humans" (published 2026-10-01 KST)”
-
16
The essay, section 1confirmed
Cost caveat comes after the comma in that headline
The page says “"...manual labor, but are not yet cheaper than humans"”
-
17
The essay, section 1confirmed
Quote: report says "mostly in limited settings"
The page says “making up 34% of working hours, but mostly in limited settings”
-
18
The essay, section 1confirmed
IFR said on Sept. 24
The page says “Frankfurt, Sep 24, 2026”
-
19
The essay, section 1confirmed
Record 5 million industrial robots operating in factories worldwide in 2025
The page says “The global operational stock of industrial robots surged 9% to a record 5 million units in 2025”
-
20
The essay, section 1confirmed
Factories are built around the machine; controlled settings are where robots do best
The page says “most deployed robots work in engineered environments ... Most robots today operate in controlled environments like factories”
-
21
The essay, section 2confirmed
O*NET sponsored by the U.S. Department of Labor
The page says “developed under the sponsorship of the U.S. Department of Labor/Employment and Training Administration”
-
22
The essay, section 2confirmed
Around 900 occupations
The page says “a database of around 900 occupations linked with descriptions of around 19,000 job tasks”
-
23
The essay, section 2confirmed
Around 19,000 job tasks
The page says “descriptions of around 19,000 job tasks”
-
24
The essay, section 2confirmed
Quote: hand packer task "Seal containers or materials, using glues, fasteners, nails, and hand tools." (10 words, attributed to O*NET)
The page says “Task: "Seal containers or materials, using glues, fasteners, nails, and hand tools."”
-
25
The essay, section 2confirmed
Claude scored each task for physical, cognitive and interpersonal demands
The page says “we have Claude score task descriptions on a rubric measuring physical, cognitive, and interpersonal work requirements”
-
26
The essay, section 2confirmed
"Dig trenches" physical; "Teach dance students" physical plus cognitive and interpersonal
The page says “The task "Dig trenches" is physical; so is "Teach dance students," though this also requires cognitive and interpersonal skills”
-
27
The essay, section 2confirmed
Claude told to search for specific robots, check where they operate, quote sources
The page says “we instruct Claude to search for specific robots relevant to each task ... assess their capabilities and operating environments, quoting sources directly”
-
28
The essay, section 2confirmed
E0: no robot can do it
The page says “Robot cannot perform task”
-
29
The essay, section 2confirmed
E1: purpose-built robotic workplace, like a factory assembly line
The page says “Robot can perform task in a purpose-built robotic work environment, like a factory assembly line”
-
30
The essay, section 2confirmed
E2: structured human workplace, like a logistics warehouse
The page says “Robot can perform task in a structured human work facility, like a logistics warehouse”
-
31
The essay, section 2confirmed
E3: unstructured setting, like a city road
The page says “Robot can perform task in an unstructured environment, like a city road”
-
32
The essay, section 2confirmed
Hospital meal trays to match diets is E1 because food-prep robots pack meals on conveyor belts
The page says “Assemble meal trays with foods in accordance with patients' diets ... Food prep robots pack ready-to-eat meals on conveyor belts”
-
33
The essay, section 2confirmed
Delivering medicines to nursing stations is E2
The page says “Deliver medications or pharmaceutical supplies to patients, nursing stations, or surgery (E2 example)”
-
34
The essay, section 2confirmed
Driving a taxi is E3
The page says “"Drive taxicabs or privately owned vehicles to transport passengers" ... E3 today (in an unstructured environment)”
-
35
The essay, section 2confirmed
Dyeing a client's hair is E0
The page says “"Bleach, dye, or tint hair, using applicator or brush" (E0 example)”
-
36
The essay, section 2confirmed
Robots find unpredictable surroundings hard; most work in environments engineered for them
The page says “it's hard for robots to adapt to unpredictable environments ... most deployed robots work in engineered environments”
-
37
The essay, section 2confirmed
E1 possible in purpose-built environments but not wider settings
The page says “Robots can do half of physical tasks in purpose-built environments but not in wider settings (E1)”
-
38
The essay, section 2confirmed
Half of physical tasks only in purpose-built settings
The page says “Robots can do half of physical tasks in purpose-built environments but not in wider settings (E1)”
-
39
The essay, section 2confirmed
22% in structured human workplaces
The page says “E2: 22% of physical tasks”
-
40
The essay, section 2confirmed
2% in the open world
The page says “E3: 2% of physical tasks”
-
41
The essay, section 2confirmed
26% no robot can do (100 minus 74)
The page says “Tasks that robots can't do (E0) make up about a quarter of physical tasks. (26% is the arithmetic complement of the stated 74%; page text says 'about a quarter')”
-
42
The essay, section 2confirmed
A task robots can already do is what the report calls exposed
The page says “A job is exposed when robots can perform its tasks. (task-level 'robot exposure' measure)”
-
43
The essay, section 2confirmed
Exposure measures what is technically possible today
The page says “Our task-level measure of robot exposure asks: can a robot today perform this task ... robots can theoretically do tasks that add up to 34% of all work time”
-
44
The essay, section 2confirmed
Quote: "difficult to predict exactly" (4 words, attributed to authors)
The page says “It's difficult to predict exactly how these efforts will affect jobs and productivity”
-
45
The essay, section 2confirmed
Around half of all work exposed to LLMs
The page says “Overall, around half of work is exposed to LLMs alone”
-
46
The essay, section 2confirmed
Rises to 81% with robots
The page says “but that rises to 81% when considering robots (key findings round to 'about 80%')”
-
47
The essay, section 2confirmed
Where robots can do the work
The page says “Figure title is descriptive of E0-E3 breakdown”
-
48
The essay, section 2confirmed
Share of U.S. physical tasks, by least controlled setting a robot can manage
The page says “E1-E3 tiers defined by environment; shares given of physical tasks”
-
49
The essay, section 2confirmed
No robot can yet (E0): 26%
The page says “about a quarter of physical tasks (26 = 100 - 74)”
-
50
The essay, section 2confirmed
Purpose-built only (E1): 50%
The page says “Robots can do half of physical tasks in purpose-built environments”
-
51
The essay, section 2confirmed
Structured workplace (E2): 22%
The page says “22% of physical tasks (E2)”
-
52
The essay, section 2confirmed
Open world (E3): 2%
The page says “2% of physical tasks (E3)”
-
53
The essay, section 2confirmed
Three-quarters within reach, most only in a space built for the machine
The page says “74% of physical tasks ... half ... in purpose-built environments”
-
54
The essay, section 3confirmed
Packers and packagers are the largest occupation where robots already beat people on cost
The page says “Packers and packagers are the largest occupation exposed to cost-competitive robots”
-
55
The essay, section 3confirmed
Hand packer tasks run from inspecting and weighing to lining cartons, sealing, labelling, moving goods onto conveyors or loading docks
The page says “Examine and inspect ...; Measure, weigh, and count ...; Assemble, line, and pad cartons ...; Seal containers ...; Mark and label containers ...; Remove completed ... placing them on moving equipment, such as…”
-
56
The essay, section 3confirmed
Robots: pack containers, move materials around warehouses, inspect visually, erect boxes, print and apply labels, seal packages
The page says “pack containers, move materials around warehouses, visually inspect goods, erect cardboard boxes, print and apply labels, and seal packages”
-
57
The essay, section 3confirmed
Over $2 million to buy and install
The page says “These robots are estimated to cost over $2 million to purchase and install”
-
58
The essay, section 3confirmed
Replace yearly work of around 14 workers
The page says “but replace the yearly work of around 14 workers”
-
59
The essay, section 3confirmed
Up-front cost spread over working life allowing for cost of money
The page says “annualizes fixed costs to reflect hardware lifespans and the time value of money ... roughly 10-year service life at an 8% cost of capital”
-
60
The essay, section 3confirmed
Cost-competitive when robot cheaper than slice of total pay and benefits spent on task
The page says “cost-competitive when its robot cost is less than its labor cost (total compensation scaled by the task's share of time)”
-
61
The essay, section 3confirmed
Packers spend 97% of time on tasks robots can do
The page says “packers and packagers spend 97% of their time on tasks that robots can do”
-
62
The essay, section 3confirmed
Each costs around $49,000
The page says “and cost around $49,000”
-
63
The essay, section 3confirmed
Robots about $2,500 a year cheaper
The page says “robots cost about $2,500 less per year to do that work”
-
64
The essay, section 3confirmed
This is the report's showcase on cost; margin thin vs $2M+ purchase
The page says “Largest cost-competitive occupation; ~$2,500/yr saving vs over $2 million to purchase and install (interpretive, consistent with page)”
-
65
The essay, section 3confirmed
Rests on estimates of robot cost, lifespan, coverage
The page says “these cost estimates are approximate ... roughly 10-year service life”
-
66
The essay, section 3confirmed
Across almost all other tasks the person is cheaper or robot can't do the work
The page says “Robots are cost-competitive for just 0.3% of job tasks ... For nearly all tasks, costs would also need to fall”
-
67
The essay, section 3confirmed
Claude read task descriptions and searched out robots, quoting sources
The page says “we have Claude score task descriptions ... search for specific robots ... quoting sources directly”
-
68
The essay, section 3confirmed
Quote: rubric "requires many judgment calls" (4 words, attributed to authors)
The page says “This rubric requires many judgment calls”
-
69
The essay, section 3confirmed
Frame was the authors': place over robot, pricing method, historical test
The page says “A backtest across 50 years validates this approach; authors define E0-E3 environment rubric and annualized cost method”
-
70
The essay, section 3confirmed
More than $2 million against 14 workers
The page says “cost over $2 million to purchase and install, but replace the yearly work of around 14 workers”
-
71
The essay, section 3confirmed
14 WORKERS (and 14 drawn figures: two rows of 7)
The page says “around 14 workers”
-
72
The essay, section 3confirmed
Seesaw almost level, robot arm vs 14 figures
The page says “SVG contains 14 person figures; 'around 14 workers'”
-
73
The essay, section 3confirmed
Packers largest cost-competitive occupation; robots win by about $2,500 a year
The page says “largest occupation exposed to cost-competitive robots ... about $2,500 less per year”
-
74
The essay, section 3confirmed
Authors call their cost estimates approximate
The page says “these cost estimates are approximate”
-
75
The essay, section 4confirmed
Half only in purpose-built settings, 2% in open world
The page says “half of physical tasks in purpose-built environments ... E3 2%”
-
76
The essay, section 4confirmed
A task is not a job (exposure is task-level)
The page says “Our task-level measure of robot exposure”
-
77
The essay, section 4confirmed
Backtest from 1977: more exposed jobs saw wage and employment declines in later decades
The page says “jobs that were more exposed to existing robots experienced wage and employment declines in later decades”
-
78
The essay, section 4confirmed
In 1977 robots couldn't do 62% of physical tasks
The page says “In 1977, robots could not do 62% of physical tasks”
-
79
The essay, section 4confirmed
Today they can do all but 24% of those tasks
The page says “Today's robots can do all but 24% of those same tasks”
-
80
The essay, section 4confirmed
Robots pick up about 2% of tasks they couldn't do each year
The page says “each year, robots have become able to do about 2% of the physical work they previously couldn't”
-
81
The essay, section 4confirmed
Cost is one of four barriers
The page says “Barriers named: capabilities, costs, regulation, human preferences”
-
82
The essay, section 4confirmed
Barriers overlap
The page says “Around 30% of physical tasks are limited either by preferences or by regulation (14% + a quarter > 30%, implying overlap); For nearly all tasks, costs would also need to fall”
-
83
The essay, section 4confirmed
Capabilities and cost are the biggest
The page says “We find that capabilities and costs are the biggest impediments to robot adoption”
-
84
The essay, section 4confirmed
Regulations wouldn't allow 14% of physical tasks
The page says “today's regulations wouldn't allow robots to perform 14% of physical tasks”
-
85
The essay, section 4confirmed
Preferences hold robots back from a quarter
The page says “Human preferences hold back robots for a quarter of tasks”
-
86
The essay, section 4confirmed
Claude applied authors' rubric that needed many judgment calls
The page says “This rubric requires many judgment calls”
-
87
The essay, section 4confirmed
Robot costs estimated with web searches starting from robots cited for each task
The page says “We prompt Claude to estimate costs using web search and the list of robots cited for each task”
-
88
The essay, section 4confirmed
Careful first map, not a precise measurement
The page says “these cost estimates are approximate ... more suggestive”
-
89
The essay, section 4confirmed
Highly exposed workers 55 percentage points less likely to hold bachelor's or higher
The page says “These workers are also 55 percentage points less likely to hold a bachelor's degree or higher (top quintile vs unexposed)”
-
90
The essay, section 4confirmed
Earn around $30 less per hour
The page says “earn around $30 less per hour”
-
91
The essay, section 4confirmed
Unemployment rate more than twice as high
The page says “face an unemployment rate more than twice as high”
-
92
The essay, section 5confirmed
Industry and government data: robot prices fell roughly 3% a year since the 1990s
The page says “industry and government data suggest that robot prices have declined roughly 3% per year since the 1990s”
-
93
The essay, section 5confirmed
Costs would need to fall about 70% for robots to be cheaper for 10% of today's work
The page says “For robots to be cost-competitive for 10% of human work today, costs would need to decline about 70%”
-
94
The essay, section 5confirmed
At 3% a year, around 40 years
The page says “At a 3% decline per year, that would take around 40 years”
-
95
The essay, section 5confirmed
A 20% fall takes around seven years
The page says “it would take around seven years to reach a 20% cost decline”
-
96
The essay, section 5confirmed
Warehouses are the report's example of a structured workplace
The page says “like a logistics warehouse (E2)”
-
97
The essay, section 5confirmed
IFR said on Sept. 30
The page says “Frankfurt, Sep 30, 2026”
-
98
The essay, section 5confirmed
Transportation and logistics stayed top use for professional service robots
The page says “Transportation and logistics stays in the lead as top application”
-
99
The essay, section 5confirmed
117,500 sold in 2025
The page says “with 117,500 units sold in 2025 (note: page subheading rounds to 117,000)”
-
100
The essay, section 5confirmed
Full-size humanoid robots sold about 7,000
The page says “In 2025, about 7,000 units of full-size humanoids (above 140 cm) were sold globally”
-
101
The essay, section 5confirmed
Of 10 most exposed occupations, nine are vehicle operators
The page says “Of these 10 occupations, 9 are vehicle operators”
-
102
The essay, section 5confirmed
Taxi drivers top the list with index 2.2
The page says “Taxi drivers lead exposure with an index of 2.2”
-
103
The essay, section 5confirmed
Self-driving cars haven't replaced drivers at scale
The page says “Autonomous vehicles have yet to upend driving jobs at scale”
-
104
The essay, section 5confirmed
But the capability is there
The page says “Drive taxicabs ... E3 today”
-
105
The essay, section 5confirmed
On O*NET OnLine you can look up an occupation and read its task list
The page says “Occupation summary page 53-7064.00 lists 12 tasks; home page: find, search, or browse across 900+ occupations”
-
106
The essay, section 5confirmed
How fast the gap closes, if prices keep falling
The page says “If robot price declines follow past trends, it will take 40 years”
-
107
The essay, section 5confirmed
Anthropic's estimates, assuming robot prices fall 3% a year
The page says “At a 3% decline per year”
-
108
The essay, section 5confirmed
1977: robots can't do 62% of physical tasks
The page says “In 1977, robots could not do 62% of physical tasks”
-
109
The essay, section 5confirmed
Today: cheaper than people for 0.3% of job tasks
The page says “Robots are cost-competitive for just 0.3% of job tasks”
-
110
The essay, section 5confirmed
About 7 years: costs down 20%
The page says “around seven years to reach a 20% cost decline”
-
111
The essay, section 5confirmed
About 40 years: costs down 70%, cheaper for 10% of today's work
The page says “cost-competitive for 10% of human work today, costs would need to decline about 70% ... around 40 years”
-
112
The essay, section 5confirmed
Read your job like the study: look up occupation on O*NET OnLine, read task list, ask three questions
The page says “O*NET OnLine occupation pages list tasks (e.g. 53-7064.00); study asks 'can a robot today perform this task, and if so under what circumstances?' plus cost”
-
113
The essay, section 5confirmed
Same three questions work for chatbots and desk work
The page says “LLM exposure measures that assess capabilities but not costs (method parallels LLM exposure)”
-
114
The essayconfirmed
Pull quote taken from the essay's own words
The page says “Identical sentence appears in sections[0].paragraphs[2]”
Checked for consistency within the issue itself
-
115
Further reading, item 1confirmed
What work can robots do? (Anthropic) - the report, with four settings, example tasks, cost method
The page says “Page title 'What work can robots do?'; contains E0-E3, examples and cost method”
-
116
Further reading, item 2confirmed
O*NET OnLine; task lists sponsored by the Labor Department
The page says “O*NET OnLine is sponsored by the U.S. Department of Labor”
-
117
Further reading, item 3confirmed
Five Million Robots now Operate in Factories Globally (IFR)
The page says “Headline: Five Million Robots now Operate in Factories Globally”
-
118
Further reading, item 4confirmed
Labor market impacts of AI: A new measure and early evidence (Anthropic)
The page says “Title: Labor market impacts of AI: A new measure and early evidence”
Anthropic labor-market post (further reading only), at the passage · whole page
-
119
Further reading, item 4confirmed
Background from March; same approach applied to chatbots/desk work
The page says “Published March 5, 2026; 'Exposure is typically defined at the task level' (LLM exposure)”
Anthropic labor-market post (further reading only), at the passage · whole page
The checker's notes
- Window: all cited facts dated on or before D (2026-10-02). Anthropic report Sept 30, 2026; IFR Sept 24 and Sept 30, 2026; DigitalToday 2026-10-01 15:16 KST; MIXED dated Oct 2, 2026 (same day as D, not after; its facts duplicate the Sept 30 Anthropic report, which is co-cited). O*NET overview shows 'Site updated October 6, 2026' but the cited facts (DOL sponsorship, 900+ occupations, 19,000+ tasks) are long-standing.
- Background: the only dated item older than the 45-day window is the March 5, 2026 Anthropic labor-market report in Further reading, labelled 'Background from March'. 1977 backtest figures are the Sept 30 report's own findings.
- No 'ahead' block; 'What to watch' section relies only on figures published by Sept 30 (Anthropic, IFR).
- Quotes all 14 words or fewer and attributed: 'can replace 74 percent of manual labor' (headline), 'mostly in limited settings', O*NET task line (10 words), 'Dig trenches', 'Teach dance students', 'difficult to predict exactly', 'requires many judgment calls'.
- Minor precision notes (not errors): E0 26% is 100 minus 74; the page text says 'about a quarter'. E1 50% is the page's 'half'. The 'highly exposed workers' comparison on the page is top-quintile vs unexposed workers. IFR 117,500 is from the release body; its subheading says 117,000. The report's own $49,000 vs $45,000 figures differ by $4,000 while it states ~$2,500; the draft correctly reports the page's $2,500.
- Final paragraph of sections[4] (no src) is general advice with no checkable factual claim; not counted.
- Fetch method: pages opened with WebFetch on 2026-10-24 (direct curl blocked by proxy); passages are as returned by WebFetch.