Cohere · Sep 3, 2026
Cohere reports that agentic tools reach different parts of different jobs
Markdown · JSON · Original source
Cohere’s article was published on September 3, 2026. It presents the Task Ecosystem (ATE) as a record of public tools developers make available for automating work, not of tools employers have adopted. Under Cohere’s test, only a small share carry out a recorded work task from end to end. The article also reports that many occupations have no agentic tools represented. Where tools match work, Cohere argues, their location within a job matters more than the amount of tooling. Tools reach toward specialized tasks in healthcare and computing, while in legal and production occupations they tend to reach routine tasks, leaving specialized work to people. Cohere argues that specialized work is not uniformly hard to automate: it is harder when physical or interpersonal, and more tractable when already conducted through software. These patterns do not establish effects on employment or wages. ATE measures supply, not adoption or economic impact. The article asks how automation may affect the way entry-level workers gain expertise. It concludes that the future of work depends on more than AI capability alone, and says early, meaningful interventions are needed for machine capabilities to expand human opportunities and raise labor’s value.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
Cohere · Article
Source details
Original source date: Sep 3, 2026 (2026-09-03T17:02:00.000Z).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-06T04:43:23.537Z. First observed: 2026-11-05T07:39:42.268Z. Retrieved: 2026-11-05T07:41:37.066Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on November 5, 2026; originally published September 3, 2026. ATE covers public directories; Cohere says bespoke internal MCP servers are absent, so the dataset may miss tools, particularly for back-office processes. Cohere says the dataset measures tool supply, not whether companies deploy tools, workers use them, or economic impacts follow. The article does not establish how the reported patterns will affect employment, wages, or entry-level workers’ paths to expertise.
Open question
If routine work is increasingly done by machines, how will entry-level workers gain the experience needed to become experts?
Source-stated
Cohere reports that ATE contains 696,291 tools from public MCP server listings collected in May 2026.
We collected 696,291 tools from 123,069 public MCP server listings across seven directories in May 2026.
Source-stated
Under Cohere’s test for carrying out a recorded work task end to end, 2.6% of tools meet the standard.
Under that standard, about one tool in forty – 2.6% – performs a recorded work task from end to end.
Source-stated
Cohere reports that 419 of 923 occupations in O*NET have no agentic tool activity.
419 of 923 occupations represented in O*NET have no agentic tool activity at all.
Source-stated
Cohere reports contrasting patterns by occupation: tools reach toward specialized work in healthcare and computing, and stay at routine edges in production and legal occupations.
In healthcare and computing occupations the tools that exist reach toward the specialized end of the job leaving humans the more routine remainder
Source-stated
Cohere says ATE measures the supply of tools, not their adoption or economic impact.
ATE is a measure of supply, not adoption nor economic impact
Source-stated
Cohere argues that specialized work is harder to automate when it is physical or interpersonal, and comparatively tractable when already conducted through software.
specialized work is not uniformly hard to automate – it is hard to automate where it is physical or interpersonal, and comparatively tractable where it is already conducted through software.
Source-stated
Cohere argues that the future of work depends on more than AI capability alone and says early, meaningful interventions are needed for futures that raise labor’s value.
Futures in which machine capabilities expand human opportunities and raise the value of labor are possible, but require early and meaningful interventions.
Source and contributor context
Cohere · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Agentic AI
AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer.
Editorial provenance
Version 1. Publication ID: 0a6f24a5-95b8-c8d9-9a66-f3cd247c73fb.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-06T04:39:23.927Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 9f07514a75fbaaea3ce61e3808564079eecce316b23451acf7c59863cd77ef5a. Source text hash: bc504b7499c6b284c35d19cea3fdf9c0fcac462633cf02b74cbb883468b1c52e.
Revision and source-check history
Version 1 · 2026-11-06T04:43:23.537Z
Cohere reports that agentic tools reach different parts of different jobs
Cohere’s article was published on September 3, 2026. It presents the Task Ecosystem (ATE) as a record of public tools developers make available for automating work, not of tools employers have adopted. Under Cohere’s test, only a small share carry out a recorded work task from end to end. The article also reports that many occupations have no agentic tools represented. Where tools match work, Cohere argues, their location within a job matters more than the amount of tooling. Tools reach toward specialized tasks in healthcare and computing, while in legal and production occupations they tend to reach routine tasks, leaving specialized work to people. Cohere argues that specialized work is not uniformly hard to automate: it is harder when physical or interpersonal, and more tractable when already conducted through software. These patterns do not establish effects on employment or wages. ATE measures supply, not adoption or economic impact. The article asks how automation may affect the way entry-level workers gain expertise. It concludes that the future of work depends on more than AI capability alone, and says early, meaningful interventions are needed for machine capabilities to expand human opportunities and raise labor’s value.
Original source
Cohere · Article
Source details
Original source date: Sep 3, 2026 (2026-09-03T17:02:00.000Z).
Date basis: Date stated on the publisher’s page.
Added to Escape Meridian: 2026-11-06T04:43:23.537Z. First observed: 2026-11-05T07:39:42.268Z. Retrieved: 2026-11-05T07:41:37.066Z.
Limits
Text as retrieved on November 5, 2026; originally published September 3, 2026. ATE covers public directories; Cohere says bespoke internal MCP servers are absent, so the dataset may miss tools, particularly for back-office processes. Cohere says the dataset measures tool supply, not whether companies deploy tools, workers use them, or economic impacts follow. The article does not establish how the reported patterns will affect employment, wages, or entry-level workers’ paths to expertise.
Open question
If routine work is increasingly done by machines, how will entry-level workers gain the experience needed to become experts?
Source-stated
Cohere reports that ATE contains 696,291 tools from public MCP server listings collected in May 2026.
We collected 696,291 tools from 123,069 public MCP server listings across seven directories in May 2026.
Source-stated
Under Cohere’s test for carrying out a recorded work task end to end, 2.6% of tools meet the standard.
Under that standard, about one tool in forty – 2.6% – performs a recorded work task from end to end.
Source-stated
Cohere reports that 419 of 923 occupations in O*NET have no agentic tool activity.
419 of 923 occupations represented in O*NET have no agentic tool activity at all.
Source-stated
Cohere reports contrasting patterns by occupation: tools reach toward specialized work in healthcare and computing, and stay at routine edges in production and legal occupations.
In healthcare and computing occupations the tools that exist reach toward the specialized end of the job leaving humans the more routine remainder
Source-stated
Cohere says ATE measures the supply of tools, not their adoption or economic impact.
ATE is a measure of supply, not adoption nor economic impact
Source-stated
Cohere argues that specialized work is harder to automate when it is physical or interpersonal, and comparatively tractable when already conducted through software.
specialized work is not uniformly hard to automate – it is hard to automate where it is physical or interpersonal, and comparatively tractable where it is already conducted through software.
Source-stated
Cohere argues that the future of work depends on more than AI capability alone and says early, meaningful interventions are needed for futures that raise labor’s value.
Futures in which machine capabilities expand human opportunities and raise the value of labor are possible, but require early and meaningful interventions.
Source and contributor context
Cohere · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Agentic AI
AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer.
Editorial provenance
Version 1. Publication ID: 0a6f24a5-95b8-c8d9-9a66-f3cd247c73fb.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-06T04:39:23.927Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model.
Observation: 9f07514a75fbaaea3ce61e3808564079eecce316b23451acf7c59863cd77ef5a. Source text hash: bc504b7499c6b284c35d19cea3fdf9c0fcac462633cf02b74cbb883468b1c52e.
passed · 2026-11-06T04:39:23.927Z · anthropic/claude-opus-5.5
Publication: 0a6f24a5-95b8-c8d9-9a66-f3cd247c73fb
attribution — pass: The source is a single Cohere blog article presenting Cohere Labs research. The reading attributes the dataset, findings and arguments to Cohere and keeps them as Cohere's reported claims rather than established facts.
At Cohere Labs, we are committed to helping assemble this puzzle
summary — pass: The summary leads with the ATE dataset and explains it in plain language as public tools that developers make available, not tools that have been adopted. It conveys the 2.6% end-to-end finding, the uncovered occupations, the contrast between specialized and routine work by field, and the software-versus-physical/interpersonal argument. It preserves the open question about how entry-level workers become experts. The central term agentic is defined beside the reading. The omission of sales occupations is minor.
The question that decides labor market outcomes isn't how much of an occupation is exposed to automation, but which parts.
date — pass: evidence.originalPublished is 2026-09-03T17:02Z, and the reading states September 3, 2026, which matches the UTC day. The May 2026 collection date is given at month precision, as in the source. There is no hindsight framing, and forecasts remain attributed. The evidence kind is publisher-page, so no archive disclosure is needed. modifiedAt equals the publication time.
We collected 696,291 tools from 123,069 public MCP server listings across seven directories in May 2026.
support — pass: Every claim is source-stated and has a matching excerpt. The caveats that the dataset measures supply rather than adoption, and that it excludes internal servers, are drawn from the source. The back-office point is appropriately hedged. The conclusion about interventions keeps the source's conditional framing. No unsupported benefits or costs are introduced.
ATE is a measure of supply, not adoption nor economic impact
media — pass: No media candidates were supplied and none was selected. That is appropriate.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
AI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer.
Definition referenceAI systems that plan and carry out a series of steps toward a goal, often by using other software, rather than giving one answer.
Definition reference