Automations Early Footprint

Skip to content The state of sovereign AI adoption: What enterprise leaders need to know. Read now

Products

Solutions

Resources

Blog

Research

Company

Sign in

Request a demo

Platform North

Enterprise-ready AI for business

Compass

Intelligent search and discovery

Models Command

Generative language models

Transcribe New

Speech recognition model

North Mini Code

Agentic coding model

Parse New

Document parsing model

Embed

Semantic representation model

Rerank

Retrieval optimization model

Models Overview

Product Products Overview

Total Cost of AI Ownership

Pricing

Featured Command: High-performance generative AI models for real-world applications

Deploy Model Vault

Dedicated model inference platform

Private Deployments

On-prem or isolated VPCs

Security

Protect your data at every stage

See deployment options

By Industry Financial Services

Public Sector

Technology

Telecommunications

Energy and Utilities

Healthcare and Life Sciences

Manufacturing

Featured Model Vault provides fully-isolated, performant inference with Saas simplicity

Insights Customer Stories

For Developers Developers

Models Overview

Docs

Discord

LLM University

Connect Partners

Events

Webinars

Merch Store

Featured How CoreWeave used Cohere North to transform its customer support in 90 days

Blog

The latest news, launches, and insights

Read more

The state of sovereign AI adoption in 2026

Cohere and the University of Waterloo launch partnership to strengthen Canada’s AI talent pipeline

Introducing North Automations: Intelligent workflow orchestration

Research Cohere Labs

Cohere’s ML research lab

Explorations Future(s) of Work

How will AI change the way we work?

Aya Models

Multilingual AI at scale

All Papers

Initiatives Research Scholars

Finding the new generation of ML talent

Open Science Community

Championing global, open science

Catalyst Grants

Supporting impactful ML endeavors

Resources Blog

Hugging Face

Events

Featured The future of work debate has an evidence problem

About

Careers

Newsroom

Sep 03, 2026

13 minute read

Automation’s Early Footprint Where AI Agents Are (and Aren’t) Being Built

Key takeaways We aggregated seven public directories of AI tools into a new dataset: the Agentic Task Ecosystem (ATE) , a corpus of roughly 696,000 published tools across 123,000 MCP servers , the largest open dataset of its kind. Under a strict test that asks whether a tool can actually carry out an occupational task rather than merely inform someone doing it, only 2.6% of tools clear the bar. Automation of recognized work is a small slice of a very large ecosystem . Among those that group into recognizable categories, three patterns dominate: existing work represented at a finer grain than occupational databases record, infrastructure for running agents themselves, and a small amount of genuinely new work, most of it the work of managing agents. Nearly half of U.S. occupations have no agentic tools represented. Of 923 occupations, 419 show no agentic tool activity of any kind. What gets built follows what can be built . Expert judgments of technical feasibility predict which occupations receive tools; workers' own preferences about what they would like automated predict nothing. The question that decides labor market outcomes isn't how much of an occupation is exposed to automation, but which parts . Across 178 occupations, we ranked required work tasks from most routine to most specialized. In healthcare and computing, we find agentic tools exist for tasks toward the specialized end of the occupation, leaving humans the more routine remainder. In legal, production, and sales occupations, tools stay at the routine edges, leaving the specialized core with humans. Specialized work resists automation when it is physical or interpersonal, and gives way when it is already conducted through software.

Hundreds of thousands of tools have been published to let AI systems take action on our behalf: write code, query databases, update records, schedule meetings, move information between applications. But what does this rapidly growing layer of agentic AI infrastructure actually add up to? What work are we building AI to automate?

Most of what we know about AI and work comes from one of two vantage points. Some studies estimate what AI could theoretically do, calculating exposure based on the percent of occupational tasks an AI system could plausibly handle. Others look at what people ask AI to do, analyzing millions of conversations to see which tasks users bring to a chatbot [ AI Observatory, 2026 ; Handa et al., 2025 ; Iscenko et al., 2026 ; OpenAI, 2026 ; Tomlinson et al., 2025 ]. Both represent distinct pieces in the larger puzzle of how AI is impacting work, however, they do not represent the full picture on their own. At Cohere Labs, we are committed to helping assemble this puzzle, and our research presents new evidence from a third, complementary vantage: the agentic-task ecosystem .

We collected nearly 700k tools from 123k public MCP servers to create the Agentic Task Ecosystem (ATE) dataset and ask, what tools have developers built to enable AI agents to fully automate certain work tasks? What we found complicates the idea that agents are simply swallowing occupations one task at a time. What is a MCP tool? In the past two years, AI systems have moved from generating text to taking autonomous actions. A model that once only answered inside a chat window can now query a database, edit a file, open a pull request, schedule a meeting, or run a sequence of these tasks in order.

The Model Context Protocol (MCP) is an open standard for connecting AI systems to external software and data. One server might let a model work with GitHub, another might give it access to a calendar, a spreadsheet, a payment system, or a company's internal documents. In practice, a developer writes an MCP server that consists of a set of tools, each with a short description of what it does. MCP tools as a record of what we are automating Economists studying automation usually weigh three variables against each other: (1) what work humans do, (2) what work gets handed to machines, and (3) what genuinely new work appears for both. Automation displaces workers from tasks machines can perform whereas genuinely new tasks that emerge, where humans hold the advantage, reinstate them. Whichever effect is larger determines whether a technology raises or lowers demand...