Cohere Labs releases largest ever public dataset of AI agent tools
Researchers at Cohere Labs – the open science team at sovereign AI leader Cohere – launch the Agentic Task Ecosystem (ATE), their new dataset of AI agent tools. The largest collection of tools created to date, this new dataset aggregates nearly 700,000 tools from public Model Context Protocol servers. It will support new research into AI and the future of work and is part of a broader programme of work in this area led by the Cohere Labs team.
Marzieh Fadaee, Head of Research at Cohere Labs, said:
"Understanding the future of work requires looking beyond what AI is capable of and studying how those capabilities actually enter the economy. What we show in this research is an emerging reorganisation of work: agentic tools are changing the boundaries between existing tasks in jobs, and even more interestingly the role that human expertise plays within those new boundaries."
Joelle Pineau, Chief AI Officer at Cohere, said:
“Leaders in businesses and governments must ground decisions in robust evidence about how AI will reshape work, workers’ lives, and our economies and societies. Cohere Labs’ research contributes to that evidence base, aiming to help shape a future that expands human potential and delivers lasting benefits for society.”
View the dataset and find out more
This new dataset has been launched as part of the London School of Economics and Political Science’s 2026 Global Forum for AI and the Social Sciences. To build momentum and unlock new forms of coordinated and collective action, selected organisations from around the world have been invited to use the Global Forum as a platform for releasing work, launching new initiatives, or making commitments related to the Forum's annual focus.
Statement from Cohere Labs
Many existing explorations of AI and work focus on either theoretical capabilities (what AI could do) or user demand (what people ask AI to do). Data is lacking on what tools are actually being built to enable AI agents to automate work tasks.
Cohere Labs have created Agentic Task Ecosystem (ATE), a new dataset of AI agent tools to measure the "supply side" of automation – what developers have judged ready enough to hand to machines.
Analysing these tools against the Occupational Information Network database of occupational tasks, Cohere Labs have found that only a small fraction, about one tool in 40, performs a recognised work task from end to end, and that for a great many occupations no one is building anything yet.
Across 178 occupations, Cohere Labs ranked required work tasks from most routine to most specialised. In healthcare and computing, they found agentic tools exist for tasks toward the specialised end of the occupation, leaving humans the more routine remainder. In legal, production, and sales occupations, tools stay at the routine edges, leaving the specialised core with humans. Specialised work resists automation when it is physical or interpersonal, and gives way when it is already conducted through software.
This matters because it says about the nature of automation within jobs, not just the volume of it. Expertise-raising automation may increase the wages of those workers, but limit the pool of qualified applicants. As the inverse, expertise-lowering automation can be viewed as democratising expertise, but this doesn’t account for how skills and expertise are shaped by experience. Entry-level workers often learn through the routine tasks experienced colleagues and mentors hand-off.
ATE will support researchers to study these impacts, as well as to track and influence what capabilities developers choose to build into tools, complementing other existing evidence sources and providing an early signal for labour market impacts, visible before impacts show up in employment or wage data. Cohere Labs have released ATE openly to support other research into AI and the future of work.
The future of work won’t be determined by AI capabilities alone, but by what tools are built to allow AI agents to apply those capabilities to work tasks. Understanding those tools is crucial for supporting workers, economies, and societies through the transition brought by AI agents.
About Cohere
Cohere, founded in 2019, is the world's leading sovereign AI company building foundation models and end-to-end AI products to solve real-world business problems. It partners with organisations to deliver seamless integration, deep customisation, and user-friendly solutions. Its all-in-one platform provides maximum security, privacy, and deployment flexibility across public clouds, private environments, and on-premises.
Cohere Labs is Cohere’s frontier research lab and open research community dedicated to contributing fundamental research in machine learning, working to solve some of the field's most challenging problems and create more points of entry into machine learning research. The Labs team has released over 100 research papers in just three years.
About the Global Forum for AI and the Social Sciences
The Global Forum for AI and the Social Sciences is a flagship event bringing together leaders across government, industry, academia, and civil society from around the world. Organised by the LSE Data Science Institute, the Global Forum aims to convene international dialogue at the intersection of AI and the social sciences and to mobilise expertise, resources, and partnerships to ensure AI serves human and societal needs.
The inaugural edition of the Global Forum will take place on 16 and 17 September 2026 with a thematic focus on how AI is reshaping work and what governments, firms, and workers must do to prepare for the changes ahead.