Neura Market
Eric Zhou & Yoko Li & Seema Amble & Jennifer Li — a16z - The Rise of Computer Use and Agentic Coworkers - August 2025 logo

Eric Zhou & Yoko Li & Seema Amble & Jennifer Li — a16z - The Rise of Computer Use and Agentic Coworkers - August 2025

Free

Exploring the rise of computer-using AI agents for autonomous digital work

FreeFree tier
Type
Open Source
Founded
2009
Company
Andreessen Horowitz
Visit Website GitHub (38 stars)

About Eric Zhou & Yoko Li & Seema Amble & Jennifer Li — a16z - The Rise of Computer Use and Agentic Coworkers - August 2025

A market analysis article from a16z exploring the rise of computer-using AI agents that can autonomously execute end-to-end digital workflows. The article argues that computer use is the key enabler for true AI agents, expanding tool accessibility and reasoning capability. It discusses current limitations of agent offerings (resembling advanced RPA) and highlights emerging projects like OpenAI's ChatGPT Agent, Claude, Project Mariner, Manus, and Context. The vision is agents that can perform tasks such as finding office space, updating CRM, notifying colleagues, and generating reports without human input, while working with both APIs and human-like computer interactions.

Key Features

Autonomous, task-oriented AI agents operating within browser and desktop environments
Computer use enables agents to interact with any software like humans (clicking, logging in, using legacy software)
End-to-end digital workflow execution without human input
Tool accessibility beyond APIs via computer-using models
Reasoning capability through end-to-end training on action sequences or reinforcement learning
Scalable and retrainable like human teams

Pros & Cons

Pros
  • Can tackle complete, end-to-end digital workflows autonomously
  • Does not require APIs or custom integrations for tool access
  • Works with existing software and legacy systems
  • Agents can be retrained or scaled as needed, like human teams
  • Emergent capabilities may allow agents to solve context retrieval by autonomously exploring
Cons
  • Current agents still fall short of full autonomy, resembling advanced RPA
  • LLM-based agents rely on intricate prompt engineering and predefined workflows
  • Not yet fully autonomous for complex, unstructured tasks

Best For

Finding new office space: identifying requirements, researching locations, scheduling tours, negotiating leasesLocating a document in a database, extracting key details, updating Salesforce, notifying colleagues on Slack, generating compliance reportsHandling legacy software and behind-the-scenes glue work that ties up human workersOperating software when programmatic access isn't possible

FAQ

What is computer use in the context of AI agents?
Computer use refers to AI agents that can directly interact with computer interfaces like humans do—clicking through user interfaces, logging in, sending files, and working with legacy software—rather than relying solely on APIs.
Why does computer use matter for AI agents?
Computer use dramatically expands both the number of tools an agent can access and its ability to reason across them. It gives agents the breadth to work with any software and the intelligence to chain actions into full workflows.
What are some examples of computer-using AI agents mentioned?
The article cites OpenAI's ChatGPT Agent, Anthropic's Claude, Google's Project Mariner, and startups like Manus and Context.