Anthropic's agent takes over your computer

Plus: Microsoft and IBM make big plays in AI agents, and xAI recruits agent engineers for elite team

Author

Michael Cunningham
October 24, 2024

Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!

You know what to look for from tech in San Francisco and New York, but if you want all the important details, we here in Chicago have got you covered

With a host of major players and tech companies announcing new agentic capabilities, it’s been a big couple of days, so let’s dive right in

In today’s issue…

  • Anthropic demos computer control agents, upgrades Sonnet 3.5

  • Microsoft rolls out agent building in Copilot Studio and releases new agents

  • IBM unveils agent-focused LLMs, teams up with CrewAI

  • xAI hiring agent engineers for “Starfleet” team

…and more

📰 NEWS

Source: Anthropic

The OpenAI rival released demonstration videos of an upcoming agentic system designed to take full control of a user’s computer to execute complex, human-like tasks, and revealed upgraded capabilities for its flagship Sonnet 3.5 LLM.

In an effort to counter Salesforce’s wholesale push into AI agents, Microsoft announced that users would be able to create their own autonomous agents within Copilot Studio and that it plans to launch 10 new agents in its Dynamics 365 enterprise AI suite.

In quick succession, IBM unveiled Granite 3.0, a new generation of LLMs designed specifically for agentic business workflows available via the company’s cloud services, and announced a partnership with leading agent framework provider CrewAI to integrate its technology into IBM’s watsonx.ai

Elon Musk’s AI company posted several positions for software engineers on a mysterious new “Starfleet” team, which seems to be dedicated to building advanced agentic AI.

If you find Building AI Agents valuable, forward this email to a friend or colleague!

🛠️ USEFUL STUFF

Source: UC Berkeley

UC Berkeley’s massive open online course (MOOC) on AI agents is capping off with a virtual and in-person hackathon featuring over $100k in prizes and credits, open even to non-participants in the MOOC.

The startup’s tech, used by AI search engine Perplexity, allows users to safely run code generated by LLM agents in a sandboxed environment.

This short course by Andrew Ng’s DeepLearning.AI educational platform—with CrewAI founder João Moura as its instructor—teaches users to build multi-agent systems.

YC-backed startup Protegee offers an API which allows customers to securely pay AI agents directly rather than through a human intermediary.

💡 ANALYSIS

Source: Openclipart.org

This piece speculates on the different impacts AI agents will have on a wide range of sectors in the coming year.

Why Anthropic’s announcement will be such a game changer—once the tech is mature enough to use.

🧪 RESEARCH

Source: arXiv

Web browsing agents which interact with the world through text-based APIs are often assumed to be limited by their inability to interact with graphical user interfaces. This new paper challenges that assumption, finding that API agents can outperform visual ones.

Thanks for reading! Until next time, keep learning and building!

If you have any specific feedback, just reply to this email—we’d love to hear from you

Follow us on X (Twitter), LinkedIn, and Instagram