Easily integrate your agents with hundreds of apps

Plus: OpenAI's bonanza of new AI agent tools, Salesforce’s agent marketplace, and more

Author

Michael Cunningham
March 13, 2025

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

Just wait until the Neuralink integration comes out.

In today’s issue…

  • OpenAI releases wealth of agent tools

  • Agent integrations for Gmail, Github, Slack, Discord, and more

  • A free course on building agents with LlamaIndex

  • A roundup of agent memory frameworks

  • Nadella vs. Benioff: the showdown for the future of agentic AI

…and more

📰 NEWS

Source: Wikimedia Commons

In a major announcement, the company unveiled a host of agent tools for developers, including a new Responses API with powerful built-in tools for LLMs such as web browsing, file search, and computer control, as well as an agent SDK that builds on its earlier Swarm release.

Salesforce launched AgentExchange, a marketplace where over 200 partners provide pre-built agents to automate a wide range of business tasks, in tandem with the release of Agentforce 2dx, the latest generation of its Agentforce platform. The company has bet massively on agentic AI, describing it as a $6 trillion opportunity by 2030.

Amazon Web Services formed a team devoted to unlocking the potential of agentic AI, with its CEO Matt Garman declaring that "agentic AI has the potential to be the next multi-billion business for AWS.” This announcement came the same day as a report that Amazon is developing reasoning models similar to those of OpenAI and DeepSeek.

The company behind web browser Opera released an integral agent capable of carrying out assigned tasks in-browser, such as booking hotels and buying socks.

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

 🤝 PARTNER WITH US!

Do you have a high-quality product or service agent builders and business leaders would like to hear about?

Building AI Agents has…

👥 Thousands of readers

📈 Rapid audience growth

🙌 High engagement and satisfaction

If you’re interested in advertising with us, please reach out here!

🛠️ USEFUL STUFF

Source: Composio

Composio’s new platform provides model context protocol (MCP)-compatible servers for hundreds of common apps such as Gmail, Github, Slack, Discord, and more, allowing agents to interface with them without needing an individual integration for each one.

Template Hub is a new repository by Convergence AI where users can build and publish AI agents, as well as easily deploy those built by others.

Hugging Face’s intro course on building agents with popular framework LlamaIndex.

Mistral OCR is a new multimodal and multilingual API for optical character recognition (OCR) that can process thousands of pages per dollar, giving AI agents the ability to cheaply query huge troves of documents, including their visual and foreign-language components.

Phoenix is a platform by machine learning experiment tracking provider Arize that enables easy logging, tracing, and evaluation of agent performance.

💡 ANALYSIS

Source: Created by the author using Dall-E 3

Agents’ stateless nature is a severe limitation to their capabilities, but a variety of memory solutions from different providers are emerging. This article compares the strengths and weaknesses of several offerings by LangChain, CrewAI, OpenAI, and more.

A detailed breakdown of the bitter fight between Microsoft’s Satya Nadella and Salesforce’s Mark Benioff for control of the AI agent space—and the two competing philosophies they embody as agents transform enterprise technology.

Managing AI agents will be a daunting challenge for IT departments, with cost, security, and organization becoming major pain points.

The author of this piece narrates their experience with the quirks of OpenAI’s Operator web browsing agent and speculates on what the future of the agentic internet holds.

🧪 RESEARCH

Diagram of the authors’ agentic framework, MARBLE | Source: ArXiv

MultiAgentBench is intended as a comprehensive benchmark for evaluating agents’ general capabilities by testing them on a wide range of tasks, from coding to Minecraft navigation.

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

What did you think of today's issue?

Login or Subscribe to participate in polls.

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