Our audience’s favorite agent frameworks

Plus: free alternatives to OpenAI's deep research, the best models for enterprise agents, and more

Author

Michael Cunningham
February 06, 2025

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

The good news: we now know the reasoning process DeepSeek R1 has been using to achieve its spectacular results. The bad news:

In today’s issue…

  • How Microsoft is embracing the agentic moment

  • MIT’s agent index

  • Free alternatives to OpenAI’s deep research

  • The best models for enterprise agents

…and more

🗳️ POLL RESULTS

On Monday, we asked all of you which agent framework was your favorite. Thank you to everyone who participated in our first Building AI Agents poll!

CrewAI came in with a convincing lead, followed by LangGraph and then AutoGen. Surprisingly, not one participant used AutoGPT, showing that being the first doesn’t necessarily make you the most enduring.

Overall, the results showed the community converging on a handful of major frameworks, but without any single one dominating the rankings.

📰 NEWS

Source: LangChain

Early-bird tickets for LangChain’s Interrupt Conference on AI Agents in May are now available starting at $499.

Agent.ai, a platform founded by HubSpot founder Dharmesh Shah which allows customers to identify AI agents potentially useful to their businesses, has surpassed 500,000 users, growing by 10x in just 6 months.

The company’s software engineering agent, which won significant acclaim and raised questions about the future of programming jobs upon its release last September, is now publicly available as an IOS app.

Firecrawl, a startup which makes websites accessible to LLMs, posted a job creating examples of its tech, available only to an AI agent or someone able to build one for the task.

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

🤝 WITH GAMMA

An entirely new way to present ideas

Gamma’s AI creates beautiful presentations, websites, and more. No design or coding skills required. Try it free today.

🛠️ USEFUL STUFF

Source: Wikipedia

Although deep research’s $200/month price tag has turned away many users from accessing its advanced capabilities, several highly capable open-source alternatives have quickly become available.

An intro to building agents within Azure AI Agent service, the Microsoft cloud platform’s framework for creating and deploying enterprise-grade agentic AI.

Salesforce maintains a detailed leaderboard showing various LLMs’ performance, price, and speed on a wide range of tasks across the company’s customer relationship management (CRM) benchmark.

💡 ANALYSIS

Source: Wikimedia Commons

A blog post by the company giving an overview of their efforts in the agent space, particularly around its flagship Copilot agents.

This piece briefly discusses what AI agents are, and walks through the various steps of deploying them in an enterprise setting, from preparing proper integrations to selecting the right frameworks to evaluating the agents once they have been built.

A discussion of the major actions companies must take in order to make their data and systems accessible to agents in a secure way.

🧪 RESEARCH

Source: AI Agent Index

A detailed database by MIT of 67 major publicly-available agentic systems, including their purpose, builder, evaluations and more, based on this study.

Multi-Agent System Search (MASS) is a new automated method for optimizing agents’ architecture and prompts to increase performance. The authors use lessons from this process to abstract some universal principles of agent design, such as the importance of optimizing individual agents before the entire system.

This paper provides a case study in the use of agents to reproduce and fix bugs in an enterprise setting, finding that the authors’ agentic approach significantly outperforms the existing state-of-the-art technique.

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