The new AutoGen is here

Plus: agents struggle to overcome "enterprise hell", agentic software engineering at Google, and more

Author

Michael Cunningham
January 16, 2025

In partnership with

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

Apparently even the original Wolf of Wall Street is launching his own AI sales agent. I’m quite confident that nothing could possibly go wrong here.

Interested in reaching our highly engaged audience of agent builders and business executives? Check out our metrics and contact us here!

In today’s issue…

  • AutoGen’s overhaul is here

  • LangChain’s new ecosystem of “ambient agents”

  • Can AI agents survive enterprise hell?

  • Agent-based software engineering at Google

…and more

📰 NEWS

AutoGen ecosystem introduced in v0.4 | Source: Microsoft

The long-awaited overhaul of Microsoft’s AutoGen agent framework, dubbed AutoGen v0.4, is now live. Initiated prior to the schism that produced spinoff framework AG2, AutoGen’s upgrade was intended to address growing pains introduced by its predecessor’s rapid evolution.

ChatGPT users can now schedule simple agentic tasks, such as compiling and sending a weekly news briefing.

Customers on the free tier of Microsoft’s 365 Copilot Chat will now be able to access Copilot AI Agents on a pay-as-you-go basis.

The luxury automaker will be one of the first to integrate Google’s just-unveiled Automotive AI Agent, which will allow equipped cars to provide user assistance through natural language conversation.

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

🤝 WITH HUBSPOT

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

🛠️ USEFUL STUFF

Source: LangChain

An effort by LangChain to build an AI agent that reproduces many of the capabilities of an executive assistant, such as answering emails and scheduling meetings. This agent has reportedly been used by the LangChain staff internally for months, and is the first of what they foresee as a new ecosystem of “ambient agents”.

Agent memory startup Letta has launched a new visual IDE for developing AI agents, with memory features and transparency into agents’ context window as major points of differentiation from prior frameworks.

The LLM provider has overhauled their guide to function calling for developers, enhancing it with best practices and illustrative examples.

💡 ANALYSIS

Created by the author using Dall-E 3

The author of this piece investigates the challenges of integrating agents into enterprise workflows through hands-on experiments with Salesforce’s Agentforce and Anthropic’s Computer Use.

This provocative piece tackles the question of whether AI agents are really the novel and transformative technology they are billed as, or merely a repackaging of robotic process automation (RPA).

The developer of E2B’s fully open-source alternative to Anthropic’s Computer Use offering describes how it works and the difficulties he overcame to build it.

Employers overwhelmed by waves of job applications are using AI agents to screen, schedule, and even interview candidates, achieving significant time savings—and finding that many applicants actually prefer them to human recruiters.

🧪 RESEARCH

Source: Wikipedia

In this paper, Google engineers sought to go beyond popular agentic software engineering evals such as SWE-Bench to test agents’ abilities to fix real bugs in an enterprise setting.

A deep dive on the process of building a new class of large language models: large action models (LAMs), which are designed and trained from the ground up to power agentic systems.

This paper presents a new method of generating synthetic data to improve LLM reasoning, in which multiple LLM agents interact with each other and are fine-tuned on the data generated from the interactions.

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