Meta joins the business agent race

Plus: GitHub's agentic IDE, the first AI Agent Congress, and more

Author

Michael Cunningham
November 25, 2024

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

Sometimes, your best bet for evaluating your agent’s performance is a fancy benchmark published at NeurIPS, but other times it’s just…vibes maaan

In today’s issue…

  • Meta’s move into business agents

  • GitHub releases an agent-powered IDE

  • The first AI Agent Congress

  • The state of enterprise gen AI

…and more

🔍 SPOTLIGHT

The epic battle between Microsoft and Salesforce for control of the AI agent market for business might soon have a third contender.

Last week, Axios reported that Meta will be creating a new unit devoted to developing AI tools for the businesses which use the company’s apps, who number over 200 million. In a major hint at the group’s focus, Meta hired away Clara Shih, the CEO of Salesforce AI, to head it. With Salesforce’s aggressive focus on AI agents a core part of its software-as-a-service (SaaS) offering to businesses, it seems likely that Meta will capitalize on Shih’s experience to develop agentic AI services offering deep automation, rather than only more basic functions such as chatbots.

The tech giant has its work cut out for it, as it is entering a crowded field. Just last week, Salesforce’s major rival Microsoft announced a dizzying array of new features for its agentic business AI platform Copilot and for its Azure cloud ecosystem. Microsoft, long a dominant player in the business-to-business (B2B) SaaS space with Azure and its Microsoft 365 platform, has been fending off a challenge from Salesforce in which both companies have made AI agents a core part of their software suite. Many robotic process automation (RPA) companies such as UiPath and Automation Anywhere have been pivoting into the agent space as well, while a plethora of smaller providers have sprung up, many seeking to capture specific verticals.

The reason for the influx of business agent builders is not difficult to identify. The existing B2B SaaS market is already sized in the hundreds of billions of dollars, which, in turn, is dwarfed by a factor of roughly 35 by the sum that companies spend annually on labor—for most, it is by far their greatest expense. With increasingly serious discussions occurring about large swathes of the labor market being automated in the not-too-distant future, the potential annual value of AI agents in business is easily in the trillions of dollars and may reach into the tens of trillions. This is the prize for which which Meta may soon be competing with Microsoft and Salesforce.

For now, Meta’s generative AI offerings for businesses are relatively modest, such as a chatbot and an image generator for advertisers—nothing which approaches the sophistication of its rivals’ comprehensive, enterprise-wide automation suites. If the company’s new unit really does represent an effort to play catch-up in the agent league, it will need to dedicate significant resources and focus to closing the gap. Nevertheless, with an operating income of over $62.4 billion dollars over the past year, Meta certainly has the financial firepower—and the nearly boundless rewards on offer give it all the incentive it needs.

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

Refer one person and we’ll send you a set of weekly bonus links to additional AI agent news for a month! See here for a sample

🤝 WITH ARTISAN

Hire an AI BDR & Get Qualified Meetings On Autopilot

Outbound requires hours of manual work.

Hire Ava who automates your entire outbound demand generation process, including:

  • Intent-Driven Lead Discovery Across Dozens of Sources

  • High Quality Emails with Human-Level Personalization

  • Follow-Up Management

  • Email Deliverability Management

đź“° NEWS

Source: GitHub

Among the many features unveiled at its Ignite conference last week, Microsoft released a preview for GitHub Copilot Workspace, which enables agentic software development—though there is some controversy over whether it or Windsurf can claim the title of “first agentic IDE”.

Blitz Ventures hosted many of the top figures in the AI agent field, such as Adam Silverman of AgentOps and Chi Wang and Qingyun Wu of AG2 (formerly AutoGen), for the first of what is implied to be more AI agent congresses.

In one of the largest funding rounds in Y Combinator’s history, startup Wordware raised $30 million to develop its no-code IDE for building AI agents.

🛠️ USEFUL STUFF

Source: Menlo Ventures

Hot on the heels of LangChain’s State of AI Agents report, Menlo Ventures released its State of Generative AI in the Enterprise report, based on a survey of over 600 professionals in the space, predicting that the next wave of AI will be agentic automation

Agent framework AG2, spun off from and maintained by much of the founding team of Microsoft’s AutoGen, released two major new features—a “CaptainAgent” to build agent teams to solve a given task, and a SwarmAgent inspired by OpenAI’s framework of the same name.

The creators of the original AI agent, AutoGPT, have released a new framework for multi-agent collaboration called Agent Blocks, designed to enable workflows containing as many as thousands of specialized agents overseen by a small set of top-level supervisors which interface with humans.

This GitHub repo by the Google Cloud team demonstrates the process of building and deploying a genAI application in GCP.

NNetnav is a new method for generating web GUI navigation demonstrations to improve internet browser agents.

đź’ˇ ANALYSIS

Created with Dall-E 3

A piece focusing on the use of agents to make meaningful business decisions, including some examples of companies which have already integrated agents into their workflows.

🧪 RESEARCH

Balrog from Lord of the Rings | Source: Den of Geek

BALROG is a novel benchmark for assessing the performance of text and visual LLMs on a set of games requiring complex, multi-step agentic reasoning.

This paper proposes LLM agents that can interact with the world by writing their own callable Python functions, enabling them to solve problems which would otherwise be impossible with a fixed tool set.

The authors of this paper surveyed 1,052 demographically-representative Americans on their age, gender, race, political ideology, and other factors, finding that LLMs told to take on their personas could mimic their responses to questions with a high degree of accuracy, implying that agentic simulations are a promising avenue for studying social behavior.

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