Salesforce's AI agent conference

All the biggest news from the SaaS company's agent bonanza

Author

Michael Cunningham
September 23, 2024

Welcome back to Building AI Agents, your now BI-weekly guide to everything new in the AI agent field!

The AI agent space is moving faster than ever, with new announcements by Salesforce, Workday, Oracle, Microsoft, and more just in this past week. To enable us to provide you with the latest news as soon as it breaks without leaving out any important stories, we’re adding a Thursday issue to Building AI Agents, which will include all the same sections as the Monday newsletter except for the Spotlight. We hope you find it valuable!

In today’s issue…

  • Salesforce goes all-in on its Agentforce platform

  • Workday bets that “agents are the future of AI”

  • A map of the 2024 agent ecosystem

  • Sam Altman teases “level 3” agents

  • Will every white collar role have an AI agent?

…and more

🔍 SPOTLIGHT

Source: Salesforce

The 38th most valuable company in the world is now built around AI agents.

Customer relationship management (CRM) provider Salesforce held its annual Dreamforce conference in San Francisco from Tuesday to Thursday last week, and the theme was agents all the way. Over the past several months, the company has reinvented itself around the technology, spearheaded by CEO Marc Benioff, who declared a “hard pivot” to agents and ambitiously envisions the company powering a billion of them for customers within the next 12 months.

Core to this strategy is Agentforce, the company’s new agent platform and the focus of the Dreamforce conference. Agentforce, announced on September 12th, allows clients to build and deploy custom AI agents with access to company data to automate a wide range of business tasks, such as sales, human resources, and customer service. Agentforce represents a bold attempt to position the company ahead of rivals such as Microsoft, whose simpler AI “Copilots” Benioff derided as "kind of a failed idea”. Agentforce will be made widely available to Salesforce customers in October.

At Dreamforce, Salesforce unveiled a plethora of new features in its Data Cloud intended to facilitate Agentforce, such as a low-latency layer to allow AI agents to respond faster to queries, support for unstructured video and audio data that can be processed by agents, and more. Slack, which was acquired by Salesforce in 2020, is adding the ability to integrate not only Agentforce agents, but also those developed by other providers. Agentforce Studio and its low-code tool Agent Builder will enable Salesforce customers to easily create and deploy these agents into their businesses. To provide the colossal computational power necessary to run a billion or more agents, Salesforce announced a partnership with NVIDIA to leverage the company’s GPUs, as well as its ACE platform for creating lifelike avatars for customer and employee interaction. In addition, Salesforce doubled its financial backing of AI startups via Salesforce Ventures to $1 billion.

While a host of large companies are adding AI agent capabilities—more on that in the News section—Salesforce stands out as being the first to rebuild its core strategy around the technology. As the field continues its rapid advance and the enormous economic value it enables becomes increasingly clear, it will likely not be the last.

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

📰 NEWS

Source: Workday

Human resources software company Workday introduced their new Illuminate AI technology, allowing users to deploy AI agents to carry out HR tasks.

Oracle unveiled an expansive set of new agents for its Fusion Cloud Applications Suite, intended to automate repetitive tasks in human resources, supply chain, enterprise resource planning, and customer service.

T-Mobile announced a partnership with OpenAI to build a new platform called IntentCX for customer service agents.

Microsoft will be rolling out a new interface for its agent Copilot Studio, allowing even amateur users to easily build Copilot Agents and deploy them to automate tasks in Microsoft 365.

🛠️ USEFUL STUFF

Source: FlickrCC

A map of the current state of the AI agent space, divided into the ecosystem, application, management, and data layers.

A long X post on how to tell which commercial problems are suitable for automation by AI agents and which are not.

💡 ANALYSIS

Source: Wikipedia

At the aforementioned T-Mobile—OpenAI colab launch event, OpenAI’s CEO discussed the future of AI agents, hinting that “level 3” on the company’s roadmap to human-level AI—autonomous agents—is just around the corner

This article by Andreessen Horowitz claims that every white collar job will experience partial or total automation by AI agents, and outlined opportunities for agent builders to challenge existing incumbents.

🧪 RESEARCH

Source: arXiv

Test-time aggregation is a method to improve LLM reasoning by generating multiple samples and having the model self-correct them, but LLMs often fail to rectify particularly complex reasoning errors. MAgICoRe is a new proposed framework which allows LLMs to identify these difficult reasoning traces and reassess them using a multi-agent loop.

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