Summary Bullets:
- Salesforce’s new AgentExchange is a marketplace for AI agents that are preconfigured to integrate seamlessly.
- Interoperability among agents and frameworks will be a key concern as organizations look to deploy multiple agents to complete more complex tasks.
Salesforce’s annual developer conference, TDX25, took place in San Francisco during the first week of March. As expected, AI played heavily in all conversations, with AI agents and Salesforce’s Agentforce platform taking a starring role. Similar to its approach with GenAI, Salesforce has been a thought leader when it comes to AI agents. Noteworthy announcements from Salesforce TDX25 included Agentforce 2dx (a suite of AI-powered tools to support building, testing and deploying AI agents), an Agentforce API (enabling customers to embed Agentforce across applications and workflows), partnerships to help scale deployment of AI agents, and customer testimonials and potential use cases.
As yet, AI agents are relatively new for most enterprises, and few organizations have developed roadmaps that incorporate the technology. Conversations with enterprises reveal that many companies have questions regarding use cases, techniques for helping employees embrace AI agents, strategies for avoiding vendor lock-in, agent interoperability, scaling AI agents, and evaluating their impact on costs and operational efficiency.
AI agents lend themselves well to Salesforce’s value proposition (as does GenAI). Tasks related to customer service, sales, and marketing are low hanging fruit for the use of the technology. Salesforce launched Agentforce, its agentic AI layer for the Salesforce Platform, at Dreamforce in September 2024. It claims that it had 5,000 customers signed up for Agentforce in Q4 2024. During TDX25 Salesforce announced Agentforce 2dx for the developer community. Agentforce 2dx is a suite of tools designed to help developers and admins configure, test, and deploy Agentforce, speeding up the build process. These tools benefit both pro code and low code developers.
By far, the most intriguing announcement from Salesforce TDX25 was the creation of AgentExchange, an ecosystem designed to help organizations scale the use of AI agents. AgentExchange is a marketplace for AI agents that includes pre-built actions and templates from over 200 partners, such as Box, Google Cloud, Workday, and Docusign. Salesforce already offers an enterprise AppExchange with over 7,000 partners and that is used by 91% of its customers. It expects to see similar success with AgentExchange as companies move from experimenting with one agent to incorporating multiple agents that interact with each other to complete complex tasks. The exchange addresses the issue of agent interoperability, one of the top challenges facing enterprises that want to deploy complex agents. Enterprises looking to develop AI agents that incorporate multiple tasks need multiple agents that work well together. Since the AI agents on AgentExchange are already configured to integrate seamlessly, the exchange paves the way for more robust AI agents that can complete more complex tasks independently, leading to greater automation, more widespread adoption, and additional use cases.
Support for AI agents is expected to evolve quickly through 2025. Platforms for building agents will offer additional tooling to assist with building agents as well as governing the relationships among AI agents. Furthermore, as the pool of large and small language models grow, enterprises will need a platform that helps them evaluate model results across vendors. (Enterprise can access multiple models with Salesforce, including models from Anthropic, AWS, and Google). Similarly, they will begin to explore models that offer the ability to dial up or dial down (or turn on or turn off) more expensive and higher latency chain of thought reasoning to balance computing costs with latency and performance requirements for specific use case. Interoperability among agents and frameworks will be a key concern as organizations look to deploy multiple agents to complete tasks.
Additionally, organizations will need to evaluate when to use ‘human in the loop’ (likely by identifying trigger points). And, as has been shown time and time again, the best technology is of no value if it isn’t embraced by employees. Enterprises are struggling to determine the best way to encourage adoption of the technology, a challenge that has been around for years, starting with the deployment of predictive AI solutions. Finally, the elephant in the room is the question of how AI agents will impact jobs. While many industry leaders communicate that the role of AI agents is to improve efficiency and free up employees so they can focus on higher value tasks, there is a large part of the workforce that sees them as potential threats to job security. The coming months will see a lively conversation on all of these topics, with customer service being an obvious place to start AI agent journeys. As a best practice, organizations should look to automate tasks that are repetitive and straightforward. And as was seen with generative AI, robust data management practices are a prerequisite for success with AI agents.