Salesforce today announced Agentforce 2dx, a significant upgrade to its AI agent platform. This new version enables AI agents to be proactively deployed within any business process. Agentforce has gained several new development tools, and the AppExchange App Store has been expanded to include AgentExchange for sharing ready-made agents.
The announcement of Agentforce 2dx follows just a few months after the introduction of Agentforce 2.0 and shows that Salesforce is keeping pace. Where many companies are still struggling to implement basic AI functionality, Salesforce is now taking the next step by allowing AI agents to operate fully autonomously within business processes.
From reactive to proactive
With Agentforce 2dx, Salesforce significantly expands the platform’s capabilities. Whereas previous versions of Agentforce were primarily responsive to user interactions, AI agents can now act autonomously based on data and business process changes. This makes deploying agents for more complex tasks possible without direct human intervention.
AI support when building agents
Developers and users building agents can now get help from an AI. Creating topics and actions can be done with AI support. This ensures that the instructions are logically structured and that the AI can handle them well. The AI also provides support in improving agents and the answers an agent gives.
Agentforce 2dx is for developers
While proactive agents and support for AI will appeal to many business managers, we believe this update is primarily one for developers. This is because the options for building and deploying agents are greatly expanded. It will be even easier for developers to get started with Agentforce. A summary of what’s coming:
Agentforce APIs:
Agentforce APIs allow developers to integrate agents into the backend systems of other solutions. If a process is kicked off in your SAP ERP, it can trigger an Agentforce agent via an API. This agent can then collect data and set up a production and ordering process, for example.
Calling Agentforce actions:
Each Agentforce agent can perform several actions. Often, the agent decides which action to perform based on the information coming in. Now it is also possible to execute a specific action directly within Salesforce Flow and Apex. This allows agents to be integrated within Salesforce’s existing workflow tools.
Mulesoft for Agentforce:
Basically, MuleSoft is an iPaaS solution. With it, you establish API connections with hundreds of applications so that they can work together. MuleSoft for Agentforce now offers the ability to create Actions and Topics for an agent based on the APIs in the MuleSoft catalogue. This allows an agent to suddenly enrol projects in Asana or Monday or send messages via Microsoft Teams or WebEx.
Agents can also be integrated within existing MuleSoft workflows. MuleSoft’s big advantage for Agentforce is that the observability offered by MuleSoft also applies to agents using MuleSoft’s APIs.
Agentforce in Slack Workflow builder:
Of course, Slack cannot be left out of this list. In the no-code Slack Workflow builder, Agents can now be added as an intermediate step. So, a Slack event can invoke an agent. For example, it can send call content back to Agentforce to add that to a customer record.
Agentforce Surfaces, Cards and Tableau integration:
Agentforce can now respond much more dynamically than just text. It can also use images in a response and will soon be able to show a video. However, it can also show a window with buttons, such as choices for dates and times to schedule an appointment. The user can easily schedule an appointment with someone at the touch of a button. There is also integration with Tableau so that any responses with lots of data can be visible in tables and graphs.
What such a response looks like depends on the data and the platform on which it is communicated. For example, a chatbot on the website may offer more possibilities than Slack or WhatsApp.
Testing center:
Salesforce has developed a Testing Center. If you are convinced that your new AI agents are ready to be deployed. You can visit the Testing Center to generate 10, 100 or even 1,000 interactions that the AI agents should start processing in a secure Salesforce sandbox. Then, you can evaluate the output to see if the agent is performing as desired.
This follows previous improvements in managing and monitoring AI agents. Although agents can now act autonomously, organizations still need to keep an overview of what the agents are doing. Salesforce takes this one step further with DX Inspector.
DX Inspector
With DX Inspector, an agent interaction can be thoroughly dissected. The DX Inspector provides insight into which data and metadata were used, which topics and actions were used, and how the agent arrived at its result. Especially in combination with Testing Center, this is a great way to test and optimize agents so they perform better.
Agentforce Developer Edition and CLI and VSCode support.
Developers and Salesforce partners can now register for a new free Agentforce Developer Edition, in which they can experiment with Agentforce and Data Cloud. Salesforce limits the environment to 10GB of storage capacity and 150 LLM runs per hour.
Finally, developers can now create, edit, and delete Agentforce configurations using CLI and VSCode. This allows them to work in the environment they are most comfortable with and is not tied to Salesforce’s graphical user portal.
Finally, with the introduction of AgentExchange, developers will soon be able to share or sell agents and actions with each other. At the launch of this new marketplace, more than 200 partners are already connected and offering agents. Each agent undergoes a rigorous security review before it becomes available in the marketplace.
Customer experiences
The impact of Agentforce is already visible among early adopters as well. OpenTable, one of the first users, reports that their AI agent now handles 73% of all restaurant-related Web queries. This is a 50% improvement over their previous solution. Salesforce itself has converted help.salesforce.com with Agentforce and only 3% of queries still end up with a human agent. Engine, a business travel agency, states that emails are now answered 15 per cent faster on average, and AI agents pick up 30 per cent of all emails. Precina, a U.S. diabetes care provider, says Agentforce is helping to automate many recurring customer contact moments. Patients now have an AI coach they can ask questions, the AI agent can answer many of these patient questions. This leaves employees with less administrative work and more time for actual patient care. In addition, Precina states that it expects to save about $80,000 per year per 5,000 patients.
The added value and profit that Agentforce delivers still clearly varies by organization. It’s a learning curve, but since organizations can now be more proactive with agents in Agentforce 2dx, this should further increase the benefits.
Availability and roadmap
Agentforce 2dx is being rolled out in phases, with the full release becoming available in April 2025. Some components, including AgentExchange and the Salesforce Developer Edition, are available immediately. The AI-enabled Agent Builder will follow in March 2025, followed by enhancements to the Testing Center in April.
With Agentforce 2dx, organizations can deploy AI agents for virtually any business process, regardless of whether they use other Salesforce products. Techzine is attending Salesforce TrailblazerDX this week, Salesforce’s developer conference, and will continue to report extensively on these new innovations.