How to Use Agentforce with MuleSoft: Architecture Guide

Good morning everyone. How do you give an agent power to take actions over hundreds of disconnected systems? Today we will walk you through the use case we have built within Salesforce. How Salesforce fuels Agent Force with mulesoft so I'll give you a brief introduction before I started. I'm Shravan Kumar Vajrapu, Senior Engineering Manager at Salesforce leading enterprise APIs and Mulesoft integrations. Moving on to this we are the customer 0 if you look at the big picture agent360 agentforce360 platform this is our vision completely for unified agentic enterprise where AI agent connect every part of our business from sales and service to operations and hr. We are the customer zero for Mulesoft. We within Salesforce are the Mulesoft Enterprise team responsible to evaluate all the products that Mulesoft releases. We closely work with the Mulesoft product team, evaluate and then test all the products before their launch so we get the ability and chance to test all the features first. We implement within our own real world business use cases. We solve all the complex problems using our own products and we provide that feedback early in the game. So this brings a lot of confidence to the customers that Salesforce is actually leveraging their own products within their own ecosystem before rolling out to various customers customers. So the customer success is all truly embedded within our own team. We truly pioneer all the products and be at the forefront of this revolution and innovation. Now how we are scaling up to the agenting enterprise is the question. So we have heard lot about the digital transformation. For years we have been procuring lot of software, we are bringing in data and we are storing in data across various places in the name of digital transformation. But that is all great. But now how do we use all that data to put into an action? Going forward the humans will be closely working with agents and then agents empower humans to boost their productivity so that business can focus more on the innovation and building products for the future rather than focusing more on the day to day activities or the redundant mundane activities. So given all this, our role as Mulesoft team is vital in all this. Agentic transformation or the agentic enterprise transformation and building APIs are the true pillars to take any action within an enterprise. Even though the data is all available, agents can only provide information, but the action part is always in. Any legacy system or any databases or any systems which are sitting outside of Salesforce is only possible through Mulesoft. Mulesoft can take actions anywhere and APIs play a key role in taking those actions. Today, in every enterprise that is using Mulesoft, it is already composable in nature. We have APIs that are built in and they are reusable. Now agents can reuse the same assets. The next part is the MCP servers. APIs can act or easily be converted into MCP servers that are useful for agents. The next is the agent fabric. We are evaluating the mulesoft agent fabric which is recently launched that is used to govern all the agents that are spread across your enterprise. You can orchestrate, visualize and also use it in every life cycle of the agent journey. So this creates our roadmap for composable enterprise to the agentic enterprise. So data plays a key role. It's the foundation of everything. Any business process relies on the accuracy of the data. We have lot of regular audits and checks and balances. We do have a lot of automated reprocessing and monitoring in place. But still, if the agents have to be successful, the data accuracy plays a key part. Now I would like to walk through our own use case where we have brought in Data360 and Mulesoft and Agentforce to work together to solve our own problems. So this is our employee portal. So this is my profile within our own Salesforce sales system where my information has all the details. So if you look at my address, so it is showing the address as New York. So recently I'm a sales representative. My information within our own sales system is listed as New York. But I have recently moved out to Oregon. I have been classified as a remote employee. Now look at the impact of this. The information is showing as New York as a sales employee, all my territory information, all the sales related quota or acv, all that information is tagged to the east coast. But I am based out of west coast now. So even though the information is minute for us to get it corrected, but the impact is high. Now I'm going to showcase in real time manner like as an employee within Salesforce, I go to our own portal called Basecamp. On the Basecamp portal. So when an employee comes before the agentic world, we used to log a case using the log a ticket option. And the employees have to wait three to five days minimum to get their information corrected. So I'm pretty much sure everyone in this place might have gone through a similar approach within your own enterprise or within your own company. So you log a ticket, wait for some time and sometimes you get a response back like oh, I'm not responsible for this case. Go to a different department, get it corrected. So this is a quite common use case in every enterprise. So I'm going to do a live demo now. So now my agent has initiated. So now if you look at behind the scenes, the agent is doing lot of work for me. The agent is able to fetch my employee number from the session and now it is going to take the action immediately, right away. Now here is the magic. The agent has quickly reprocessed my record. It has given me the transaction number and it shows me the status as well in a real time manner. Now if I go back to my profile and refresh it. Wow. My information is now accurately mapped as Oregon. So this is the real magic of agent, how it is impacting the real world. Now the employee is is not no longer required to log a case. They are no longer required to wait for three to five days. The customer experience is now growing because the action is in real time. It's not a chatbot, it is taking action because without mulesoft it would have just responded with the information that your information is inaccurate. But the action part is the key here. Mulesoft has taken the right action correcting the data. How did we architect this solution? And even though the agent worked very smooth, the architecture was required, well thought. And I would like to go through a step back and provide you the composable architecture that was there prior to AgentForce. So if you look at it, the employee information within our own company is not in one system. Their information is distributed across various sources. So now how we resolved is by bringing the agent in the current state. We have replaced all that operations work with our own agent. That is again with the magic of mulesoft. So what we have done is we have aggregated all the reprocessing endpoints into the reprocessing API, which is now the intelligent orchestrator. So it can take any actions connecting to any system within the mulesoft layer. So the action support part is now being done by mulesoft and the reprocessing API is now being called by the agent. So anytime the agent invokes the reprocessing API and then it takes the action, this is the mulesoft part of it. Now I would like to go through the in depth architecture of this. So this is the overall architecture. Now on the left side the user is getting all the information from basecamp. So in basecamp I am typing my request. Now agent has two things to do. One part of it is it has to understand the context from the request it is receiving and it has to take action. So it is twofold because I just showed you the happy path. Because in real time users can type in anything and the agent should have right information to understand it properly. So for that what we have done is like we have ingested all the agent enterprise information into our data360. It has all the structured data, unstructured data, it has all the requirement documents, it's mapping documents how the data is brought across various systems. So that knowledge is now available in the data 360. It is all being done using MuleSoft direct another MuleSoft out of the box connector that is being provided by MuleSoft. So we are ingesting all the data into Data360 within that. So once the data is available, Agent is able to get the context. So now in our previous demo we talked about arg62. How will the agent know what that is? That is all because in Data360 it has information, it understands the context, it understands which object is being mapped and which fields are being retrieved. Once the context is all retrieved, now the agent has to perform the action. So now the API catalog and topic center finds the right tool and the right tool is it finds the reprocessing API that we were talking about in the previous slide and it is able to take action. So now we are aggregating the API LED connectivity approach with the Agentic approach within few simple integrations, all within the Salesforce platform. Why this architecture works is we are able to get the right secure data, reliable data within our own enterprise. We know it all and we are doing it all with mulesoft and it is all within Salesforce governed platforms where security, governance, all the features are embedded within the platform. So this architecture will now refer as the futuristic blueprint. We will be able to scale well. We can add more features now. Coming on to the next. Now this is how the agent is looking like the employee sync agent receives the prompt, the natural language prompt, hey, correct my location information. In Salesforce profile, it is getting the context from Data360, it is taking actions from Salesforce flow, it is retrieving the enterprise APIs and it is able to connect with all the systems and take actions. Is it not amazing? We had all the reprocessing APIs in the past. We had all the information in data Cloud. We just aggregated and then stitched everything together to work well for the agentec enterprise. Now there are three key takeaways from this architecture. Now our end users are very happy with this customer experience. They are no longer waiting for the actions to be taken. The second part the business is very happy because they are now no longer receiving any escalations at all. The third part is we are all secure because of the capabilities that MuleSoft, AgentForce and Data Cloud are always bringing in within the Salesforce ecosystem. This is this architecture not only works for one part, it also works for several other use cases within your own organization. You can apply the same thing for any of your customer resolution activities. So our Customer Zero journey is still exciting and it is going to be more fruitful in experimenting and resolving all these enterprise use cases with this. I would like to thank each one of you for listening to this and please do provide your feedback.