Diving Deep Into Agentforce 3 With Salesforce SVP Sanjna Parulekar

Agentforce 3.0 was officially launched on Monday, bringing some groundbreaking new features that will help optimize your overall agent performance and ultimately drive more people to adopt Agentforce into their workforce.

The early feedback from the release has been mostly positive. Many feel as though a lot of the pre-existing concerns around agent monitoring have been directly addressed, and the new Model Context Protocol (MCP) has really opened the door for companies to connect Agentforce with wider tools and workflows they already use.

But, as expected, some skepticism remains around the Agentforce product in general and whether the pace of innovation is outpacing customers’ ability to implement, govern, and extract real value from it. 

To discuss this further, I spoke to Sanjna Parulekar, SVP of Product Marketing at Salesforce, to find out the motivations behind introducing Agentforce 3.0, and why these new features should help improve the agentic experience for customers.

The Visibility Fix: Why Command Center Matters

Since Agentforce was released at Dreamforce last year, a lack of observability has been a major customer pain point. Many have wanted to understand how their agent is performing and acting, but haven’t had that real understanding, which has had a direct impact on Agentforce adoption rates.

The Agentforce Command Center is Salesforce’s answer to this. This new feature provides customers and users with key insights, such as health monitoring, use frequency, and interaction analysis, amongst other things – all to help you understand what’s working with your agent and what needs improving.

According to Sanjna, introducing this feature was all about taking on constructive feedback from customers, understanding their use cases, and building something concrete to work with.

Sanjna said: “The idea is that you now have full observability and governance across all of your Agentforce agents, down to the individual interaction level. A lot of the early feedback we’ve heard from customers is, ‘It’s great that we can deploy agents quickly with Agentforce, but we really want to take it to the next level and closely monitor how they’re performing.’

“For example, imagine you have an agent running on your website, and you want to know if a competitor is being mentioned on your help site. How does your product compare to another? With these new tools, you’re not just able to see that someone asked that question; you can replay the full back-and-forth and inspect exactly how the agent responded. That kind of visibility is incredibly powerful because it helps you continuously refine and improve how your agents operate.”

Another important aspect of understanding agents and how they’re performing is the return on investment (ROI). Agentforce is arguably quite expensive to implement, and many executives were deterred by the fact that there was no way to track what each agent was costing them. 

But now, you can observe how much each agent is costing and act accordingly – all within the Command Center.

Sanjna said: “Our Digital Wallet is integrated, so IT leaders can see how much each agent is costing the business. If, say, the marketing team’s agent is burning through credits for a low-priority use case, you have the governance controls to step in and say, ‘Hey, we need to optimize this,’ or even pause usage. It’s pretty robust tooling.”

Model Context Protocol and Interoperability

Described as the “USB-C for AI agents”, the MCP allows you to easily connect agents to external systems without custom code. You can take real actions across your tech stack while staying within enterprise security controls.

This feature in particular is a huge deal for many in the ecosystem. The opportunity to integrate Agentforce at scale while removing the heavy lifting of custom code is re-instilling trust in the product for customers, which has been a struggle in the last few months.

From Sanjna’s perspective, introducing MCP was all about being realistic about Salesforce’s customers. 

She said: “We’d all love to pretend that everyone’s data lives solely in Salesforce, but that’s just not realistic…the same goes for agents.

“Some agents will absolutely be Agentforce-native and built specifically for Salesforce workflows, but there are always going to be processes that live outside the platform. For example, if you want to process a transaction through a system like PayPal, that’s not something that happens within Salesforce, but with secure MCP client access, you can have agents work together across systems like that.

“That’s really what we’re optimizing for. From a Salesforce perspective, it comes back to our tech roots. [We] describe our vision as an ‘open but coherent’ platform. That means customers should be able to use different technologies, connect to any data, use any model, but we make it all meaningful to the customer experience and outcome.”

While many AI platforms promote their “openness” as a strength, Sanjna was quick to point out that not all openness is created equal. She drew a clear distinction between offering raw access and delivering real, usable integration, suggesting that Salesforce’s approach is less about giving customers tools for the sake of it, and more about ensuring those tools lead to actual outcomes.

She said: “There are so many platforms out there that say they’re open but are not coherent. They’re open in the sense of just raw APIs, and you end up with a DIY kind of disaster where you’re maintaining APIs with custom development. Whereas in Salesforce, we need to ensure that customers are able to connect to different systems and make it useful for them.”

AgentExchange: How Do I Choose the Right Agents?

AgentExchange’s introduction in March resonated with many across the ecosystem, tapping into the spirit of collaboration that has long defined the Salesforce community.

Fast forward to now, and this latest release shows how Salesforce is working to turn that promising concept into something with real business value, introducing features like plug-and-play MCP integrations, curated use cases from trusted partners like PayPal and Box, and a roadmap for measuring agent quality and performance. 

It’s a clear shift from early-stage vision to practical utility, giving customers more confidence to build with and invest in AI agents.

These new features are, of course, exciting – but how are Salesforce helping customers confidently choose the right agents from the marketplace?

When I asked Sanjna about this, she was very transparent about the fact that while these new features are a great next step, Salesforce still have some work to do to help their customers understand how to use it.

She said: “What’s interesting about this space is that it’s so present in our language and thinking, we sometimes assume it’s further along than it actually is. We’re all still trying to figure out what the right metrics are for defining a ‘great’ agent.”

Sanjna drew comparisons to the well-known AppExchange, and how they may take inspiration from their existing product to shape the AgentExchange experience.

“On AgentExchange, you can imagine something like a quality score or customer reviews, just like you’d see on the AppExchange. There are a lot of ways we can help surface that kind of information, and we’re actively exploring what really matters.

“Just like when customers browse AppExchange today, they look at reviews, they understand the use cases, and that’s how they make decisions. In the agent space, the signals might be different. So we will provide ways to evaluate agents, but we’re also learning from customers what kind of information they need to make more confident choices.”

It’s refreshing to hear how closely Salesforce are looking to work with the community to get this ready, and that they’re ready to admit it still needs some work to reach its enterprise-standard potential. 

The Adoption Gap: Technical Debt, Expectations and Best Practices

Despite all the new capabilities arriving with Agentforce 3.0, adoption is still being slowed by deeper structural challenges, such as technical debt, data readiness, and uncertainty around how to design and deploy AI agents.

Sanjna acknowledged these issues head-on, saying: “This is where the art and science of agents really come together. There’s a lot of tech out there that looks the same – everyone’s talking about agents – but Salesforce’s difference is on the ‘art’ side. We’re not just offering a platform; we’re providing best practices and close partnership when it comes to deploying agents.”

To help fill these knowledge gaps, Salesforce have launched a public best practices guide for Agentforce implementation, with more educational resources on the way.

Sanjna said: “Yes, tech debt is real, but there’s also a lack of understanding around how to actually architect an agent. Some customers believe they need perfect data to get value from an agent, but that’s not true. Of course, data quality matters, but it doesn’t have to be perfect.”

A recurring theme in our conversation was this aspect of realism over perfectionism. Salesforce has now seen thousands of Agentforce customers go live, and they’re using those lessons to show what “good enough” looks like, not just what’s ideal on paper.

On the product side, Sanjna also mentioned that Salesforce is building “low-lift” ways for customers to get started with Agentforce.

“If you want to connect an agent to a corpus of data, but you don’t have a perfectly harmonized Data Cloud setup, you can use our Agentforce Data Library,” Sanjna explained. “It’s a simple drop-down in Setup that lets you connect to structured or unstructured data and quickly get a first version of your agent up and running.”

This overall flexibility should make it easier for teams to experiment without needing a full-scale data transformation project up front.

Final Thoughts

This really positive and insightful conversation with Sanjna highlighted that while there is significant excitement around Agentforce 3.0, Salesforce is being realistic about the current maturity level of the technology. Openly acknowledging adoption hurdles like technical debt and data readiness, among other things, is what is going to drive trust with customers.

It’s also positive to see steps are being taken to address these concerns, with clearer governance tools, best practice guides, and flexible product features now also in their inventory.

Agentforce will continue to scale, but with 3.0, there now seems to be more in place to help customers get up to speed and better understand how to use agents effectively.