Agents are leading the next wave of artificial intelligence, and Salesforce Inc. sees its agentic AI platform as the natural culmination.

The company recently released a series of updates to Agentforce that enable its AI agents to be more proactive and autonomous, as well as an AI agent builder for low-code and professional or pro-code developers. A critical component of building agentic AI is giving agents the context they need to operate autonomously, according to Muralidhar Krishnaprasad, president and chief technology officer of Salesforce.

“One of the key things for your agents to succeed is not just having the data and the actions, but making sure you teach the agent what to do,” he explained. “This we do with every human person too. When you bring in anybody on board, you have a whole onboarding … we have created our agent platform to actually have all of that stuff. We call that topics, we call that instructions, we call that guardrails. That is part of the learning that we all have to do as a business.”

Krishnaprasad and Alice Steinglass, executive vice president and general manager, platform, at Salesforce, spoke with theCUBE’s Savannah Peterson and George Gilbert on the “Road to Intelligent Data Apps” podcast. They discussed the role of data and applications in Salesforce’s agentic AI platform.

Constructing a ‘multi-agent’ world

AI lives and dies on its data, and Salesforce supports its AI agents with DataCloud, which unifies all the data across the Salesforce platform. When it comes to AI agents, developers need to combine structured and unstructured data for the best results, Steinglass emphasized.

As an example, one of Salesforce’s customers, Heathrow Airport, has large amounts of both structured data, such as the layout of its restaurants, and unstructured data, such as the movement of people’s luggage. Salesforce achieved a 95% accuracy rate on AI results with Agentforce at Heathrow by streaming real-time data, according to Steinglass.

“What we want to do is give the business the flexibility to be able to update those policy guidelines and then have an agent that can reason through those in real time and take advantage of that while combining it with that structured deterministic [database],” she said. “The power of AI and agentic reasoning to bring that structure and unstructured data together and respond to the business in real time.”

Salesforce has continued to refine its agentic AI platform by putting a metadata lens over its database, followed by a security and governance layer, and a semantic model on top of that. The JSON Graph enables agents to give efficient responses, while the semantic model creates consistent KPI standards. The company also “agentified” its application programming interface management software, MuleSoft, allowing users to ask questions about their enterprises’ APIs and converse with them.

“You can get your mainframe to talk English … you can actually start conversing with it,” Krishnaprasad said. “That brings us to the next step, which is: OK, now if you can start conversing with it as a human, you can have other agents also converse with it. We are slowly then leading that revolution onto agents now talking to agents, which we call multi-agent behavior.”

An agentic AI platform that can experiment

Part of Salesforce’s goal in building agentic AI is not just to improve individual agents, but to make the whole system smarter by turning insights into deterministic, repeatable rules. Steinglass compares the process to following a recipe and making tweaks in real-time.

“We want to enable ourselves to be able to say, ‘Hey, I just came home from work and I’m going to figure out what I’m going to make based on what’s based on the ingredients that happened to be in my fridge today,’” she said. “That’s what AI allows us to do. AI allows us to take these different deterministic capabilities, these APIs, these functionality and compose them dynamically at runtime to solve the problem that’s in front of us.”

As agents grow smarter, AI has the potential to harmonize APIs across different platforms, as well as tackle problems that have yet to be answered with deterministic coding. Even though multiple companies have invested in agentic AI, Krishnaprasad believes that Salesforce’s strategy will prove superior.

“A year from now, I think you will look back and say it wasn’t about who came first, it was about who got it right,” he said. “With our comprehensive platform, you might have seen that hemisphere slide and we are seeing a lot of our competitors just copy it … but basically that slide is real. For any enterprise agentic to succeed, you need that. You need all those ingredients. We believe we are at the forefront of getting all of that together into an integrated platform.”

Here’s theCUBE’s complete video interview:

Image: SiliconANGLE