3 Salesforce Leaders on Why Context Is King in the Agentic Era

Data is the backbone of enterprise software - without a strong data foundation, businesses can't fully leverage their platforms. With the advent of AI agents, data isn't enough by itself: If agents don't have an understanding of how all the data fits together, they can make errors or hallucinate.

That's why, according to three Salesforce executives, the key to unlocking the next phase of agentic evolution is in context - the harmonization of data that transforms it into meaningful, trusted, and actionable intelligence to allow for a single, shared view of the business. Think of context as AI's working memory with institutional knowledge that enables agents to act reliably - without context, agents can't understand the meaning behind all of the data and may fail to take appropriate action.

Imagine you call your customer contact center or chat with an AI agent. They know everything about you and they're able to proactively fix the issue versus you being handed off to various service reps or AI agents and having to repeat every single thing 10 times. That is the difference in context.

Rahul Auradkar, President of Data Foundations at Salesforce, defines context as the bridge that connects disparate data to a specific action. Without this link, he argues, data is just "simple text floating around" - lacking the guardrails and interaction necessary for decision-making.

As Muralidhar Krishnaprasad, President and CTO of Engineering at Salesforce, put it: "Imagine you call your customer contact center or chat with an AI agent. They know everything about you and they're able to proactively fix the issue versus you being handed off to various service reps or AI agents and having to repeat every single thing 10 times. That is the difference in context."

Context Is the 'New Currency'

While models are able to complete actions with human prompts like generating text, agents go a step further and act autonomously. Context acts as the long-term memory for agentic AI, putting the varied puzzle pieces together so these "hands-free" agents accomplish their tasks effectively. In the Agentic Enterprise, where humans and agents work together, context has become the "new currency," according to Auradkar. Unlocking this value is what enables a business to become a successful Agentic Enterprise with agents that perform reliably.

Context is therefore essential - scattered information alone can't drive action. For example, with Salesforce's Agentforce, trusted context comes to life through a knowledge base powered by Data 360, MuleSoft, and Informatica. "To bring that currency to life, you need Data 360 to unlock trapped data and provide real-time actions with a unified profile," Auradkar explained. "Then you need the enterprisewide catalog that needs to be unified along with customer profiles, which is what Informatica brings. And then MuleSoft allows you to drive the connectivity across all applications and also does the last-mile integration into third-party applications as well."

Architectural Foundations (Iron Man vs. Wolverine)

In order to access the benefits of this rich context, it's important for companies to take a holistic, structural-level approach. Agents should be integrated into the architectural foundation of companies' platforms, not just thrown on as mere add-ons or treated as isolated experiments. Integrating a secure context engine into the foundation allows for the seamless translation of raw data into actionable intelligence and also allows companies to scale agents more quickly.

Shibani Ahuja, SVP of Marketing for Enterprise IT Strategy at Salesforce, noted that "models are smart things, but they need a body, a platform that provides the channel, the shell, and the rails. As you think about the evolution of moving from a generative to agentic use case, the architecture necessary for that is very, very different. Generative to me is like Iron Man's arm - it can be replaced. Agentic is like Wolverine's Adamantium in that you can't take all of that architecture out and re-architect it." While generative AI can be added on to existing platforms as a tool, agentic AI must be "built into the bones" of the data architecture to function safely and autonomously.

Implementing agentic AI at the architectural level also allows companies to scale quickly. Ahuja added, "CIOs that are thinking about the broader architecture that's necessary for a virtual agent/colleague to traverse the entire halls of an organization will have greater success in being able to scale some of the capabilities that they're building." Or, as Auradkar explained, "Once you have the platform foundations right in place, things become much faster."

The 'Stalling' Myth vs. 'Slowing Down to Go Fast'

One criticism of agents is that their development is stalling, with companies often ending up in "pilot purgatory," where technology gets stuck in testing phases. Companies are moving past the "sugar rush" phase of AI demos into the hard work of building trusted foundations. Instead of stalling, Auradkar contends that companies have "slowed down to go fast" by taking a more deliberate approach to agentic AI implementation. It's important to invest up front in context so agents can effectively perform, rather than produce early but unreliable outputs that are devoid of a comprehensive understanding of the business.

Your job is not done when the agent is launched. Your job starts when the agent is launched. You need to observe, fix, iterate, so that the agent is fresh.

In reality, agents aren't stalling but rather maturing through an iterative process of trial and error paired with ongoing product development, as teams learn how to provide the right data, guardrails, and contextual grounding for real-world use. As Krishnaprasad stated, "Your job is not done when the agent is launched. Your job starts when the agent is launched. You need to observe, fix, iterate, so that the agent is fresh."

Our "Agentic Maturity Model" necessitates "a bias for action focused on specific use cases," according to Ahuja. She continued:"Find the biggest, most boring, Level 1 use case. Start to explore what would need to be true to implement that. What are the minimum data attributes that you need to harmonize and cleanse?" When businesses commit to this intentional foundational work, the definition of success shifts from mere outputs - deploying AI for the sake of production - to measurable outcomes that deliver tangible ROI. This approach allows organizations to right-size their data readiness and tech investment, incrementally strengthening their stack to meet specific goals rather than attempting to harmonize all enterprise data up front before seeing value.

The New KPIS: Adaptability, Agility, and Trust

To move beyond "pilot purgatory," organizations must stop chasing isolated production use cases and start building for long-term outcomes. Instead of deploying one agent, the goal becomes building a system that can adapt to future changes and scale from the first agent to the hundredth. Agents need to be reliable decision makers and action-takers, and this can't happen without a unified platform and trusted context.

Salesforce leaders emphasized adaptability, agility, and trust as key measures of success in the agentic era. For Ahuja, adaptability will be key: "We have all heard of IQ and EQ, but I predict AQ will play a greater role - the adaptability quotient of leadership mindsets and of organizational tech stacks. How adaptable, how interoperable, how composable is your tech stack today?" Additionally, Auradkar highlighted how scaling fast will be key: "It's not your first agent or your second agent being successful. How can you, in a scaled way, have an impact across your business? Once you have the platform foundations right in place, things become much faster. To do that, you need agility."

As for trust, Krishnaprasad predicts it will become even more important as agents continue to evolve: "You've got to make sure, out of the box, that the business reputation is protected. This is why context is really important for an agent: for it to solve the problem, solve it the right way, and also for us to make sure that it's giving you the right ROI."

Go deeper: