Beyond AI Models: Data Platform Requirements for Agentic AI
Two years ago, ChatGPT and large language models (LLMs) dominated the headlines. But the conversation has advanced from the creation of generative AI models to how agentic AI can utilize them to take specific actions with minimal human supervision and create real business value.
The benefits of agentic AI systems include being able to perform multistep problem-solving tasks, generate nuanced and context-dependent responses to queries, and learn from experience to improve accuracy and customer satisfaction. According to PwC’s AI Agent Survey, 79% of senior executives say AI agents have already been adopted in their companies. Two-thirds note that their agents are already delivering measurable value through increased productivity. As a result, nine out of 10 said that agentic AI has convinced them to raise AI spending in the next 12 months.
But the success of agentic AI implementation is dependent on data. Without a firm foundation in the form of a comprehensive data platform, organizations are hesitant to trust the conclusions and actions of AI and often question the reliability of its outputs. Hence, they fail to attain the goals of their AI deployments.
Bringing Structured and Unstructured Data to Agentic AI
Most organizations do a good job of providing AI applications and agents with access to structured data in databases, spreadsheets, web forms and other systems where everything exists in a standardized format. But that applies to only a fraction of the total knowledge base of any business.
More than 80% of all enterprise data is unstructured: social media posts, PDFs, manuals, onboarding docs, FAQs, charts and tables, customer interaction records, emails, and multimodal data like audio and video. Without integrating this treasure trove of enterprise know-how, agentic AI implementations go astray. They lack the trusted, real-time data foundation that is the essence of agentic AI performance.
Organizations, then, must advance from the stage of storing unstructured data in data lakes or trapping it in information silos. What is needed is a tailored platform that provides unified and harmonized data; confidence and trust in AI outcomes; real-time and scalable data foundations; and seamless integration for coexistence with existing enterprise infrastructure.
Unified and Harmonized Data With the Power of Metadata
AI agents need access to diverse data sources. This must include all areas or forms of data, including structured and unstructured data. By unlocking the many silos that contain enterprise knowledge, agents can glean context that was never apparent before. Only by being able to pool data from policies, contracts and sales information with the power of metadata can you create a unified, harmonized and trusted data profile.
Take the case of an AI agent aiding a human in a decision related to raising a credit limit. If the agent has access to a limited span of data, an incorrect decision is likely — either denying credit to a valuable customer or extending credit to a high-risk individual. However, when the agent can access policy, all relevant customer information, the person’s credit history and credit rating, and their history of sales and marketing contacts, the chances of an agentic AI error diminish sharply.
Most customer service agents have access only to passive data. Agentic AI seeks to take this to the next level by shifting from passive data to an active ecosystem of data that adds to customer context in real time.
Confidence and Trust in AI Outcomes
As well as gathering all available enterprise context, the autonomous actions taken by agents must be trustworthy. They must respect all privacy and governance policies and be based on access to complete data. Trust comprises three distinct aspects:
- Accuracy and relevance to eliminate errors and hallucinations.
- Security and privacy concerns, such as masking sensitive data like social security numbers in real time through automated tagging, classification and policy enforcement.
- Data lineage/provenance by tracking origin and changes for compliance to GDPR and other regulatory frameworks.
Trust becomes all the more essential as AI systems scale. If agents are to act autonomously across millions of transactions, the repercussions of mistakes are enormous. Agentic AI raises the bar on trustworthiness. Done correctly and based on the right data platform, the organization gains confidence in the actions taken by autonomous agents.
Real-time and Scalable Data Foundation
Context should extend to “agentic memory” i.e., real-time contextual awareness within the AI agents themselves, so they remember previous interactions and respond relevantly. This is crucial for both system connectivity and ongoing human–agent interactions.
Harmonization, unification and trustworthiness cannot be done slowly or for a small number of transactions. They must be supported by speed, in real time and at massive scale. Whether an organization has 1,000 or more than 100 million customers, the data foundation must be robust and scalable enough to support agentic AI activity at the moment humans supported by agents deal with customer requests.
The data platform should seamlessly power both insights and actions with data fluidity, distinguishing it from passive storage inside a storage array or data lake. A real-time and scalable foundation means data is available at the “right time for the right need,” ranging from sub-second real-time needs for personalized web experiences to within-second delivery for customer service agents. Multiply this for high-frequency interactions, and the scalability of the underlying data platform becomes imperative.
Seamless Integration With Enterprise Infrastructure
System design must enable agents to easily access and leverage both short-term context and long-term knowledge across a wide range of tools and systems. Any and all enterprise applications must be able to connect effortlessly with the underlying data platform. Due to the sensitive nature of some data, though, a zero-copy framework might be required to respect privacy and protect trade secrets. In this way, agents are permitted data access without any movement or compromise of the organization’s data.
This necessitates a data platform that has connectors to all enterprise applications and systems. Hence, many organizations struggle when they try to build such a platform internally. They often underestimate the level of complexity and the amount of time required. As a result, it tends to divert them from their core business.
Trustworthy, Empowered AI Agents
The rise of agentic AI demands a data platform that goes beyond storage to actively empower autonomous AI agents. It must be able to supply them with intelligent, contextual data from a great many data sources (whether structured or unstructured) in real time and at massive scale. The data being used and the actions taken by AI agents must be unified, harmonized, trustworthy and accurate based on automated data governance, security and ethical guardrails.
Salesforce Data Cloud meets all of these criteria. It’s the intelligent activation layer of the Salesforce Platform, coexisting with any existing IT infrastructure to drive agentic transformations. For more information, visit Salesforce Data Cloud.
