Cloud company Salesforce is moving fully into what it calls the “Agentic Enterprise,” a model where artificial intelligence (AI) does not just assist workers but can take actions, work alongside people, and help run everyday business tasks. The company sees this shift as a major change in how organizations operate, not just another technology upgrade.
A few months after introducing its Agentforce platform in the Philippines, Salesforce said businesses are starting to look beyond simple AI tools and toward systems that can work with employees in real time.
“AI assistants are small tools we added on, but agentic systems are different,” said Gavin Barfield, VP and CTO, Solutions, ASEAN at Salesforce, during a media briefing at Salesforce Innovation Day Manila. “Agentic systems involve AI and humans working together, hand in hand. It is about AI that can take action and actually do things.”
The Agentic Enterprise idea is described as the “third wave” of AI inside companies. The first wave focused on predictive AI, which used past data to forecast what might happen next. The second wave introduced copilots, such as chatbots and digital helpers that guide workers but do not change how work is done. The third wave involves AI agents that can act on reliable company data and help complete tasks automatically.
Salesforce said these agents can process customer inquiries, update records, help sales teams prioritize leads, or connect information across systems.
According to Barfield, turning into an Agentic Enterprise is not easy. Companies must have strong data foundations or risk getting stuck in early test phases.
“Organizations must have high-quality data, strong governance, and seamless integration to avoid stagnation in trial phases,” he said.
He warned that the “agentic divide,” or the gap between ambition and real implementation, is growing as many companies try AI but fail to put it into their everyday operations.
Salesforce emphasized that poor or untrusted data weakens any AI system. Barfield said AI tools must be built into company platforms instead of being stand-alone add-ons. Effective governance is also needed so that teams know where data is coming from, how it is being used, and how AI decisions are monitored.
He added that modern businesses store information in many places (documents, emails, chat messages, and internal systems) so AI must be able to work with both structured and unstructured data. This requires a flexible technical setup that can support communication between multiple AI agents and connect to external systems when needed.
Inference
Barfield said LLMs should not be relied on for factual information. Instead, they should only be used for reasoning and generating natural language. The facts themselves must always come from trusted enterprise data.
“You see toxicity, hallucinations, and bias incorporated into the response,” he said. “The way we work is through grounding those problems with trusted data.”
Grounding means connecting the AI model to the company’s validated internal databases so any answer it gives is based on correct information. Salesforce uses tools such as Data Cloud to check data quality, label data sources, and ensure the system selects accurate information before AI produces an output. This reduces the risk of the AI generating answers that look correct but are inaccurate.
Speed also plays a role in inference. Barfield said earlier AI systems often struggled with slow response times, which made them less useful in real customer interactions. Salesforce’s Agentforce Voice was engineered specifically to keep latency low, allowing AI agents to respond almost instantly.
“We didn’t want to roll out a product that had latency issues where customers were waiting on the phone for responses,” he said.
He added that recent engineering improvements brought response times down enough to make conversations feel more natural.
Salesforce said companies aiming to become Agentic Enterprises must treat LLMs as reasoning engines while keeping factual answers tied strictly to internal data. Strong data governance, reliable technical foundations, and tools that maintain accuracy and speed are essential to building trust in AI-driven decisions.