As more vendors rebrand automation as agents, established giants and startups alike are challenged to show agentic AI delivers real business value — not just hype.
Among the most ambitious is Salesforce, which aims to do for AI what it did for enterprise software.
Last week, Salesforce highlighted growth for its Data Cloud and Agentforce platforms, reporting $900 million in annual recurring revenue for fiscal 2025 and 3,000 net-new Agentforce deals in the fourth quarter. The company is pitching these tools as the foundation tapping into what it sees as a potential $6 trillion “digital labor market” — which comprises AI agents powering marketing, sales, and service workflows.
Agentforce and Data Cloud are the platforms driving Salesforce’s agentic AI strategy. Agentforce, which debuted last fall, enables companies to deploy autonomous AI agents. Meanwhile, Data Cloud, its customer data platform, first debuted in 2020 as Customer 360 Audiences before evolving through several rebrands before becoming Data Cloud in 2023.
Companies like CaixaBank, Maserati, OpenTable, and Pfizer are already using Agentforce and Data Cloud to power agents with CRM and other data. Other examples include Equinox and Wyndham Hotels, which are leveraging the platforms for real-time, personalized marketing and customer experiences across global markets.
The challenge for companies like Salesforce and others is to show how new agentic workflows actually differ from past forms of automation. The old model was driven by bots and deterministic rule-based systems merely pre-wired to take specific actions, said Rahul Auradkar, Salesforce’s GM and EVP of Unified Data Services.
“The new world of AI is rewriting those rules, and the re-writing of those rules is determined at the moment based on the information you have provided and you’ve grounded it with,” Auradkar Told Digdayay.
A key factor for Salesforce’s agentic strategy is its cloud and agent architecture. Auradkar said customers can build agents using their existing data infrastructure investments along with customers’ preferred large language models and data sources. Salesforce is also adding bi-directional integrations with Databricks (in beta) and Snowflake (coming this summer) to support AI-driven personalization.
Diverse, grounded data is essential for AI agents’ overall effectiveness. It’s also key to reducing inaccuracies — known as “hallucinations” when AI models generate confident but incorrect or fabricated responses.
Salesforce’s Data Cloud is now powered by more than 270 data connectors, including its Zero Copy tool, which enables near-real-time activation by linking to original data sources — such as data lakes, warehouses, CRM systems and other sources — without copying or moving them.
Data Cloud has had massive growth, according to Salesforce. In fiscal 2025, Data Cloud processed 50 trillion records — double the previous year — with 65% from external data sources.
“If an agent powered by AI is trying to give or make recommendations or take actions on a customer’s behalf, it had better be accurate,” Auradkar said. “…For example, if the system is coming and telling you about an action it would perform, as opposed to what would [previously] already be wired in. It’s the most optimized action you would perform, but for that optimized action to be accurate, you need to have the right data.”
Making agents easy to use creates a unique tension, said Gartner analyst Andrew Frank. Ease of use makes adoption less daunting, but it also makes it harder to prove value when a lot of the complicated back-end engineering is behind the scenes.
“Just because capabilities seem incredibly powerful doesn’t mean they’re necessarily going to create value or improve customer experience or any of the things users are actually trying to do,” Frank said. “It’s easy to get trapped into the gee whiz feeling of automating this stuff.”
There’s also the risk-reward trade-off.
“The proactive agents can now act on changes in data so you can initiate processes without any human being being involved,” Frank said. “That sounds really cool, but when you think about if it really creates value versus when it could create circumstances I can’t even anticipate because I don’t know what the data is going to do.”
Salesforce has been successful with branding agents and Agentforce just like Microsoft has been able to own the brand of Copilots, said IDC analyst Roger Beharty. Although category growth being slower than market expected, he noted Salesforce stands out for its overall strategy for Data Cloud, especially given the importance of clean, governed, managed data sets that are centralized.
There’s also the question of what clients are actually willing to pay for AI, versus what they say they’re willing to pay, what vendors hope to charge and what the market expects in terms of returns. Beharty also thinks personalization might still be oversold across the sector, while the sprawling array of agentic options can feel overwhelming. (Just like it did with the earlier generative AI hype cycle back in 2023.)
“It makes for really savvy keynote presentations when you can wave the magic wand and then the apple becomes a grape and the grape becomes a peach and becomes black,” Beherty said. “Like that’s a great presentation, but now you flip that around and does that have an ROI? Will that impact sales?”
Salesforce isn’t the only company looking to power agents using a range of datasets. Other examples include Hubspot, which earlier this month expanded its AI agent options to more than 200. Another is Twilio, which is building AI agents through its open ConversationRelay platform that combines real-time data, speech recognition, and LLMs.
Using contextual data also requires an interoperable tech stack, said Andy O’Dower, Twilio’s vp of product for voice and video. He also noted Twilio’s approach emphasizes interoperability over prepackaged tools, with flexible integrations instead of silos: “Your AI agents are only as powerful as their data, and that data needs to be real-time and contextual.”
“This should be an aggregate of a person’s behavioral data, historical records, and the insights gleaned from past conversations,” O’Dower said. “The latter is especially important: each interaction with a customer should be a learning opportunity that helps refine and iterate on the AI agent.”
https://digiday.com/?p=576086
