Salesforce's ambitions for Agentforce platform come to light

Not just for Salesforce users

During Dreamforce, which begins Tuesday, Salesforce typically releases its own products to legendary fanfare. Agentforce is no different, as the user faithful -- developers, admins and consultants -- will come primed to cheer on the demos and keynote hoopla.

But in some not-so-subtle language, Salesforce makes clear that Agentforce can be a platform on which automations can be built for most all enterprise cloud applications and services -- not just Salesforce's. The company's Agentforce Partner Network -- kicked off with a Workday partnership announced in late July, and expected to be up and running by year's end -- commingles Salesforce data with other apps' data in a zero-copy, access-controlled fashion.

"From a security standpoint, the sharing model is defined at the application layer, so it's only at the application layer that you [define, for example,] which sales reps have access to which customer records," said Clara Shih, CEO of Salesforce AI. "And so that's why we're seeing some of these security issues with [Microsoft] Copilot and from some of these other competitors. You just can't take that risk of data leaking into the model or giving employees and customers access to data that they shouldn't have."

Salesforce's Einstein Trust Layer purports to maintain protection of a user's customer data inside and outside the Agentforce platform. According to Rahul Auradkar, executive vice president and general manager of Data Cloud at the vendor, that is how the company plans to persuade non-Salesforce users to try the no-code Agentforce AI automation platform. Since privacy and compliance policies are built into metadata, they can transfer through data stores into Agentforce intact.

I think what we want to focus on with new funding is really on some smaller models, smaller case-specific models, which we think also will require less GPUs to be actionable.

Another component of Salesforce's expansion of its automation platform to work with other apps is a new $500 million infusion of capital into Salesforce Ventures' Generative AI Fund for a total of $1 billion to be invested so far into the technology. While Salesforce's venture capital fund made early investments in horizontal-reaching generative AI models and platforms such as Anthropic, Cohere, Mistral AI and Hugging Face, this round will likely focus on verticals, said John Somorjai, chief corporate development and investments officer at Salesforce Ventures.

"I think we did a good job helping the large language models -- those companies in our portfolio work more effectively with Salesforce ... and we feel very good about those investments," Somorjai said. "I think what we want to focus on with new funding is really on some smaller models, smaller case-specific models, which we think also will require less GPUs to be actionable. That's ... very interesting tooling that helps models and agents be more effective and more accurate."

And on top of all that, Salesforce and Google revealed a partnership Tuesday that enables Agentforce agents to work across Google Workspace and Salesforce Customer 360 apps. This bidirectional integration enables actions such as creating Google Slides and Docs from Salesforce data, as well as sending Google email and calendar data into Salesforce records.

Pricing still evolving

Auradkar said Salesforce isn't trying to become a competitor to data warehouse vendors such as Databricks or Snowflake. Instead, the idea is for Agentforce to make data -- whether it's in Salesforce, other cloud apps or large repositories such as Snowflake -- more usable.

"Data Cloud is the biggest upgrade to the Salesforce platform in our history, and I would argue that it's also the biggest upgrade for these lakehouses and warehouses," Auradkar said. "It's an upgrade for Snowflake, it's an upgrade for Databricks -- because we're bringing their data to life for it to be used in the flow of work."

What remains to be seen is how much all of this is going to cost, Constellation Research analyst Liz Miller said. Benioff said $2 per Agentforce conversation is a pricing benchmark for which Salesforce is aiming, with volume discounts added to that. Miller said a cost of $2 per conversation compares favorably with what companies spend for human conversations in the contact center -- which industry estimates say cost $2.70 to $5.60 per conversation. But Agentforce isn't human, and some Agentforce conversations will still be complex enough that they must be elevated to a human anyway.

More information will be needed before customers can understand exactly what they are buying.

"Will customers have a calculator or way to estimate what '$2 a conversation' really means? At some point, we need to admit we don't really know, and that might be OK right now," Miller said. "But we might have real bottom-line impact we aren't prepared for. Let's say that thanks to agents, I can triple my capacity to engage in conversations about returns or simple purchases. Sounds great, right? But what does [that] look like? Are we ready for that cost?"

Don Fluckinger is a senior news writer for TechTarget Editorial. He covers customer experience, digital experience management and end-user computing. Got a tip? Email him.