Trending Articles
Trending Topics
Trending
In its effort to shift enterprises from using AI at the model level to the deep-down system level, Salesforce today launched AI Foundry. Its goal: to help AI researchers, customers and partners to collaborate and roll out enterprise AI products that solve pain points in a secure manner.
“We’re announcing AI Foundry, an initiative from Salesforce AI Research to focus on our big bets for the next few years,” said Itai Asseo, VP of Salesforce AI Research during a pre-launch briefing.
”The goal is to take foundational research on things like simulation environments, multi‑agent systems and ambient intelligence, and turn it into system‑level AI that we can bring much more quickly into our products.”
The core idea behind AI Foundry is to provide an environment where Salesforce can work closely with its most strategic customers to validate new technologies against what Asseo termed “real enterprise problems. That incubation model lets us partner tightly with our engineering and product teams, so we can move from breakthrough research to production‑ready capabilities much faster than before.”
So what are the three big bets that Salesforce believes will be the foundation of enterprise AI in 2027? It describes them as:
Let’s tackle Simulation Environments first. “Learning from experience means learning by absorbing positive feedback from positive behaviour and penalising mistakes,” said Silvio Savarese, Chief Scientist at Salesforce.
“We can do this by true simulation environments. In the context of enterprise AI, simulations environments are critical for creating thousands of realistic scenarios and which can be populated with synthetic data, which mimics real customers’ data or real business logic. And within those environments, we can measure how agents can handle those complex business cases and how we can use feedback to reward positive outcomes and penalise mistakes.”
For the second bet, Savarese pointed out that enterprises have already embraced AI agents and “that we will soon see agents that not only operate with the same org but across orgs. We soon will have personal agents [and] local agents interacting with business agents.”
Rather than rush into this world, we need to prepare for it. “We need to not only look at how two agents will interact at the protocol level, but also at the more semantic level,” he said. “And semantic level means essentially, what are the rule of conduct that agents will need to have in order to operate in a safe and proactive manner? How [do we] ensure that these negotiations are kept within the right legal and safe boundaries?”
Then we come to what Salesforce calls Ambient Intelligence, where the agents recede into the background.
Today, said Savarese, “they operate based on prompts and specific instructions. But very soon, and already now, we have seen that agents are being seamless, integrated, embedded in the background.
“They will be fundamentally aware of the context of what’s happening, what is the workflow in which they are operating. They will be proactive and they will be capable of delivering insight, assistance, information without being specifically prompted.”
This, he explained, is what Salesforce means by Ambient Intelligence.
If you’re interested in finding out more about what Salesforce AI Research does, head over to its dedicated website.
