
“The agent will see you now”: Inside Salesforce Israel’s mission to build responsible AI
Startup Nation is home to an 800-person Salesforce R&D center dedicated to managing the security, privacy, and governance features across Salesforce AI tools.
Salesforce Israel is home to 1,000 employees, with roughly 800 working in its R&D center. Led by Oren Winter, who is also Senior VP of Engineering of Salesforce Trusted Services, the site is responsible, among other things, for the company’s AI trust, safety, and governance - an issue increasingly critical in the age of autonomous agents.
“You need to have two basic things to actually run agents,” said Winter. “One is a reasoning engine: that piece of code that knows how to effectively talk to the LLM and make it perform…And then the last piece that you need to have is the data. You need to have the enterprise data.”
Israel has one of the highest concentrations of AI talent globally, ranking 6th overall and 1st among women. Startup Nation is ranked first in industry-specific areas such as Education, Financial Services, and Manufacturing, meaning that plenty of global companies are ready to set up an R&D shop to access the talent pool. Salesforce joins the likes of Apple, Intel, Amazon, Meta, and NVIDIA - all of which have varying projects surrounding artificial intelligence in areas like chip technology, cloud infrastructure, cybersecurity, and productivity tools.
The company hopes to not just contribute code, but shape the global conversation around agentic AI. Its investment in the tech started roughly a decade ago and is evolving from Einstein, its own AI for CRMs, to Agentforce, helping businesses build and customize autonomous AI agents to support employees and customers.
According to numbers provided by the company, the last quarter of fiscal year 2025 saw more than 5,000 Agentforce contracts closed globally with all types of companies, such as OpenTable, Indeed, and more. Winter says the upgrade from traditional chatbots means that the agents will communicate in natural language and depart from traditional text-based tools, and could even appear as video-based synthetics to look like people and help companies “augment your human headcount with those to actually perform basic tasks.”
The new age of agentic AI is a departure from more historic chatbots, especially given the jump in LLM technology that paved the way for more conversational AI through tools like ChatGPT. Whereas traditional chatbots would help with basic tasks such as restaurant bookings or scheduling for appointments, Agentic AI refers to AI systems that act as autonomous agents, capable of initiating actions, pursuing goals, and interacting with their environment, often with minimal human intervention. These systems are not just reactive or predictive but are proactive, self-directed, and often multi-step in their reasoning and decision-making.
Agentic AI is largely seen as a stepping stone toward Artificial General Intelligence (AGI), but it’s not without risk. Guardrails need to be in place to ensure safety and control to prevent ‘going rogue’, ethical decisions made by the tools, and the security of the data it relies on. According to a Salesforce study, 54% of consumers say they trust the use of AI agents, but only 10% say they ‘completely’ trust them. This is due, in part, to people being naturally cautious about new technology, but also because many customers are concerned about how their personal information is used and protected, and are becoming more careful about safeguarding it.
Therefore, the presence of AI, agents, and the information security for them have both potential and challenges for a company that may choose to adopt them: Even if they may open up more opportunities for information security, deploying them in an organization increases the attack surface, placing a strong emphasis on information security and leak prevention.
One of the ways Salesforce has achieved adequate IT security for Agentforce is through the $1.9 billion acquisition of Own, a DLP company that provides companies with backup to protect against the loss of essential data. It also purchased Zoomin to help it gain deeper insights into customer behavior and preferences through Zoomin's unstructured data management capabilities. Salesforce also intends to be part of the conversation in how to set global standards in agent interoperability. Winter predicts that LLMs will eventually become a commodity, and companies will be encouraged to collaborate toward two major trends: MCP (The Model Context Protocol) and A2A (Agent-to-Agent).
Some large tech companies have been taking note, attempting to steer the race. Earlier this month, Google announced A2A, its new open protocol empowering developers to build interoperable AI solutions. Early work done in achieving a level playing field will position this as essential infrastructure for the future of “plug-and-play” enterprise agents.
“The world is going to be flooded with agents, and we need to have a better way to go and manage that interaction,” said Winter. “Whether it's agents-to-real world, whether it's agents-to-agents, you are seeing some sort of open source. Right now, it's open source initiatives to go in and align that.”
Startup Nation's experience in cyber, fintech, productivity, and chip technology makes Israel uniquely suited to tackle the dual challenge of innovation and regulation in the AI space. And companies like Salesforce are ensuring their R&D centers are front and center of the new arms race focused on talent and trust.