How Salesforce Is Using AI to Power the Enterprise

Overview

This episode features a conversation with Silvio Savarese, Chief AI Scientist at Salesforce, recorded at Dreamforce 2025. The discussion explores Salesforce's pioneering role in enterprise AI, including their development of foundational AI models, the AgentForce platform, and the use of simulation environments to ensure trust and reliability in AI deployments. The conversation also delves into the future of AI agents and their transformative potential for businesses and individuals.

Notable Quotes

- Agents are not just large language models. They are complex systems with memory, reasoning, actuators, and interfaces. – Silvio Savarese, on the multifaceted nature of AI agents.

- An agent without data, without context, is kind of useless. – Silvio Savarese, on the importance of integrating AI with enterprise data.

- We have to imagine a future where agents talk to each other. It’s going to be almost like thinking about the internet before the internet. – Silvio Savarese, on the transformative potential of AI agents.

🧠 Early AI Innovations at Salesforce

- Salesforce began developing large language models (LLMs) for developers nearly four years before the release of ChatGPT.

- These models were designed to enhance developer efficiency, enabling faster and more effective code generation.

- Salesforce also developed a trust layer to ensure AI outputs are safe and reliable for customers.

- The Atlas reasoning engine, introduced in 2024, powers AgentForce by enabling advanced task reasoning and execution.

🤖 How AgentForce Works

- AgentForce is a platform for creating agentic systems, allowing enterprises to deploy specialized agents for various tasks.

- It integrates four key components:

- Memory: Enables retrieval and contextualization of data and conversations.

- Reasoning Engine: Breaks down complex tasks into actionable steps.

- Actuators: Executes tasks through API and function calls, such as sending emails or making reservations.

- Interface: Facilitates communication via text, voice, or other modalities.

- AgentForce is designed to connect enterprise data with agent functionality, making it a practical solution for businesses.

🏢 Challenges in Enterprise AI Deployment

- Enterprises often face challenges with siloed data across departments, such as sales and marketing, which hinders AI's ability to operate effectively.

- Trust is a critical concern for enterprise AI, particularly regarding the accuracy and reliability of outputs.

- Salesforce addresses these challenges by focusing on data harmonization and building robust systems to ensure consistent and trustworthy AI performance.

🧪 Simulation Environments for AI Development

- Salesforce uses enterprise-grade simulation environments to test and improve AI agents.

- These environments simulate real-world scenarios, such as customer service interactions, using synthetic data and realistic variables like background noise, accents, and incomplete requests.

- The goal is to identify and address potential failure points, ensuring agents are robust and reliable in diverse situations.

- This approach helps bridge the reality gap between simulated and real-world environments.

🌐 The Future of AI Agents

- Silvio Savarese envisions a future where personal AI agents perform tasks on behalf of individuals, such as negotiating loans or making reservations.

- These agents will interact with other agents, necessitating new communication protocols and guardrails.

- The widespread deployment of AI agents is expected to transform society, akin to the advent of the internet.