Salesforce’s agentic AI platform to transform business automation

Salesforce has unveiled its Agentforce platform in a bid to provide enterprises with intelligent conversational capabilities and autonomous agents designed to handle a wide range of day-to-day tasks.

This move by the customer relationship management (CRM) giant marks a significant shift towards agentic artificial intelligence (AI), where autonomous AI agents act on goals and decisions, pushing the boundaries of business process automation.

According to Forrester, a technology research firm, AI agentic architectures require multiple models, advanced data architectures and specialised expertise, which can be challenging for organisations to build.

Salesforce hopes to address those challenges with a sophisticated reasoning and learning engine dubbed Atlas, the driving force behind Agentforce’s ability to continuously adapt and improve to address each organisation’s unique needs.

The key to Atlas’s capabilities lies in its architecture, including the use of specialised embeddings models that enable Agentforce agents to understand the nuances of different business processes, data formats and industry-specific requirements. This allows the agents to operate seamlessly within the customer’s existing workflows and systems.

Unlike traditional reinforcement learning approaches that rely on human feedback, Atlas was designed to continuously monitor the real-world impact of its actions and automatically adjust its behaviour to achieve better results.

This “reinforcement learning from customer outcomes” approach is made possible by Salesforce’s position as the world’s largest database of customer data and outcomes, said Salesforce AI CEO Clara Shih, ahead of Dreamforce in San Francisco this week.

“The more you use it, the smarter it gets for your business,” she said. “It’s looking at things like, ‘Did it improve the conversion rate? Did it handle the customer support issue faster, with higher customer satisfaction?’ Those are outcomes that live in Salesforce and Atlas is constantly tuning.”

All the data that powers Agentforce’s capabilities comes from the Salesforce Data Cloud, which enables businesses to build a unified view of their customers by bringing together structured and unstructured data from across the organisation.

“Knowing your customer across all touchpoints, channels and modalities is not trivial,” said Rahul Auradkar, executive vice-president and general manager of Salesforce Data Cloud and Einstein. “We’re seeing over 70% of our customers face this challenge today.”

If an agent is going off the rails and customers are upset, they can immediately take over and seamlessly escalate it to a team member

Data Cloud solves this problem by integrating data from multiple sources, including CRM systems and external data lakes, as well as unstructured content like audio and video. This ensures Agentforce agents have a comprehensive understanding of each customer, allowing them to deliver personalised and contextual support.

Data Cloud also features a new sub-second, real-time data pipeline, enabling businesses to instantly activate customer data and trigger timely, AI-driven actions.

“Data Cloud is the core of how Agentforce works,” said Auradkar. “Our agents are differentiated because we have a foundation of accurate, reliable and trusted data and CRM context.”

Early adopters of Agentforce have already seen results. Wiley, a publishing company, experienced a 40% improvement in case resolution during its busy back-to-school season by scaling Agentforce agents. OpenTable, the restaurant reservation platform, is using Agentforce to respond to customer enquiries with personalised regional insights. And BACA Systems, a small business that produces equipment for the global natural stone industry, reduced average handling time for repeat issues by 26% with the help of Agentforce.

Salesforce is making a significant investment in Agentforce, with plans to onboard 1,200 customers at Dreamforce this week, allowing them to build their first AI agents in just minutes.

The company is also launching the Agentforce partner network, enabling third-party developers to create specialised agents for various industries and use cases.

IBM, for example, is building banking agents that can handle credit checks and create client onboarding documents to speed up loan approvals, while Google is building Agentforce actions to automate tasks in Google Workspace, such as generating Google Docs, searching across Gmail and triggering Google Calendar events.

“With our new partner network, we’re leaving the era of disconnected copilots behind and moving into a future of a broad, open, connected network of interoperable third-party systems,” said Shih.

Gartner noted that while AI agents are a groundbreaking technology set to autonomously execute complex actions across a multitude of industries, they also raise significant security and ethical concerns.

With Agentforce, agents inherit the existing sharing models and permissions that Salesforce customers have already defined for their various user roles and profiles. Shih said this ensures agents only have access to the data and business processes that are appropriate for their designated responsibilities, preventing unauthorised access to sensitive information.

Auradkar added that policy-based governance in Data Cloud also allows for AI-driven tagging of data, such as identifying personally identifiable information. Administrators can then define policies that govern the use of and access to this data, ensuring that it is handled in compliance with relevant regulations and security best practices.

Finally, Salesforce’s existing Omni Supervisor feature, which provides sales and service leaders with a command centre to monitor the activities of their human teams, has been extended to Agentforce. “If an agent is going off the rails and customers are upset, they can immediately take over and seamlessly escalate it to a team member,” said Shih.