Salesforce first introduced its AI layer called Einstein back in 2016 to provide predictive AI services across the Salesforce family of products. In March, just months after the release of OpenAI’s ChatGPT, it introduced Einstein GPT to bring the ability to ask questions about the software in natural language across the platform.

Today, at the Dreamforce customer conference, taking place in San Francisco, the company announced the next step in its AI journey, introducing Einstein Copilot, which embeds this ability to ask questions in the context of whatever users are doing, regardless of product.

Clara Shih, CEO of Salesforce AI, a pretty important title given the role of AI in the company these days, says Einstein GPT was the first attempt to spread generative AI across the platform. It was designed to provide a kind of generalized automated assistance, like generating a response for customer service or writing an email for a salesperson, but for many customers that wasn’t enough and that’s why they are introducing Copilot.

“We are launching Einstein Copilot which is a conversational AI assistant for companies, employees, as well as their customers to securely and safely be able to access generative AI to do their jobs better, faster, more easily, to augment and amplify their their own abilities, their skills, their work, be more efficient and be more productive,” Shih told TechCrunch.

The idea is to let users ask the bot in a conversational way to get information that would often take several clicks and knowledge of how to do it.  A salesperson could research new accounts, a newer customer service rep might ask how to deal with a return over 30 days and in commerce, a product manager could create a customized storefront for a new product launch. Instead of asking someone or searching a solution, they can just ask Einstein Copilot in plain language, and it will find the information for you, assuming it’s been trained to answer the question.

Brent Leary, founder and principal analyst at CRM Essentials, says that while just about every software company is embedding generative AI at this point, he sees Salesforce with one distinct advantage. “What could potentially separate Einstein Copilot [from the rest of the enterprise software pack] is the coverage across the customer touchpoint spectrum, including commerce. That creates an opportunity to impact a plurality of customer interactions and impact not only customer experiences, but also the experiences employees have while engaging customers at the time of need,” Leary said.Shih recognizes these large language models have issues, and she is pretty open about them. “We know that there’s an AI trust gap. There’s a gap and there’s reasons for this…Islands of data can result in hallucinations and incorrect or incomplete outputs,” she said in a press event this week.

Salesforce believes that by linking Einstein Copilot AI tooling to data coming from its own Data Cloud (introduced last year at Dreamforce as Genie), and building its own model, it can reduce some of the issues we have seen with large language models, particularly around hallucinations where the model makes up an answer when it doesn’t have enough information.

The company also has introduced the concept of what they are calling ‘a trust layer,’ essentially an underlying security, governance and privacy architecture to give customers more confidence as they start using Salesforce generative AI tooling internally and externally with customers.

That said, it is widely believed that there is currently no known way to completely eliminate hallucinations in large language models. 

Einstein Copilot is currently in beta with customers. Salesforce did not provide a projected release date. Einstein Trust Layer will be generally available across the Einstein platform next month, according to the company.