Salesforce Inc. today rolled out Einstein 1 Studio, a set of low-code tools that will allow developers and administrators to customize the company’s artificial intelligence assistant Einstein Copilot and embed it across any customer management app or AI experience.

The company announced the new tools during TrailblazerDX, the company’s annual developer conference, as part of a broader embrace of AI enabling deep connectivity to enterprise data. Salesforce first announced Einstein 1 Studio in September as a way to customize and build secure AI applications using Einstein Copilot and access company data from within the Salesforce Data Cloud.

The company’s AI copilot, said Salesforce AI Chief Executive Clara Shih, differs from other copilots and consumer AI models because consumer AI models are trained on public datasets from the internet and enterprise models are grounded in trusted business data. The biggest problem that businesses face is that they have a great deal of data trapped in islands – databases, PDFs, email, files and so forth – that is largely inaccessible.

“This is why we’ve developed the Einstein 1 platform,” said Shih. “Einstein 1 platform is how every business can build trusted AI apps of the future built on trusted data.”

The Einstein Copilot AI assistant launched into beta mode last week and the new studio brings three different toolsets to the new platform. Developers will gain access to a prompt builder that will allow them to define prompts using low-code natural language, a Copilot builder that will allow them to configure and customize the AI assistant, and a model builder that will allow them to “bring their own model.”

With the Prompt Builder, developers and lay people will be able to construct prompts using their own words, so they don’t need to be data scientists, programmers or engineers to build AI prompts. Prompts are what users write to an AI to get it to provide them with results, for example, asking a question or telling it to summarize the results of a stream of data. Prompts can be simple or complex and sometimes must be written in specific language to elicit a particular response.

“If you can write a mail merge email, you can write a prompt,” said Alice Steinglass, executive vice president and general manager of Salesforce Platform.

The benefit of the prompt builder is that it allows anyone, including nontechnical users, to create prompts quickly without needing any training and connect data sources from within the company. Once a prompt is created, the user can modify it and iterate on it as much as they like by adding more data sources or iterating on it to fine-tune it to wherever they need it. Once complete, those prompts can then be copied out and reused elsewhere.

With Copilot Builder, now in beta mode, companies can tailor Einstein Copilot to fit their business needs better. Copilot has several capabilities above and beyond the usual conversational AI aside from being connected to a large number of data sources and the ability to read, analyze and summarize documents and emails, and connect to Slack, MuleSoft and other third-party sources. Copilot can also take automated actions on behalf of users, such as completing tickets, sending emails, updating database records and more.

Using the power behind Copilot Builder’s low-code interface, users will be able to tailor the actions that Copilot can take and add custom actions to complete tasks that the Copilot isn’t already capable of. For example, Copilot may already be able to email a customer upon request that a reservation is complete, but it might not be able to send a custom SMS text. Creating an action to send a customer a text would be a simple task of going into the builder to teach the Copilot how to send one.

“AI isn’t just changing how we build; it’s changing what we build,” said Steinglass. “Today. Without AI, I need to hardcode every possible action in my code, every button and every workflow. AI is unlocking new possibilities for developers; it will let us have a conversation with the customer at runtime and take action.”

Model Builder will allow companies to choose their own large language model, providing businesses the flexibility to connect to any third-party provider they need, said Steinglass. The tool allows customers to pick a model from Salesforce, or bring in an unmodified LLM from Cohere, Databricks, Google Cloud’s Vertex AI, OpenAI and others. Developers can also fine-tune those models using Salesforce Data Cloud using their own trusted enterprise data without it leaving secure enclaves.

According to Steinglass, even when enterprise customers choose to use Model Builder to bring in outside models, they still gain the benefit of the full trust and security of the Einstein 1 platform’s capabilities. This includes data masking, additional data security, audit trails and additional privacy measures built on top of models for safety.

Photo: Salesforce