TrailblazerDX 2024 - build your own custom AI automation with Einstein 1 Studio

Alice Steinglass speaking (Zoom screengrab)

Salesforce today announces availability of Einstein 1 Studio, a set of low-code tools for customizing Einstein Copilot, the conversational AI assistant for CRM that entered beta just last week. The announcement coincides with the the opening of TrailblazerDX, the annual Salesforce conference for developers and admins, where attendees will have the opportunity to try out the new capabilities and begin to assess how much it will change their roles. Alice Steinglass, EVP & GM of Salesforce Platform, believes the impact will be significant. She says:

AI isn't just changing how we build. It's changing what we build. Today, without AI, I need to hard-code every possible action in my code, every button, every workflow. AI is unlocking new possibilities for developers. It will let us have a conversation with the customer at runtime and take action. It will let us create dynamic, hyper-personalized experiences. And what we create with code today, we'll create with data tomorrow.

Einstein 1 Studio was first unveiled at Dreamforce under the slightly different name of Einstein Copilot Studio. The tweaking of the name links the toolset more closely to the underlying Einstein 1 platform, which brings together all the enabling components the AI works with — the Salesforce core platform and metadata framework, the unified Data Cloud, and the protection of the Einstein Trust Layer.

Three main elements

The three main elements of Einstein 1 Studio are as previously announced, with some elaboration of capabilities and timescales:

• Prompt Builder, now generally available — a no-code tool for creating and testing custom, reusable AI prompts which can then be included in applications and workflows. For example, an organization might want their contact center agents to be able to quickly call up a snapshot of all escalated cases for a customer — this could be created as a custom prompt and embedded in a contact record as a button, for one-click access to the information.
• Copilot Builder, now in beta — allows Salesforce admins and developers to configure, customize and extend Einstein Copilot to meet the specific context and needs of their organization. They can combine prompts with existing tools such as Apex, Flow and MuleSoft APIs to enable the AI assistant to complete tasks across any Salesforce application or external system. For example, an organization might want to automatically send an SMS message to a customer after a specific event occurs — this can be achieved by simply adding the action along with instructions that tell Copilot what data inputs are required.
• Model Builder, now generally available — rather than being restricted to Salesforce's own models, customers can choose to connect into other LLMs or their own custom models. Supported predictive and generative AI models include Anthropic, Azure OpenAI, Cohere, Databricks, Google Cloud’s Vertex AI and OpenAI. Later this year, enterprises will also be able to fine-tune LLMs on Data Cloud data, starting with Amazon Bedrock, Google Vertex AI, and OpenAI LLMs.

Other customization options include the ability to configure data masking in the Einstein Trust Layer, with admins able to select the fields they want to mask. In addition, audit trail and feedback data collected from AI prompts and responses is now stored in Data Cloud, where it can be easily reported on or used for automated alerts through Flow or other tools.

Customer experience

One customer that's already been benefitting from the new capabilities is BACA Systems, a manufacturer of robotic machines for the natural stone industry. Its entire business runs on the Salesforce core platform, with ERP, accounting, stock and deliveries provided by Salesforce-native AppExchange partners Rootstock and Bringg. Prompt Builder has helped automatically generate opportunity summaries, which has halved the time sellers take to get up to speed on a customer's relationship with the company and tailor the right personalized interaction. On the service side, it is helping to surface information from historic case comments that was previously difficult to find. Andrew Russo, Enterprise Architect at BACA Systems, says:

If you want to go and have an asset and find out what happened over the last three years it's been in the field, you really need to know, 'Okay, this component failed.' But you can't read every case. By giving all the context to the LLM, and having it go through all of it and just return, 'Hey, this case and this case are the two major ones where something had an issue,' that's magical for us, because it saves literally 20 minutes of time that you would read through all of those cases. Now they can respond back to the customer 20 minutes faster. So it's not only time saved, the customer's happier.

Also announced today — although not related to the AI theme — Slack has launched a free developer portal with access to sandboxes for testing, as well as a beta for creating new Bolt for Python and Bolt for JavaScript apps and support for scripting with the Slack CLI, making it easier to automate testing and deployment.

My take

Throughout the history of SaaS, the mantra ‘configuration not code’ has been a byword. But each configuration has still been a painstaking separate task. The huge impact of AI that is now starting to emerge is that application builders no longer have to craft every possible step in a given collection of workflows. They simply have to set up the prompts and workflows and then let the AI figure out what needs to happen when in conversation with the user at runtime. This is a very different way of thinking about application development and I suspect it may take some getting used to before everyone becomes adept at it. It will be interesting to see the reaction of attendees here at TrailblazerDX over the next couple of days.