One of the promises of generative AI is the ability to create code based on a written request. This can lead to the ability to generate SQL queries and simple API calls to move work between systems.
Salesforce is laying the foundation for a generative AI-driven workflow system, and it’s in the early stages of working with customers on this capability, says Liam Doyle, senior vice president and general manager for MuleSoft Automation.
“It’s really the culmination of a lot of different work we’ve been doing in different domains coming together. And it’s really just a powerful end-to-end stack,” Doyle told TechCrunch. This includes Data Cloud (formerly known as Genie), the company’s data warehouse tool; Einstein GPT, the company’s generative AI tool; and Flow, the company’s workflow tool.
It actually makes a lot of sense to pull these three products together because the combination enables users to simply enter something in plain text that they want to accomplish, and the software will take care of the request for them in a similar manner to ChatGPT, but in a business context.
“The convergence of these tools is the necessary kind of endpoint here, which is how can customers start to take advantage of all of the data that is out there that all of their business creates and harmonize it and make sense of it, and then be able to deploy really complex multi-system workflows from that data,” Doyle explained.
What this means in practice is a couple of new tools that the company is working on with early design customers right now. That includes Einstein GPT for Flow, which combines this generative AI textual interaction to create a workflow with little or no code simply by typing an action you want the system to take.
For now, that means if it’s a simple workflow, the program creates it on the fly, but for more complex ones, it could require a human programmer to help.
The other tool is called Data Cloud for Flow, which uses data-driven event triggers to drive actions in Flow. Doyle gives an example: “When a salesperson reaches 100 sales, I would like to trigger a notification in Slack and write some celebratory text to support that workflow,” he said. That’s a pretty simple example, but you could create more complex interactions based on multiple data triggers to generate reports, write emails, send texts or Slack messages and so forth.
The underlying model is a blend of available models including those from OpenAI, Google and Meta and their own model as well, and combining these with the Einstein intelligence layer the company has been working on since 2016, according to Doyle.
The tooling is still in development at the moment, but is expected to go into early beta release in June with the ultimate goal of releasing it next year. Doyle says that while it may feel premature to publicize it before it’s even in beta, he says the goal is to help customers understand what’s coming in this rapidly evolving tech world we’re living in, and in this way, they can begin to prepare for and think about the next generation of tools.
