Power Automate, Microsoft’s Power Platform service that helps users create workflows between apps, is getting new AI smarts. During its Ignite conference, Microsoft rolled out capabilities powered by OpenAI’s Codex, the code-generating machine learning system underpinning GitHub Copilot. Starting today (in public preview), Power Automate users can write what they want to automate in natural language and have Codex generate suggestions to jumpstart the flow creation.
It’s Microsoft’s latest move to more tightly integrate the various technologies from OpenAI, the San Francisco AI startup in which it has invested $1 billion, into its family of products. Two years ago, Microsoft introduced a Power Apps feature that used GPT-3, OpenAI’s text-generating system, to create formulas in Power Fx, Power Platform’s programming language. Microsoft also continues to evolve Azure OpenAI Service, a fully managed, enterprise-focused platform designed to give businesses access to OpenAI innovations with governance features.
“Our goal is that anywhere in the ecosystem that a person would need to write code they have the flexibility to start with natural language too, and Codex is core to that strategy,” Stephen Siciliano, VP of Power Automate, told TechCrunch in an email interview. “[These are] new tools that will help users eliminate tedious work and free up time for workers to focus on more high value projects.”
Image Credits: Microsoft
Using the new Codex-powered tool, Power Automate users can describe the type of workflow automation they’d like to create in a sentence. Codex will then translate this into flow recommendations, which — when set up with the appropriate connectors — can be fine-tuned within Power Automate’s flow designer to create an automated workflow.
Siciliano says that the feature will support “key” Microsoft 365 connectors at launch, and that there will be additional integrations in the coming months.
“We have fine-tuned Codex primarily with the thousands of templates that we have for Power Automate cloud flows today,” he added. Originally, Codex was trained on billions of lines of public code in languages like Python and JavaScript to suggest new lines of code and functions given a few snippets of existing code. “These templates are a combination of Microsoft-built and community submitted scenarios, so they cover a breadth of use cases and everything from very simple to more advanced flows.”
When asked about the longer-term roadmap, Siciliano declined to reveal much. But he suggested that Codex might come to more places within Power Platform in the future.
“[T]here are many different places in the Power Platform where natural language may be useful, so you’ll see a broader rollout,” he continued. “Moreover, we will continue to enhance the accuracy of the [system] over time as well.”
The new Codex-Power Automate integration dovetails with enhancements to AI Builder, which also landed this morning. (AI Builder, a built-in Power Automate feature, lets users add AI capabilities and models to automated flows.) AI Builder now offers users the ability to train AI systems on the data they might want to extract from documents, allowing Power Automate to pull data in freeform documents such as contracts, statements of work and letters, even from tables that span several pages. Microsoft says the document-processing capabilities of AI Builder now support 164 languages, including handwritten Japanese.
