16/4/2024

AI Actions

Previous blogs have touched on the subject of AI as a tool that is good at analyzing information and generating insights. But what if AI could not only think, but also do? Our partner Robocorp has now introduced a first step in this with AI Actions.

AI so far

Generative AI focuses primarily on information retrieval and understanding and helps with tasks such as coding, marketing and sales. However, its potential extends beyond just generating and analyzing data. The future of AI lies in its ability to interact with real systems. While we are familiar with AI applications such as ChatGPT, which use Python interpreters or Google searches, the next step is to integrate AI into business systems, allowing it to not only provide instructions but also complete tasks on behalf of users.

AI actions: closing the gap in enterprise systems

To bring AI safely and effectively into enterprise environments, Robocorp has come up with something new, AI Actions. AI Actions consist of solutions, workflows and automations that can be safely exposed to AI apps or agents. These actions are defined by their input and output, described in natural language, allowing AI apps to understand and use them. Actions are predefined code fragments that are executed in a controlled environment, equipped with access rights, audit trails and necessary controls.

Although actions share similarities with tools and function calls in frameworks such as Langchain, Llamaindex and Microsoft Copilot Studio, they offer additional advantages. Actions can enforce strict user approval before execution and operate within an environment separate from the AI app, providing all required controls, authentication and access rights. Actions can range from simple tasks such as retrieving data via an API to complex operations such as creating a new purchase order in SAP.

The benefits and open-source approach

Using predefined AI actions instead of generating code on-the-fly has a number of advantages, including:

  • Ensuring the accuracy of results
  • Understanding and strictly enforcing the desired behavior
  • Reusing actions across multiple AI apps
  • Enforcing human validation
  • Managing authentication for third-party systems
  • Achieving fast and scalable performance
  • Providing visibility for actions

This approach also addresses the needs of business decision makers who stress the importance of robust AI integration into their systems. To foster innovation and collaboration, Robocorp has made their AI Actions open-source and is inviting developers to contribute and expand its capabilities.

In summary, AI actions represent a significant step forward in the evolution of AI, providing a controlled and efficient way to integrate AI into business systems. If you would like to explore these opportunities for your organization, please contact us.