The synthesis of generative machine learning models and digital agents, which many have dubbed Agentic AI, may well prove the most advantageous development for enterprise-spanning deployments of Artificial Intelligence.
The tandem of these technologies drastically extends the benefits they offer. The natural language understanding and generation of AI models are considerably enhanced by the action enabled by digital assistants. When properly implemented, ChatGPT does much more than summarize, write, or answer questions about documents.
This model, and other generative models paired with bots, can actually perform tasks to accelerate workflows and spur process automation to achieve mission-critical objectives.
The relatively recent launch of Foxit’s AI Assistant, in the Foxit Admin Console, provides a tangible demonstration of these capabilities. Based on ChatGPT, it allows end users to streamline document management by simply prompting the bot—not for answering questions about, or writing, content, but to manipulate the actual documents themselves.
The implications of these capabilities are far-reaching. According to Deedee Kato, Vice President of Corporate Marketing at Foxit, this functionality may soon result in AI agents performing intricate tasks, like readying pdfs for court filing.
“We know that there’s a bunch of steps that have to happen that law firms have to do, in order to submit court cases,” Kato said. “You have to have bookmarks. You have to have it compressed. There’s about 10 different things you have to do before you can file it with the court. We’re thinking in the future, we might have [such] a multi-step workflow in these prompts.”
Smart commands
At present, Foxit users can issue smart commands—natural language prompts to the system’s digital agent—to perform single actions at a time. This feature enhances the codeless motif presently typifying Intelligent Document Processing solutions, many of which are based on drag-and-drop, point-and-click methodologies. Instead, one can simply direct the system via natural language to perform any variety of tasks for managing pdfs and e-signature activities.
“Our differentiator is that within AI Assistant you can fully execute a pdf command, like ‘rotate pages one through nine 90 degrees’,” Kato commented. “You can put that as a prompt and it will execute the command versus like Adobe or ChatGPT. If you put in that command or prompt, it’ll just explain what ‘rotate’ means.” With natural language prompts, it’s possible to implement agents to convert, edit, issue comments, initiate e-sign activities, add certifications, and implement other actions on pdfs.
Kato estimated there are approximately 80 activities AI Assistant can complete. Foxit also relies on ChatGPT to supplement user activity by rewriting pdfs in different tones of voice and lengths, translating and summarizing them, and more. “It’s anything in the ribbon, like password protect,” Kato mentioned. “There’s a lot of stuff you can do for pdfs; I only use about 10 percent. Law firms probably use a lot of the things that I don’t.”
License management
In addition to modifying pdfs to prepare them for the next phase of a workflow, the AI-powered digital agent can complete a number of administrative tasks pertaining to licensing Foxit products and services. Instead of utilizing manual methods, users can simply prompt the agent to perform activities related to licensing in bulk. “We see the AI Assistant as automating these manual tasks,” Kato remarked. “Instead of just doing it one by one, rotating each page, or signing each individual license, we see it as automating multiple tasks.”
These capabilities are invaluable for performing these tasks at enterprise scale, particularly for situations in which organizations are engaged in training sessions and onboarding new employees. There are specific features for assigning licenses to particular groups and domains, and for information in CSV files or spreadsheets. “Like for Baker McKenzie, there’s thousands of users and there’s a license for every user,” Kato pointed out. “So, to manage those licenses, let’s say someone leaves Baker McKenzie and then another employee starts. With the admin console you can move that license to the new user.”
The greater significance
Eliciting action from digital agents infused with language models represents the next progression of enterprise AI—and the premise for what has come to be known as Agentic AI. It’s possible to build agents for respective tasks, combine agents for specific workflows, and increase the overall sophistication and complexity of deployments of these intelligent bots.
Within this realm of possibilities, facets of intelligent document processing may seem relatively tame. Nonetheless, they herald the enormous potential this technology has to offer to the enterprise in its inexorable march to benefit from these expressions of cognitive computing.
