Around a decade ago, as AI gained renewed attention, one of the first areas to receive overinflated focus were business process automation and desktop automation tools that aimed to automate repetitive user interface interactions. These tools, known as Robotic Process Automation (RPA), started as humble tools for automating repetitive user interface interactions, like keyboard and mouse actions. Over time however, RPA became over-hyped, capitalizing on market confusion about AI and misconceptions about robotic intelligence.
Of course, robotic operations do not imply any sort of intelligence. Indeed, factory robots have been performing repetitive, automation tasks for over half a century, without utilizing any sort of intelligence or AI system. For sure, the goals of automation and intelligence aren’t the same. Automation is all about predictable repetition of tasks to gain efficiency and perform so-called “3 K’s / 4 D’s” tasks that are otherwise dull, dirty, demeaning, and dear (expensive). Intelligence, on the other hand, is all about variability, adaptability, and probabilistic versus deterministic operation. You use intelligence when you can’t have guaranteed repeatability in inputs and outputs. Whereas you use automation when you can and should have guaranteed repeatability.
While RPA tools didn’t necessarily promise the capabilities we can now clearly realize in generative AI solutions, the market viewed them as a “gateway” tool to more intelligent capabilities. In part because adding AI capabilities to these otherwise unintelligent tools is becoming easier every day. Organizations are coming to depend on automation tools to increase efficiency but want to avoid the headaches of setting up and managing these tools. These headaches include complex and brittle setups, constant management, and monitoring to handle inevitable changes, bottlenecks, and process exceptions that constantly arise.
Moving up the Levels of Automation to Adaptive and Autonomous Process
AI can provide a lot of value in the context of process automation giving organizations not only the efficiency benefit of repetitive automation tools, but also enable intelligent, adaptive systems that can handle the constantly evolving, changing needs of handling data and information, especially as they involve multiple systems.
In 2019, research firm Cognilytica championed the idea of the Levels of Intelligent Process Automation (IPA). In the same way that autonomous vehicle capability can be measured in multiple levels with Level 0 signifying no autonomous capabilities, and Level 5 having full no-human involved self-driving ability, so too can we identify the capabilities of process automation from Level 0 with the human doing all the process development and driving, to Level 3 in which intelligent systems are able to not only define and manage complex integration and automation tasks, but also optimize them keeping them constantly agile.
The vast majority of today’s automated processes and integrations operate at Level 0 or Level 1 because a bit of natural language processing (NLP) is thrown in to handle tasks such as reading or generating documents or processing user interface needs. Few organizations are at Level 2, the most advanced level currently available with off-the-shelf tools, having AI systems that can intelligently discover and define processes and deal with a limited set of process exceptions without human intervention.
The future, however, is with Level 3 intelligent and adaptive systems. Many of today’s business processes, even with the use of RPA solutions, are utterly absurd. Taking data that is from PDFs, then re-entering that data into an online CRM, ERP, or EHR system, and then performing yet more automation tasks to re-enter that data into yet more systems is software enabled Kafkaesque ridiculousness. The challenge is, legacy systems have proven to be much more stubbornly entrenched within organizations and industries than might have been expected when they were originally implemented. This is why we still have COBOL-powered mainframes existing side-by-side with AI-powered Large Language Model (LLMs) operating in the cloud.
Agentic AI and Intelligent Automation
While legacy systems are here to stay, what most likely will disappear are complex, brittle, hand-stitched process integrations often referred to as “spaghetti-ware” by virtue of the entanglement of their complex interactions. As its replacement, we’re going to see AI-enabled autonomous business processes that leverage the strengths of LLMs and foundation models to understand when, where, what, and how to move and integrate data between systems. This however simply describes the “what” of what we want our adaptive systems to become, and now the “how” it gets done.
Recently making waves is the concept of Agentic AI. As defined by OpenAI, Agentic AI systems are “AI systems that can pursue complex goals with limited direct supervision”. While many people use LLMs and foundation models for the purposes of generating content outputs or analyzing and processing data inputs, many of the powerful Foundation Models being built also have the capability of executing complex, multi-step tasks.
Agentic AI systems can autonomously take defined objectives or goals, and then determine the steps necessary to accomplish those goals with minimal, if any, human interaction and oversight. These autonomous systems are also adaptive in that they can handle changes to systems they interact with and any problems they might encounter. Of course, using the power of LLMs, these systems can also handle a wide range of text, image, audio, and video inputs and outputs in a wide range of languages and levels of domain-specific context. Furthermore, these systems can also implement an amount of self-optimization and reasoning so they can constantly improve their interactions. This sounds a lot like the goals of Level 3 Intelligent Automation, doesn’t it?
Currently, Agentic AI is still in its infancy, in much the same way that fully self-driving vehicles are still in their early days. However, it’s clear that the momentum and speed in which Agentic AI capabilities are emerging combined with the strong desire by organizations to advance their automation activities will lead to a new wave of more-intelligent automation capability than is currently available on the market. Perhaps now technology reality can finally live up to the earlier promises of more intelligent adaptive automation that got the industry so excited.
