The rise of AI agents, powered by generative AI and Large Language Models, offers businesses an opportunity to enhance decision-making and automation processes, though challenges such as problem identification, prompt creation, and action monitoring remain.
The growing excitement around AI agents is palpable. The concept of agents that can collaborate to autonomously solve business problems has been around for years, but the Large Language Models (LLMs) that power Celonis.com/blog/generative-ai-unleashed-how-genai-and-process-intelligence-are-driving-business-transformation/">generative AI have finally made that concept a feasible reality. With their potential ready to be unleashed, they are widely expected to “define the next great wave of progress in AI”.
Speaking at the World Economic Forum’s Annual Meeting of the New Champions in Dalian, Darko Matovski, Co-founder and CEO of causaLens, outlined the transformative impact these agents are expected to have:
“There’s this tremendous opportunity to use AI agents and more advanced forms of AI, as we develop them, to transform completely how we make decisions in our society, which will lead to more equitable, more efficient, better societies.”
So, what does all this hype around AI agents mean for your business? What do AI agents do? What’s currently preventing us from handing every decision over to them? And, most importantly, is there a path to make them work for your enterprise? Let’s find out.
What is an AI agent?
An AI agent is a software program that can interact with its environment, collect data, and use that data to perform self-determined tasks to meet predetermined goals.
As McKinsey Senior Partner Lari Hämäläinen outlines: “When we talk about gen AI agents, we mean software entities that can orchestrate complex workflows, coordinate activities among multiple agents, apply logic, and evaluate answers. These agents can help automate processes in organizations or augment workers and customers as they perform processes.”
Using AI agents to automate processes in this way helps enterprises increase productivity, reduce costs, and improve the customer experience. And there’s no need to write code or build complex workflows because agents can be instructed in natural language.
An AI agent has three essential components:
- It can take directions, such as a human-provided objective given in natural language
- It has multiple tools to both retrieve context and/or take action
- It can autonomously apply reasoning to decide how and when to use the tools available to accomplish its objectives
For example, Celonis worked with a global manufacturing company to develop an AI agent that enabled the company to significantly reduce the time required to review and release credit blocks—a critical part of the Order Management process. Using data pulled from the company’s existing business systems and the process digital twin in Celonis, the agent analyzed blocked orders and comparable process flows, made a recommendation on the appropriate action for credit managers to take, and provided detailed information on why it made the recommendation. Having this information all in one place meant managers could quickly accept the agent’s decision with a single click or reject and provide a reasoning to learn from.
Find out what other AI tools businesses are already using today.
What are the challenges in deploying AI agents effectively?
With the technology now available, and clear benefits within reach, it’s easy to imagine businesses readily handing everything over to AI agents. But that’s not yet the reality as enterprises come up against a number of challenges. These include:
Challenge one: Identifying the problem to solve. AI agents perform well when the problem they’re supposed to solve is very clearly defined and narrowly scoped. Otherwise they struggle. For most businesses, identifying the right problem to solve will be key to realizing real value from AI agents.
Challenge two: Writing prompts that work. As LLMs improve, writing work instructions for AI agents will become easier. But right now it’s a time-consuming and iterative process that makes agent building hard work. Given that the value an agent delivers depends largely on the quality of the prompts and configuration settings it receives, agent builders need the right tools to build effective agents at scale.
Challenge three: Monitoring actions and decisions. AI agents act like humans more than other automation or software systems do. While this has benefits, it also means they are easily duped. They’re more likely to give inconsistent responses and forget things, or even hallucinate from time to time. Agents require governance and continuous improvement, so businesses must be able to track agents’ actions and decisions, and understand the impact on business processes. Agents will also need to be implemented according to the principles of responsible AI.
How Process Intelligence overcomes these challenges
Process Intelligence helps overcome the challenges identified above, and therefore make AI agents useful, scalable and reliable for your business. This power stems from its ability to provide enterprise-specific process context.
To understand how this works, it helps to consider the difference between consumer AI and enterprise AI. Consumer AI relies on a wealth of training data, such as web pages, but also on archival resources like Wikipedia to provide context and explain how those individual data points relate to one another. Enterprise AI has access to a similar wealth of raw data from business systems like ERPs and CRMs, but it lacks the Wikipedia-like reference layer to connect all those data points together and provide valuable context.
Celonis provides businesses with this connective tissue in the form of the Process Intelligence Graph – a semantic layer for business processes that sits above all the raw data coming from your business systems to provide context and meaning. This context can be used to feed AI agents so they can speak the language of your business. And third parties can contribute to this contextual layer through an extensible data model, creating a Wikipedia equivalent for enterprise AI.
Find out more about how Process Intelligence takes LLMs to the next level.
How does Celonis work with AI agents?
Celonis offers integrations that help companies build AI agents, and then orchestrate those agents once they’re up and running. The Celonis platform sits between your business data, systems and people and the agent, using Process Intelligence to support the following three-step process:
- Step 1: Discover relevant business problems. Process Intelligence surfaces value hotspots in your processes.
- Step 2: Build the agents. Process Intelligence supports you in understanding how problems are solved today and generating work instructions for AI agents.
- Step 3: Run the agents. Process Intelligence allows you to track the agent’s decisions and actions, overlaying the process steps they run with the traditional process steps in your ERPs, workflows, and other tools.
The use of AI agents to maximize productivity, cut costs, and deliver the best possible customer experience is set to surge in the coming months. If you’re using Process Intelligence to power AI agents you’ll be able to identify the right business problems to solve, build more effective agents at scale, and continuously monitor performance to responsibly drive continuous improvement. This will give your enterprise a vital edge over your competition as everyone races to make the most of this next wave of AI progress.
