What’s agentic AI and why is it so disruptive?
GD: Agentic AI is the next evolutionary step of generative AI. We're all familiar with chatbots, where you can communicate and ask questions in natural language and receive responses. Well, agentic AI goes beyond that because it's not just a chat interface; it acts as an additional resource within your organization, taking over your tasks and completing them. It understands your objectives and tries to achieve them with AI.
And it may find very creative ways to meet that objective. For example, the agent might communicate with other agents to get help because they might know more or have more experience with the topic. In a work context, agentic AI can autonomously complete quite complicated tasks, which is why companies are so excited about it. Additionally, it can operate continuously, with the agent working 24/7 in the background towards an objective.
What are the challenges of deploying AI agents?
GD: Deploying an AI agent is like introducing a very smart and well-educated junior colleague to the team. In a sense, it feels like you're working with someone who knows everything the internet knows, but without the decades of socialization and context that humans acquire over their lifetime.
This means agents can surprise you. Sometimes, these surprises are positive because the agent finds creative solutions and innovative ways to complete tasks. But at other times, the surprises can be negative because the agent might cut corners or take shortcuts that humans would avoid due to ethical considerations or common sense.
The challenge lies in harnessing the agent's creativity and efficiency while ensuring it does the right things.
How does SAP Signavio support here?
GD: When we look at AI agents, we want to know three key things:
- Is the agent efficient in reaching its objective?
- Is it compliant? Does it exhibit the right behavior while reaching its objective?
- Does it create overall positive business value, and if so, how much?
These are the three dimensions that our value proposition, AI agent excellence, addresses. More specifically, it’s mining that helps in all three areas.
For the first two dimensions—efficiency and compliance—we need to look inside the agent’s behaviour; it’s what we call agent mining. We want to understand exactly what the agent is doing, including the steps it takes and any creative strategies it develops. By reverse-engineering the agent's actions, we can identify inefficiencies and ensure it operates at peak performance.
Agent mining is also crucial to ensure compliance. To assess whether an agent is compliant, we need to look inside it. For example, does it use private social media data for decision-making, which might be forbidden? Does it access data it shouldn't? These are behaviors that we need to monitor and correct to ensure the organization stays compliant with the rules and regulations in place.
Additionally, we monitor how the agent evolves over time. Since agents are autonomous and creative, they may change their behavior when they find a more efficient way to complete tasks. It's crucial to spot these trends, especially if the agent starts deviating from the desired path.
Then, with process mining, we go beyond just looking inside the agent. We examine everything happening outside of the agent within the broader process. This is essential to assess the overall business impact and value. Only by looking at the end-to-end processes can we understand the real contribution of AI agents.
Oh and by the way - we can do all of this for all agents, no matter if it’s SAP, third party or custom agents.
How can we help companies build their own agents?
GD: Building an agent is easy. However, building a good agent is difficult and highly iterative, and because agents operate within complex processes, they need to know what is happening around them. Agents cannot be truly effective if they miss context.
While SAP Joule Studio, a capability of SAP Build, can be leveraged to develop custom agents, SAP Signavio can infuse those agents with the right process knowledge and process context, derived from process models, decision models and process insights. That way agents know what’s up and how to behave.
This ability to add context isn't just important for initial setup; it's also vital for ongoing governance and compliance. If you observe an agent performing actions that are inappropriate, you need to introduce new boundary conditions to prevent such behavior in the future. You can provide additional context, ideas, hints, preferences, or instructions on what to do or avoid. This continuous feedback loop ensures that the agent remains compliant and operates effectively within the desired parameters.
How does SAP Signavio address adoption barriers, ensuring that employees are comfortable and proficient in using agents?
GD: We tackle this challenge on two fronts. First, we leverage agents for process management work itself. For all SAP Signavio users, we aim to provide the best possible agents. We deeply embed the capabilities we've built into our products, allowing users to leverage these agents with a simple click to support process analysis and optimization endeavors.
Second, we help organizations figure out the best places to use agents along their processes—like spotting inefficiencies, bottlenecks, and repetitive tasks—and where agents can make the biggest difference. Starting from this analysis, we can suggest the most appropriate agents for the business processes that users are analyzing, managing, and improving.
When doing so, we leverage the AI Agent Hub by SAP LeanIX, which acts as a system of record for all AI agents across internal and external sources.
To this point, can you explain how the AI Agent Hub by LeanIX complements SAP Signavio’s AI agent excellence proposition?
GD: Firstly, why is the AI Agent Hub so exciting? The reason is that customers are rapidly adopting agents at scale and want to understand which agents they have, what technologies they are using, which applications they have access to, and what data they are leveraging. This helps in understanding how these agents fit into the overall IT architecture, and this is precisely what SAP LeanIX does.
With SAP LeanIX, customers can capture all the agents they have and stay on top of the full list, understanding how these agents relate to the rest of their IT stack. However, SAP LeanIX does not look inside the agent to judge its behavior.
As this behavior is inherently a process dimension, this is the focus area of SAP Signavio. For the more critical agents, organizations can apply SAP Signavio's agent excellence approach to look inside these agents and understand exactly what they are doing: are they efficient, are they compliant, are they providing the expected business value.
The synergy between SAP LeanIX and SAP Signavio is clear. Together, our solutions help ensure AI agents are functioning optimally and contributing positively to the business.
How do you see AI agents' role evolving in businesses, and how is Signavio preparing?
GD: AI agents represent a fundamentally new paradigm, and a key question remains: How should we perceive them? Should we view AI agents as advanced technical constructs or as non-human humans?
It's a complex question that organizations must address as they deploy AI agents at scale. Future organizations will consist of both human employees and AI agents working alongside each other. Just as with human teams, there needs to be clear responsibility and accountability. Who feels responsible if an agent misbehaves? Who will “fire” the agent? There’s no doubt that existing organizational mechanisms need to be adapted for the world of AI agents.
At SAP Signavio, we are preparing for the future by focusing on the organizational and process dimensions of agentic AI adoption. Our AI agent excellence approach aims to ensure that AI agents are integrated seamlessly, efficiently, and compliantly into the broader enterprise. In this rapidly evolving landscape, no company can afford to stand still. Our vision is to be an engine for innovation and reinvention, helping companies grow and adapt.