Most companies are still thinking about AI in terms of productivity and efficiency at the level of the individual. The real opportunity? Using AI to reshape how entire processes run. Our Global AI Lead Sven Arnauts explains.
If 2023 was the year of the copilot, 2025 is the year of the agent. Over the past two years, I’ve seen many organizations experiment with generative AI tools like ChatGPT and Microsoft Copilot to support individual productivity. The idea was simple: give employees a powerful assistant, and you’ll unlock major time savings.
But that big wave of productivity? It turned out to be a smaller wave. Not because AI isn’t working – but because we’ve been looking at AI as a tool. The productivity wave isn’t coming from using AI as a tool but rather from how we infuse AI-capabilities in our way of working.
As agentic AI rapidly becomes more mature, we’re entering a new phase. A shift from helping individuals be a little faster, to redesigning entire processes to be radically more intelligent and autonomous. For CIOs, the real conversation about productivity is only just beginning.
To understand the transformative potential of agentic AI, we need to move beyond the current productivity paradigm — and look at how AI is used today.
AI copilots help individual knowledge workers write faster, search quicker, or summarize smarter. That’s valuable, but fundamentally limited: one copilot, one person. Your efficiency gains are constrained by how well that one person leverages the assistance of their copilot-assistant. And that’s exactly the problem. When AI is only used as a personal assistant, the individual becomes the bottleneck.
Agentic AI shifts the focus. They don’t focus on supporting the individuals, but operate at the process level, supporting all individuals involved in that process. Agents collaborate with systems, people, and other agents to complete end-to-end tasks across teams or departments. Their value doesn’t scale with the number of users, but with the complexity of your workflows.
That’s where real efficiency begins.
At delaware, we group agents into three types:
find and summarize information. They’re similar to the copilots you already know.
trigger workflows based on defined inputs. They act when prompted but still rely on humans to identify the actions and their conditions.
independently make decisions and coordinate tasks, continually enhancing their performance through experience. They operate with broader goals and fewer constraints.
The real strength of agents emerges when you combine them. Think of them as a network of specialists that hand off work to each other.
For example: a client protests an invoice. A retrieval agent checks the invoice. An autonomous agent decides whether the claim is valid and updates the system. An action agent then drafts a credit note. The final agent then writes a specific email to the client which includes the credit note and the customer specifics. Each expert-agent plays a focused role. Together, they deliver a seamless, intelligent response.
These expert agent networks replace rigid, hardcoded flows with flexible modular operations that adapt in real time. That’s the power of agentic AI.
Whereas Intelligent Process Automation (IPA) is the combination of robotic process automation (RPA) enriched with artificial intelligence, Agentic AI moves away from the hard-coded RPA rules and embraces reasoning capabilities to adapt its processes and actions based on unique contextual intelligence. For example, an IPA chatbot answers routine questions, while an Agentic AI chatbot analyzes sentiment and escalates critical issues automatically when the chatbot deems it necessary.
With Agentic AI you assign a goal and agents figure out how to reach that goal. You orchestrate intelligence instead of scripting every step in a process. This shift from giving instructions to setting intentions defines the next wave of automation.
Organizations typically start with Agentic AI in one of three ways. Both have value and are complementary.
are available in enterprise platforms like SAP S/4HANA, Microsoft Dynamics and Salesforce CRM, and offer a fast, accessible entry point. These agents support high-volume, repeatable processes such as purchase order approvals, marketing segmentation, or service desk triage. They’re embedded inside the tools, built to scale and offer proven impact with relatively low effort.
For many organizations, embedded agents cover a large portion of their initial automation needs.
are made available in configuration platforms like SAP Joule Studio, Microsoft Copilot Studio and Salesforce’s Agentforce, and offers business users a fast way to configure their own agents in a drag-and-drop DIY configurator. These agents move away from the embedded “generic” agents. Configurable agents will allow business users to use available building blocks to configure a more relevant agent for your organization-specific workflows.
are made available in developer platforms such as SAP Build, Azure AI Foundry and Salesforce Platform. Custom agents address use cases that are unique to your organization. Particularly when competitive advantage, customer experience or internal culture are at stake. Custom agents are the most relevant when they contribute to strategic differentiation as part of your business processes and context, to unlock new levels of value.
For example:
We’re building these today. And when deployed thoughtfully, they do more than save time. They help you serve clients better, support your people, and transform your operating model. Many organizations will blend embedded, configurable and custom agents over time. The key is to start from business needs and grow your agent ecosystem accordingly.
You don’t need to build a full agent ecosystem on day one. But you do need to start smart. At delaware, we use a few core principles to help clients get going.
In our workshops, we ask three questions:
These questions don’t surface flashy ideas. But they do point to the ones that matter.
Even the best agent won’t help if no one wants to use it. To get buy-in:
Start small if you need to. Many clients begin with embedded agents inside tools they already use. Others build custom agents for high-value or customer-facing tasks. Often, it’s a mix — and that works just fine.
You don’t have to automate everything. But you do need to orchestrate smartly, with business goals and people in mind. No AI can outmaneuver a frustrated employee with a clever workaround.
Throughout this piece, we’ve talked about agents making processes smarter, faster, and more autonomous. But even as these Agentic AI systems grow more capable, one principle remains constant: human oversight is essential.
All of us at delaware are strong believers in human-centered AI. Not out of nostalgia or resistance to change – but because trust, transparency, and compliance are business-critical. Especially under evolving regulations like the EU AI Act, auditability is not a nice-to-have. It’s mandatory.
That’s why we advocate for keeping a human in the loop. Whether it’s to review critical decisions, provide final approvals, or steer exceptions, people must remain part of the equation. Agents should work together with your teams, not around them.
Because if your teams can’t understand or trust what the system is doing, they won’t use it. And if you can’t explain why an agent took a certain action, neither will your auditor.
The more intelligence you embed in your processes, the more important it becomes to embed responsibility too.
Copilots helped individuals work faster. Agentic AI will reshape how organizations operate. This motion of process-oriented Agentic AI in addition to your individual productivity copilot is already underway.
Start with real pain. Focus on the processes that slow you down, cost you money, or frustrate your teams. Then build from there. You don't need a perfect plan. Just a meaningful first step.
The AI revolution isn't about faster emails. It's about building a smarter, more resilient organization — one agentic process at a time.
You don’t have to figure it all out alone. If you're exploring where agentic AI could fit into your IT strategy, looking for practical starting points, or trying to make sense of what’s hype and what’s real, Sven and his team of experts can help.
Sven Arnauts
AI Lead
Connect with Sven on LinkedIn
What topics keep you awake at night? What would you like to hear our expert's opinion about? Feel free to leave your suggestions in the box on the right.
/2-AdobeStock_329249481-(1).webp?mode=autocrop&w=1000&h=750&attachmenthistoryguid=7ca4a9d8-ac6e-4bd5-9a2e-4f856af4d9bf&v=&focusX=1917&focusY=1204&c=f9abf8199aecce6bc218ce3f130bf6354fac19c24e3b0f2b43cbcee287f19d7e)
/Salesforce_Einstein_Copilot-banner-3200x500-(1).webp?mode=autocrop&w=1000&h=750&attachmenthistoryguid=6dc4f084-a5e9-4e80-bfb9-5b7fbb433b53&v=&focusX=2870&focusY=227&c=91190b3d371e33fe2e045f814b07692df2c0444b1acf090e92d5caa9a420e1ee)
/Copilot-banner-3200x500-(1).webp?mode=autocrop&w=1000&h=750&attachmenthistoryguid=541888f4-0c28-4258-b2d2-2d204d069870&v=&focusX=2543&focusY=221&c=9e2c9d2325fdd3eeaa40cbb5fd964ae78e6817e87bac0a77d312fc66d9011c7e)
/AdobeStock_637051681-(1).webp?mode=autocrop&w=1000&h=750&attachmenthistoryguid=6d61829a-e74d-40a8-aa8f-9d3f0a31cf76&v=&focusX=2261&focusY=732&c=2bc5ed58e4e9bb588424ef44c29b16e94d1750897a01ea2aeadf7f2b81a64fb2)
/GenAi-banner-3200x500-(1).webp?mode=autocrop&w=1000&h=750&attachmenthistoryguid=12e7f669-85b5-4505-b24e-a81fbd32e055&v=&focusX=2676&focusY=293&c=302fa3823b6ff5f269c36f9a0477c6af1c063704766edf644b5d523c40fd190a)
/RAD_solution_page-Banner-1920x355-(1).webp?mode=autocrop&w=1000&h=750&attachmenthistoryguid=14c795e4-c4ab-49ae-b11f-9d46d78a4076&v=&c=0ee81656016c3302d33321dec13eb7801773d4396d5ce8364ca715ac1ba2415b)