Businesses today are coming to a harsh reality with their AI initiatives. For organizations to be successful and deliver quantifiable value that shows up on the balance sheet, they need to take a fundamentally different approach. But it’s not just changing their technical approach. It requires an entirely new mindset.
The management consulting firm McKinsey & Company confirmed this observation in the article, “Seizing the agentic AI advantage.” McKinsey’s take-home message was this: While many organizations have adopted AI, the value remains elusive, signaling the need for a fundamentally different architectural approach.
Watch Build with Boomi Episode 2: Integrations to Agents: How to Architect for Agents
Think about where we are today. We’ve connected System A to System B. We’ve tried to synthesize data from a multitude of sources into unified experiences that (hopefully) ensure desired tasks and outcomes are executed with precision, every single time. This is the integration mindset, and it’s a crucial approach that thrives on structure and repeatability. It has become the bedrock of the modern enterprise.
On the other hand, agentic systems represent the next step in this integration evolution. They’re not rigid assembly lines. They’re autonomous digital workers built to accommodate change. We provide them with the goal and the tools necessary to get the job done, and they just accomplish it. When data formats or anything else change, agents adapt.
I want to make clear what we’re discussing here. While the concept might be new, no one is proposing that enterprises “rip-and-replace” the infrastructure they’ve worked hard to build with an integration mindset. My colleague Chris Hallenbeck has written about an insightful analogy between integration and travel. The current tech stack is like a train: powerful, but it can only travel where the rails take it. But agents are like all-terrain vehicles. They have the autonomy to go “off the rails” to reach a destination.
I’ll add that we don’t want to tear up all the tracks we’ve laid down and give everyone their own all-wheel-drive Jeep because these new paradigms don’t necessarily make older ideas obsolete. It doesn’t mean that everything you’ve done in the past was wrong. Instead, this new approach complements what you’ve done, just like the introduction of cars didn’t make trains obsolete. In fact, trying to apply the integration mindset directly to problems better solved by agents is like buying a high-performance Jeep, then putting it on a circular track and letting it go round and round.
Those rails represent our integration skills. They allow us to navigate towards decisions and actions. But while critical, they’re only one part of what it takes for IT to deliver value to stakeholders, whether they’re executives, customers, employees, or partners. Operating in this new dynamic environment requires a new plan, tool set, and approach to meet a broader range of needs. We need to expand our definition of automation and shift from a lens focused on individual tasks to the expansive view of a role.
With traditional integration, you might build a solution to perform a single specific task, such as “Let’s create an integration that focuses on how to process an email that contains an order number.” But in the agentic world, however, we design an agent to fill a higher-level role. An example might be, “Let’s design a customer support agent that has the skills, access, and authority to resolve several different customer problems.”
In this case, you need to give the agent a goal and a toolbox of capabilities – whether APIs, data access, or knowledge bases – that provide context for the agent to take the appropriate action. This is how a single agent, like a human employee, can solve more than one problem. The customer support agent can check an order, but also process a return, update an address, and answer a product question.
As you approach any new business problem, consider five key questions to help you expand on the integration mindset and determine whether a more resilient role-based agentic design approach is appropriate.
- Task vs. Role? This is a foundational test. If your design document describes a rigid process like A to B to C, you’re likely looking to accomplish a task. But if it’s something that typically a human would handle with a multitude of steps and tasks, then it’s probably better suited for an agent.
- Structured vs. Adaptive? Decide if the design requires rigid dependencies. Integration developers value clean, structured data, the type that looks the same every time. But by using agents powered by large language models (LLMs), we can fundamentally handle the inherent messiness of the real world. We’re talking about unstructured data. That’s what most organizations encounter at scale with unpredictable customer messages, unpredictable ways data flows in or out of systems, and so on.
- Static vs. Dynamic? In the traditional integration approach, we design static, reliable data flows between systems. We map from Field A and System One to Field B and System Two. If either API changes, the brittle integration breaks. In an agentic system, data flow is dynamic. An agent decides what data is needed and how to get it. If one API fails, a well-designed agent can detect the failure and look for an alternative path.
- Start/Stop vs. Always-On? Traditionally, integration developers design processes to be linear and finite. They have clear start and end points. In the agentic world, processes often are event-driven and continuous. A goal or event activates an agent, and it remains active until achieved. The end isn’t when the workflow diagram is completed. It’s when the problem is actually solved.
- Single Source of Truth vs. Triangulation? A core goal of integration has always been to create a single source of truth, often by building large data warehouses and complex integrations to ensure consistency. But agents are just like employees. They understand that “truth” can be synthesized from a variety of sources. It’s all about understanding the proper context. Just as a payment specialist understands the need to look at an ERP, a bank statement, and a CRM to get the complete picture and status of a payment, you can architect your agent the same way.
If you come from a deep background in integration automation, your experience building reliable connections is priceless. You understand that systems, data, and security are all integral to deriving value. You don’t want to throw that away. You want to harvest that skill and technology to apply it in a new way. Integration is the foundation of agentic. We’re evolving from building rigid, predictable assembly lines to designing flexible, autonomous, digital colleagues.
Use these five questions to guide your journey toward an agentic mindset. You can move from writing logic and code to automate tasks to designing agentic systems that perform roles that exist within the business. It’s time to unleash your creativity as you imagine, architect, and deploy these role-based agents to unleash new power within your enterprise, reducing manual effort across multi-step processes.
That speed, agility, and scale offer value today and tomorrow by future-proofing the business for whatever comes next in our fast-changing world.
Turn AI ambition into real outcomes.
See how Boomi Agentstudio helps you design, deploy, and scale intelligent agents across your business.