Agentic AI and Solution Architects

November 12, 2025 Holt Hackney Artificial Intelligence, Applications & Technology, Data, Elevating Architecture, Machine Learning, Software, Strategy & Planning 0

Courtesy of Shelly Palmer

By Gunnar Menzel, Master Architect at Capgemini

Seven years ago, we predicted the impact AI will have. In our TechnoVision 2018 Edition we noted “Because AI augments – or even replaces – our intelligence, its use will become ubiquitous in all walks of life and work.”  Fast forward to 2025 and AI has clearly leapfrogged all technology trends, and whereas GenAI gained significant attention from 2022, triggered by the release of ChatGPT, Agentic AI has taken over since last year. Agentic AI is and will be having a significant impact, and in my short note I will be focusing on how it might change the way we (IT) architects work[1].

1) What is Agentic AI?

As we noted in our latest AI Lab Paper “Business, meet Agentic AI” issued a couple of weeks back, agents are central to modern AI discussions, and terms like “agent” and “agentic” are used everywhere, often with varying meanings. Despite their recent popularity, these concepts have deep roots in computer science with well-established definitions.

Article content
Figure 1: Image taken from our AI Lab paper “Business, meet Agentic AI”

An agent is any entity that works on behalf of another entity, working to accomplish high-level objectives often using specialist capabilities. Agents have a degree of autonomy and authority to take actions that modify their world.

Related to AI this means that Agentic AI is a class of artificial intelligence that focuses on autonomous systems that can make decisions and perform tasks without human intervention. The independent systems automatically respond to conditions, to produce process results.

2) What is the difference between GenAI and Agentic AI?

GenAI mainly focuses on creating new content, while Agentic AI is more action-oriented, focusing more on autonomous decision-making. Agentic AI systems often leverage GenAI models but integrate them with other technologies to enable more complex, goal-directed outcomes.

As opposed to one-time prompting of a large language model using Copilot where a user types into an open-ended text field and gets a result without additional input, Agentic AI allows for more complex interaction and orchestration. Agentic systems have the ability of planning, loops, reflection and other control structures that leverages the model’s inherent reasoning capabilities to accomplish a task end-to-end.

As we noted in “Business, meet Agentic AI”, the ability to take action is what makes something an agent. An AI system might provide sophisticated analysis and recommendations, but if it can’t execute actions on its own, it’s an assistant or a co-pilot, but not an agent.

3) Does Agentic AI Always Use An LLM Tool?

Converse to common beliefs, Agentic AI and large language models (LLMs) are not the same thing. Whilst GenAI tools that are based on LLMs can be used as an agent, an agent doesn’t have to use any artificial intelligence. Many non-AI systems are agents, from simple thermostats to your car’s driving assistance systems to complex industrial control systems.

Another misconception is that Agentic AI is a single tool, like Copilot is a GenAI tool. Agentic AI is the broader concept of solving issues with limited supervision and an Agentic AI would typically involve a system of agents. This means it is more appropriate to refer to an Agentic AI system instead of tool.

Developing an Agentic AI application requires the use of a programming language such as Python, C++, or Java, as these languages provide the essential infrastructure for building intelligent behaviour.

They enable developers to implement decision-making logic, integrate machine learning models, process data, and interact with both digital and physical environments. Python, for instance, offers a vast ecosystem of AI libraries and frameworks that streamline development, while C++ delivers the performance needed for real-time applications. Without a programming language, it would be impossible to define how an agent learns, adapts, or responds to its surroundings, making such systems infeasible to construct or deploy.

4) How Do Agentic AI Systems Work?

Agentic AI tools are intelligent systems designed to operate with autonomy, agency, and authority—three foundational concepts that define their ability to act independently, pursue goals on behalf of users, and make impactful decisions within defined boundaries. These systems are often built using a multi-agent architecture, where multiple specialized or generalist agents collaborate, either in centralized or decentralized environments. Each agent maintains an internal model of the world, enabling it to interpret context, predict outcomes, and adapt to dynamic conditions. Whether deterministic or non-deterministic, agentic AI tools align their actions with user intent and ethical standards, continuously learning and evolving to handle complex tasks with minimal human oversight.

Article content
Figure 2 : Image taken from our AI Lab paper “Business, meet Agentic AI”

5) How We Solution Architects Could Use Agentic AI?

For many of us Copilot is a key tool when it comes to research a particular subject or finding a document. However, as detailed in my post “ChatGPT – the Solution Architect of the Future? “ two years ago, GenAI tools, like ChatGPT have the potential to help and assist an architect, however are not advanced enough to replace an architect.

With the emergence of Agentic AI this might change. As (IT) architects we drive change that creates business opportunities through technical innovation. One of the key activities of a Solution Architect is to design solutions by applying methods and techniques combined with technical and business expertise. The actual solution design process will follow a similar pattern to that of a creative technology design process. An architect will combine and group the different components together according to stakeholder group and will, over several sessions, develop concept views related to key architectural components, establishing different options.

Deciding the “right” option will mean balancing the various criteria like functionality, value for money, compliance, quality, and sustainability. IT architecture design involves complex decision-making, planning, and problem-solving that require human expertise and experience. That is where most of the architect’s work is focused on – using knowledge and experience to research a particular subject, to apply design thinking and to solve problems to establish a solution. And this is where LLM based GenAI tools like Copilot or ChatGPT have several characteristics that limit the use for a Solution Architect.

6) Comparing Agentic AI systems With LLM Based GenAI Tools

Agentic AI systems can overcome the limitations LLM based GenAI tools have. In particular the following (see my post):

Article content
Figure 3: Agentic AI and GenAI Comparison

7) Agentic AI & Solution Architecture Use Cases

As there two main aspects of uses cases: 1) to use Agentic AI during the solution design process and 2) using Agentic AI as a component part of the solution.

Agentic AI during the solution design process: A large part of our work is to draw artefacts together, to research, to analyse and to interpret. And this is where Agentic AI can really help. A system of agents that can assist us solution architects in aspects such as[2]:

  • Strategic planning An agentic AI could propose a phased migration plan to microservices based on current monolith architecture and business growth projections
  • Requirements gathering and design It could convert a product manager’s notes into a structured backlog with prioritized epics and user stories
  • Knowledge management When asked about a legacy system’s integration pattern, the AI could instantly retrieve and summarize the relevant architecture decision record (ADR)
  • Testing and QA automation After a design change, the AI could automatically update test scenarios and validate them against the new architecture

The list of use cases of Agentic AI during the solution design process goes on incl aspects such as proactively suggests improvements (e.g., “This API design may lead to tight coupling—consider an event-driven approach.”) or learning from interactions to tailor its support to the architect’s preferences and domain and collaborating across tools (e.g., Jira, Confluence, GitHub) to maintain continuity and context.

The second area is related to the fact that Agentic AI can be used as a solution component. For example, in areas such as autonomous system orchestration, DevOps & infrastructure automation, enterprise Integration, data pipeline management, support agents and security monitoring. And of course, the list goes on as these are just examples to illustrate where Agentic AI might be valuable.

Article content
Figure 4: Image generated using Copilot

8) Summary

Agentic AI is more than just a LLM based tool; it is a full agent-based system framework that that makes use of artificial intelligenceand focuses on autonomous systems that can make decisions and perform tasks without human intervention. Related to the role of a solution architect, Agentic AI can be applied in two broad aspects areas, one to use Agentic AI during the solution design process and two by using Agentic AI as a component part of the solution.

References

[1] I will be focusing on the Role of a Solution Architect

[2] but not limited to