The role of a Solution Architect has always been to balance vision and reality. The question is, is there a change in how that process begins with the use of generative AI? The answer is, indeed, there is a change, but not in replacing the architect, but in using a powerful co-architect in the beginning of a design process.

For many decades, Solution Architects have been using their experience, documentation, whiteboards, and lengthy discussions with stakeholders for designing system’s. Architecture has always been collaborative, but historically it has also been very manual. As architects, the typical process for us usually starts with requirements collection and whiteboarding ideas with understanding existing process. Only afterwards we do use tools like Draw.io, Visio, Lucidchart, or sometimes even Mermaid scripts to transform those ideas into diagrammatic forms. This process can be laborious and time consuming.

Designing architecture is not just about conceptualizing the system’s behavior; it is also about creating each artifact manually, whether it is a diagram, a presentation, or documentation. In fact, much of the Solution Architect’s time is spent converting their concepts into visual or technical artifacts.

Today, generative AI is starting to alter this process, not by replacing Solution Architects. But by becoming something which helps Architect: a Co-Architect.

Before AI: Architecture Process Was Manual and Meeting-Heavy

The early stages of the design process, prior to the advent of generative AI tools, followed a pretty standard process.

A new system initiative would normally start with meetings with the stakeholders involved. The requirements would be gathered from the product teams, operations, security teams, and engineering teams.

Only after these meetings would the architect start drawing the diagrams. In many cases, the process would be as follows:

Stakeholder meetings → requirement gathering → system impact analysis → manual diagram creation → architecture reviews → presentation preparation for Level 1 followed by Level 2 & 3.

Even the drawing of a simple system interaction diagram would involve significant effort. Writing the Mermaid syntax, making the flows correct, making the visual layout correct, and making sure the diagram communicated the design effectively would involve multiple iterations.

In my own experience with large-scale enterprise systems, the early stages of the design process would involve many hours of meetings before the first diagram was even drawn.

It is the exact process that is now being augmented with the help of AI tools.

Enter AI: The Co-Architect Era

Modern generative AI tools like Gemini, Copilot, ChatGPT, and NotebookLM have revolutionized how the design process can start for architects.

The fundamental shift is not in the fully automation of the process but the shift is in the speeding up of the starting point and speed the time to market.

Architects do not start with a blank page anymore; they start with a partially formed architecture draft created by AI tools.

Architects do not start with version 0 of the design; they can start with version 0.7 of the design.

The evaluation and refinement of the design are still performed by the architect, and the initial phase is sped up significantly.

1. Instant Architecture Diagrams from Requirements

One of the most useful applications of AI in the field of architecture is the quick generation of diagrams.

Architects can ask the AI to generate the code using the system requirements and ask it to generate a sequence diagram or interaction diagram.

For example, they can type the following in the prompt window:

“Generate a Mermaid sequence diagram for the telecom prepaid recharge system using the API Gateway, authentication service, fraud detection service, and billing microservice.”

Within a matter of seconds, the AI generates the code in the required format, explaining the interaction between the systems.

The final diagram is not always the final design, however. This is merely the starting point that the architect can modify and change as desired.

Example 1- Requesting Chatgpt to desing a block diagram

Example 2 - Requesting to design a sequence diagram

2. Faster Brainstorming of Architecture Patterns

Brainstorming sessions in early stages of architecture design were traditionally carried out through open discussions and whiteboard exploration.

Though these discussions are still relevant, the application of AI can now include design suggestions at the beginning of the discussion.

Architects can pose questions such as:

“Given our requirement for high availability, PCI compliance, and 10 million transactions per day, can you suggest some architectures?”

Some possible answers that the AI can provide include:

It is essential to note that these are not the final answers to the problem; they are simply starting hypotheses that can be compared with the constraints of the environment.

Since the invent of AI tools the entire approach of the traditional brainstorming sessions changed to choosing the right approach between multiple design suggestions.

3. Faster Creation Architecture Presentation

Architecture communication is as important as architecture design.

Architects have historically invested considerable time in preparing a presentation to articulate the system design to stakeholders and leadership teams.

Preparation of architecture decks often involved writing on slides, copying diagrams, and preparing bullet points with comparison of approaches.

However, with the introduction of generative AI tools like NotebookLM and Gemini, it is now possible to accelerate this process.

For example, an architect can ask the following prompt:

“Create a 10-slide executive presentation to articulate this architecture to non-technical stakeholders.”

The AI can help with:

Though the results are far from perfect, it is a significant step towards reducing time and effort in architecture communication and allowing architects more time to focus on explaining the design trade-offs.

4. Faster Exploration of Design Trade-Offs

In my experience as an architect, almost every architecture decision ends up being a trade-off.

Architects often need to compare:

With AI tools, this process can be significantly sped up.

For example, an architect may want to compare:

“Kafka vs Amazon SQS for high throughput telecom transaction processing.”

In just a few seconds, the AI can produce a comparison that includes factors like latency, scalability, operational complexities, and cost considerations.

5. Visual Generation for Architecture Storytelling

Architecture is not only technical but also communicative.

Senior management tends to understand visual representations more easily than technical representations.

With the arrival of AI image generation tools, it is now possible for architects to generate conceptual images that explain architecture in a more interesting way.

These images can be used in architecture storytelling and can be used to show the interaction of systems, modernization strategies, and migration paths.

A Real Example: When AI Architecture Suggestions Fail

The recharge system handled millions of transactions every day, and the billing platform required a synchronous confirmation within a few seconds.

The preliminary architecture created by AI was a beautiful event-driven architecture where all the recharge operations would be performed asynchronously through messaging services. From a modern architecture perspective, this was a perfectly designed architecture.

The telecom billing platform we were integrating with was a legacy synchronous platform where immediate confirmation was required for each recharge transaction. The architecture designed by AI was technically sound but operationally incorrect for a legacy platform.

A powerful lesson learned:

AI can create architectures but Architecture must be validated by architects.

The Risks of Blindly Trusting AI Architecture

AI systems do not inherently understand the context of an organization & it’s current architecture.

AI systems don’t inherently understand organizational context such as legacy dependencies, regulatory requirements, enterprise architecture standards, operational maturity, or budget constraints.

An AI system would propose modern technologies such as a Kubernetes cluster, event-sourcing architecture, and global distributed architectures.

These are technically very good architectures.

However, they are not very practical if the operational maturity level of the organization is not high enough.

Architecture needs to be contextual of the current system.

Architects Must Still Evaluate ROI

Another key responsibility that architects have is the evaluation of the return on investment.

Architecture decisions have the following impacts:

Even if the AI is able to design complex solutions, it is not necessarily true that the solutions will be valuable to the business.

Some of the key questions that architects have to ask include:

These questions cannot be answered by the AI unless it is aware of the organizational objectives.

The Right Way to Use AI as a Co-Architect

The best way to design architectures with the help of AI is as follows:

In effect, the steps involved in this are quite simple. For instance, architects normally start by providing requirements to the AI, create an initial architecture draft, explore alternative patterns, and then analyze these options before finally aligning with enterprise architecture standards.

AI is used to speed up the generation of ideas. Architects must validate the decisions.

The Real Impact: Cognitive Offloading

The biggest impact of AI in the world of architecture is not diagram automation.

It is cognitive offloading.

Architects can now spend less time:

And can now spend more time on:

Architects handle the strategic layer.

Lessons from Using AI in Architecture

Having experimented with AI tools in architecture design, a few lessons are learned. First and foremost, AI is useful in creating initial drafts of architecture design but is not useful in overcoming legacy constraints and organizational context. In this regard, it is important to note that AI is useful in exploring design rather than being a tool of authority in creating design.

The Future of Architecture Is Augmented

Solution Architects are not becoming obsolete; instead, their role is adapting to the changes in the development of artificial intelligence systems. Today’s Solution Architect is no longer just a manual diagram creator, nor is he or she just an artifact builder, a requirement translator, or even an architectural validator; instead, he or she is becoming an AI-assisted system thinker. Instead of fighting AI, the most successful Solution Architect in the future will be one who uses AI to his or her advantage, using AI to explore and come up with architectural ideas. One thing, however, is certain: whereas AI can propose architectural patterns and design choices, AI can suggest architecture patterns, but deciding what should actually be built still requires experience and context and understanding the existing designs in place.

So, from my own point of view, the actual benefit of using AI is not in automatically designing systems but rather speeding up the initial phase of system architecture. While it is true that AI can offer architectural patterns and design alternatives, it is still necessary to have experience and context to understand what should be built and what already exists.