How genAI is changing the work of a software architect

A software architect is a professional who translates business requirements into technical design. In smaller and medium-sized companies, this role is often an overarching one, with the architect acting as a "construction planner". He or she is responsible for translating the brief into a specification, selecting technologies, designing databases, dividing the system into modules and determining the principles of their compatibility. In corporations, this activity is often divided among several specialized roles, but the principle remains the same: the architect defines the plans by which the development team builds the final solution.

The comparison to the construction industry is straightforward in this case. The architect designs the house, prepares the documentation and divides the tasks between the builders and craftsmen. In IT, it's similar, where the design, specification and plan become the basis by which senior developers and programmers divide the work between the frontend, backend and other parts of the development.

The design and development process

There are multiple approaches to development, but whether a waterfall or agile methodology is used, there is always a design phase. In this phase, after requirements gathering, a software specification is created that includes functional and non-functional requirements, acceptance criteria, and definition of project goals. Based on the specification, test scenarios, unit tests for individual components, and integration tests are created to verify the functionality of the system as a whole. These tests are often directly linked to client acceptance.

What is the waterfall development method?

The Waterfall Model is a classic, linear approach to software development. In it, a project proceeds step by step from analysis through design, implementation, testing, and deployment. Each phase must be completely finished before moving on. This model is based on the principle that the plan is known in advance and changes during development are minimal.

The key output of the architect's work is the design of the data model and the description of the processes. The data model defines how data will be structured and stored, while the process model defines how users will interact with the system and what behavioral scenarios need to be covered. Typically, these processes are modelled using UML diagrams, in particular use case diagrams and activity diagrams.

What is the agile development method?

Agile development, on the other hand, is an iterative and flexible approach that breaks the entire project into smaller parts, called sprints. Each sprint delivers a functional unit that can be immediately tested and improved based on feedback. Instead of a long-term plan, agile teams focus on rapid delivery and adapting to changes that come from the client or the market. Agile development requires close collaboration between developers, analysts, designers and the client, who is involved in the process continuously, not at the end.

How AI enters the process

Today, artificial intelligence is becoming a tool that is fundamentally changing the efficiency of an architect's work. The first major contribution is in database design. The architect prepares the general structure of tables and relationships, and AI helps him consult them for correctness and completeness. For example, when designing a delivery system, requirements for address points, logistic units and time windows can be described. The AI then generates table and session designs, which the architect validates and completes. For us at TRITON IT, this assisted process reduces the workload of database design by approximately 50%.

The next step is to use AI to assist in the design of the overall system architecture. The architect presents the AI with textual descriptions of processes (use cases) and a data model, giving the AI a complete picture of the application. Thanks to its extensive knowledge of technologies and best practices, it is able to suggest possible architectural options, recommend specific technologies and have a dialogue about their suitability. In practice, this means that AI can suggest, for example, using the Kafka message broker instead of RabbitMQ, recommend a specific programming language, or split the backend and frontend into logical modules.

However, the key point is that AI is not an authority, but a partner. The architect must judge the proposals, reject inappropriate options, and assert their own intent in light of the larger context of the project. It is in this dialogue that the greatest strength of the synergy between the information architect and AI is demonstrated - the AI provides the broad knowledge base and generates the options, the architect provides the critical evaluation and final decision.

Saving work and generating documentation

The biggest savings come in the final phase, when AI can generate UML diagrams, documentation and work estimates. Once the design is complete, the model has a complete description of the database and processes and can generate diagrams of classes, components or use cases. Similarly, it is able to generate work estimates for specific development teams according to their experience and composition, including schedules and budgets. Even other activities, such as developer work, can now be handled assisted with AI.

If designing a database meant saving around 50% of the time, then working with architecture reduces our time consumption by around 70% thanks to AI. In the area of documentation, traditionally one of the most time-consuming and administratively burdensome tasks, the savings are even more significant, in some cases reaching up to 90% at TRITON IT.

Utopia vs. reality

Nowadays, with the rapid emergence of GenAI, new so-called application generators are constantly emerging, which in their advertisements entice that they will produce the entire application themselves in a matter of minutes. At first glance, this seems like a revolution that can completely replace development teams and software architects. But the reality is considerably more sobering.

While generative AI can fundamentally speed up the analytical, design and documentation parts of development, it still cannot replace the human decision-making behind the stability and security of the entire system. AI can be a great assistant to help architects and developers design databases, code structure or processes, but it remains a human task to create a complex, stable and sustainable system. For now, AI is only his strong right hand, not a replacement.

The growing importance of standardised text formats

Large Language Models (LLMs) are tools that primarily work with text. So at first glance, they may not seem ideal for creating visual elements such as UML or ER diagrams. However, this is where the power of text-based graphical formats such as Markdown, DPML (database description languages) or PlantUML comes in. These standardised formats make it possible to describe even complex structures in purely textual terms, which LLMs understand perfectly.

As a result, they can work with the architect to generate accurate and consistent outputs that can then be easily visualised using specialised tools. This approach greatly speeds up the work, where instead of drawing manually, most designs are created with assistance and the resulting diagrams are simply displayed, exported or printed. At TRITON IT, this process saves us up to 90% of the time.

The new role of the architect

Artificial intelligence is therefore not moving the software architect out of the game, but changing his role. The routine work of drawing diagrams and writing documentation becomes a curator of dialogue with AI, who must understand all aspects of the design, ask the right questions and critically evaluate the answers. The architect remains the one who holds the direction of the project, but instead of manually preparing all the documents, the focus is on the quality of the brief and the decision-making process.

The result is a major increase in efficiency, as well as higher quality documentation that is produced in less time. In this way, AI becomes not only an assistant but also a catalyst to change the work of the architect, much like CAD tools have changed the work of architects in the construction industry.