How AI Agents Changed My Mind About Vertical Slices in Software Architecture
An insightful look at why working with multiple AI agents shifted a developer's preference from traditional layered architecture to vertical slices.
Stock photo for illustration only, not from the actual event
- Traditional layered architecture groups code by technical layers like controllers and services.
- Multiple AI agents working simultaneously cause layered code features to scatter everywhere.
- Vertical slices give each AI agent total ownership of a cohesive feature folder.
- Architectures requiring less cross-coordination perform better for both humans and AI.
For years, my go-to choice was a traditional layered architecture. The standard layout featuring Controllers, Services, Repositories, DTOs, and Models made codebase navigation straightforward because if I needed a repository, I knew precisely where to look.
Stock photo for illustration only, not from the actual event
Everything shifted once I started collaborating with multiple AI agents simultaneously, which abruptly altered the optimization target. Imagine tasking one agent with implementing Shipping while another handles Billing. Under a layered approach, both agents inevitably end up touching controllers, services, repositories, tests, and dependency injection despite working on completely separate features. Consequently, implementing a single feature becomes scattered across the entire project, and code reviews turn into a fragmented chore.
A much more effective model is organizing code around features rather than technical layers. Shipping and Billing cease to be scattered concepts that must be pieced together from multiple directories, transforming instead into visible segments of the project structure:
- Features/Shipping/Controllers, Services, Repositories, DTOs
- Features/Billing/Controllers, Services, Repositories, DTOs
Now, each agent owns an entire feature rather than a few isolated files spread across the solution. Reviewing changes simply means opening Features/Shipping instead of hunting through controllers, services, repositories, and tests scattered in different locations. This ownership boundary naturally extends to verification as well; if an implementation belongs to a feature, the tests proving it works should live right beside it.
It is worth noting that vertical slices were originally created to maximize code cohesion rather than to accommodate AI development. However, agentic workflows act as a magnifying glass for architectural friction, proving that software agents face the exact same coordination challenges as human teams. Minimizing the inter-module communication required by your design directly enhances productivity across the board.
I previously argued that event-driven architectures pair well with AI because they render behavior observable. I now view vertical slices as the structural equivalent of that principle. They make feature ownership immediately transparent, instantly revealing which agent governs a specific domain and where its corresponding test suites reside.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment