I use AI coding tools for scaffolding, tests, docs, and boring boilerplate. Architecture and module boundaries stay human-led. The way to keep a growing app clean is to break it into small vertical slices, add guardrails, and ship faster without letting the codebase drift.
The Real Problem
Large applications slow down as complexity grows. Code reviews take longer as dependencies multiply. Once the codebase exceeds the model context window, AI becomes unreliable unless the system is organized around clear boundaries.
I worked on a similar application and learned how to keep a mid-to-large system clean as it scales.
The Core Approach
Break the application into small, focused modules such as Auth, Payments, Notifications, and Search. Give each module a clear contract through OpenAPI, AsyncAPI, or typed interfaces.
Each module needs explicit boundaries: what belongs inside it, how it communicates with other modules, who owns the data, how consistency works, which cross-cutting guardrails apply, how testing is handled, how security testing is handled, how delivery works, and where AI should not be used.
Where AI Helps
AI is useful for scaffolding controllers, services, repositories, DTOs, validation pipelines, logging, error handling, configuration, dependency injection, authorization setup, unit tests, contract tests, test data, basic security tests, READMEs, API examples, module guides, renaming, restructuring, and cleanup.
Where AI Hurts
AI should not own architecture decisions, complex business logic, security-critical components, or anything involving sensitive code. It is an assistant, not the architect.
Pros
The upside is speed, consistency, cleaner naming, better source-control hygiene with structured prompts, faster documentation, and less time wasted on syntax issues.
Cons
The risks are quality drift, contract drift, shallow tests, privacy and licensing concerns, and architecture erosion. Code can look right while being unsafe. Modules can misalign if contracts are not enforced. AI often defaults to happy-path tests.
Without reviews, expertise, contract tests, and security checks, you will hit a wall.
My Workflow
Start with strong context and knowledge files: a clean system prompt, templates, and cleanup routines. Write one-page module ADRs for Users, Auth, Payments, Notifications, and Search. Use a contracts-first approach with OpenAPI and AsyncAPI.
Then scaffold with AI one module at a time. Add guardrails through coding standards, formatting, security scans, and contract tests. Ship vertical slices: API, UI, and tests per module. Document continuously with READMEs, API examples, and ADRs.
Final Thought
AI accelerates productivity. Architecture protects the product. Use both intentionally to build fast and build clean.