Building Software with AI: A Developer's Honest Experience
For the past two years I've used AI-assisted tools in nearly every project — building mobile apps with Flutter and web apps with React alike. In this post I want to share my honest experience, without the marketing language.
The real gain is focus, not speed
The greatest value of AI tools is that they take over boilerplate and repetitive work. A form validation, an API client, a model class… Instead of typing these by hand, I describe and review them. The time saved goes into the actual problem and architectural decisions.
Where AI genuinely makes a difference in my daily workflow:
- Repetitive structures: model classes, DTO mappings, form validations
- Writing tests: quickly enumerating edge cases for an existing function
- Refactoring: getting a first draft when splitting a large file into meaningful parts
- New libraries: starting from a working example instead of getting lost in docs
Where does it mislead?
Mostly by sounding confident when it isn't. It may use an API method that doesn't exist, or write code against an outdated library version.
A compile error never costs me anything; the real danger is code that compiles but behaves incorrectly.
That's why I write the tests myself for critical business logic and run AI-generated code through them. Shipping code you haven't read is like deploying borrowed code without testing it.
Responsibility increases, not decreases
I can say this clearly: using AI doesn't reduce a developer's code-review responsibility — it increases it. When I open a pull request, I need to account for every line. "The AI wrote it" is never an excuse.
Three rules I follow in practice
- Architecture first, code second. I don't ask for what I haven't thought through; I set the direction.
- Read every line. Code I don't understand doesn't enter the project.
- Tests are mandatory in critical flows. Payments, auth, data writes — I write those tests myself.
Conclusion
AI is not a technology that replaces developers; it's a developer's lever. And a lever only works in the hands of someone who knows the direction. That direction is still engineering: understanding requirements, designing the right architecture, anticipating edge cases, and owning every line you ship.