What AI Can Build on Its Own, and What Still Requires a Developer

When I decided to take up coding myself at 61 and build formspace.design, without a traditional IT education, I understood that I couldn't do without a powerful tool. And that tool became artificial intelligence. Today, I can confidently say: AI is capable of independently generating a significant portion of the code, but fundamental architecture, debugging, and understanding complex business logic are still the responsibility of a human developer.

My Journey to Code with AI

My background includes teaching, 90s business, and landscape design since 1997. I've seen how technology evolved, but I was never a professional programmer myself. At "Dom i Sad" (House and Garden), where I was a director, I assigned tasks to programmers but didn't write code. However, when I started creating formspace.design, I realized I wanted to build a product that I myself lacked in practice, and to do it as efficiently as possible, including with the help of AI. It was a real challenge, but also a huge opportunity. I started learning, using AI as my main mentor and co-author.

What AI Can "Build" on Its Own

AI is an incredible generator of boilerplate code and drafts. It excels at tasks that require a large volume of routine work but not a deep understanding of the context. Here are a few examples from my practice:

• Boilerplate code: Creating project structure, boilerplate for REST APIs, basic CRUD operations (create, read, update, delete) for data models. AI can generate hundreds of lines of such code in minutes, saving hours or even days. For example, for a new module in formspace.design related to plant accounting, AI quickly drafted all necessary endpoints and database schemas.
• Basic UI components: Generating simple user interfaces, buttons, forms, modal windows. I can ask it to "create a form for adding a user with name, surname, email fields" — and get a ready-made skeleton that only needs styling and integration.
• Code translation or refactoring: AI handles adapting existing code snippets to new standards or frameworks quite well, as well as basic refactoring to improve readability. For example, rewriting old JavaScript functions to more modern ES6 syntax.
• Writing tests: For simple functions, AI can generate unit tests, which is very useful for maintaining code quality, especially when working on a project alone.

Where AI is Still Useless

Despite all its capabilities, AI has significant limitations, especially when it comes to creating a complete, complex product:

• System architecture: AI cannot design the overall application architecture, choose the right technologies for the stack, determine how microservices or modules will interact, or how to scale the system. This requires strategic thinking and a deep understanding of future product needs. For example, the choice between a monolith and microservices for formspace.design was my decision, based on the long-term development perspective.
• Complex business logic: When it comes to logic that accounts for many nuances, exceptions, and specific rules, AI often makes mistakes. It doesn't understand "why" something should work a certain way and not another. For example, the algorithm for zoning a plot in formspace.design, which considers insolation, terrain, user preferences, and dozens of other factors, cannot be entirely entrusted to AI.
• Performance optimization: AI can suggest general optimization methods, but deep profiling, identifying bottlenecks, and fine-tuning database queries or algorithms are tasks for an experienced developer.
• Unique UX/UI: Creating a truly intuitive and pleasant user experience that distinguishes a product from competitors requires human vision and empathy. AI generates standard solutions, but not the "soul" of the interface.

Critical Stages Requiring Human Involvement

My experience shows that there are several critical stages in development where a human developer is indispensable:

1. Planning and design: This is the very first and most important step. A human must define exactly what needs to be created, for whom, how it will work, and what problems it will solve. AI can help with ideas, but not replace strategic vision. It is at this stage that I make maximum use of my experience from business pain to a working module with AI.
2. Debugging and testing complex code: AI can generate code, but it can also generate errors. Especially in complex systems where multiple modules interact with each other, finding and fixing errors, understanding their causes — this is purely human work. I sometimes spend more time debugging AI-generated code than writing a simple module manually.
3. Refactoring and maintenance: Maintaining clean and readable code, adapting it to changing requirements, eliminating technical debt — this is an ongoing process that requires human evaluation and decision-making. AI can suggest refactoring options, but choosing the best one and applying it correctly is up to me.
4. Security: Designing and implementing robust security systems requires deep knowledge of vulnerabilities and best practices. AI can help with generating authentication code, but the overall strategy for data protection and attack prevention is a task for an expert.

The Role of Architecture and System Thinking

AI is a tool that enhances a developer's capabilities, but does not replace them. I often compare it to an architect and a builder. AI can be an excellent builder who quickly erects walls according to blueprints. But who creates these blueprints? Who plans the foundation, utilities, the overall layout of the building to make it functional, safe, and beautiful? That's the architect — a person with systemic thinking and vision.

In product development, the human is this architect. As the developer of formspace.design, I am responsible for ensuring that all parts of the system harmonize, that the product solves real user problems, and that it is resilient to changes. AI helps me translate my ideas into code faster, but the ideas and their structuring are my task.

Common Mistakes When Working with an AI Developer

Working with AI as a co-developer, I've identified several common mistakes that I try to avoid:

• Insufficiently precise prompts: If the request to AI is too general, the result will be the same — general and unsuitable. You need to detail the task as much as possible, provide examples, specify the technologies used, and data structure. My experience with step-by-step prompt creation for garden visualization has been very useful here too.
• Blind trust in code: You should never simply copy and paste AI-generated code without thorough review. There is always a possibility of errors, outdated methods, or suboptimal solutions. I always review every line.
• Lack of iterations: Development with AI is a dialogue. If the first response isn't perfect, you need to refine the request, ask to change details, and ask clarifying questions. It's not a one-time request, but a series of iterations.
• Ignoring project context: AI doesn't know your entire project, its history, specific dependencies. It generates code in a vacuum. A human must be the one to integrate this code into the existing system, considering all its peculiarities.

How I Use AI in formspace.design

In the process of creating formspace.design, I actively apply AI at different stages, but always under my careful control. For example, when I develop the plot zoning planning module (module 1.x), AI helps me generate initial templates for frontend components responsible for rendering and interacting with objects on the plan. I can ask it to create a basic component for dragging elements or for resizing them, which significantly speeds up the start of work.

When I move to more complex functions, such as the AI-concept module (2.1), AI acts as a "brainstorming partner." I describe the desired result — for example, "create a Japanese-style garden concept considering a flat 10-acre plot and minimal maintenance" — and AI generates ideas, possible elements, and palettes. Then I use these ideas to formulate prompts for visualization (module 3.1 "design by photo"), and the AI engine itself, which generates these concepts, I write independently, using it to speed up the writing of individual functions related to text and data processing. This allows me to focus on unique business logic and user experience, entrusting routine tasks to AI.

FAQ

Can development be fully automated with AI?

Fully automating product development, especially complex and innovative ones, is not yet possible. AI can be a powerful tool for accelerating routine tasks and generating drafts, but strategic planning, deep understanding of business logic, debugging critical errors, and ensuring security remain the prerogative of the human developer. The human sets the direction and controls quality.

What type of projects are best to start using AI in development?

It's best to start with small, well-defined tasks, such as generating boilerplate code, writing unit tests for simple functions, translating code between languages or frameworks. This allows you to master interaction with AI, understand its strengths and weaknesses, without risking the integrity of a large project. Gradually, you can move on to more complex tasks.

What skills become most important for a developer in the age of AI?

In the age of AI, critical skills include systemic thinking, the ability to formulate precise and detailed prompts, and the capacity to critically evaluate and debug generated code. A deep understanding of the subject matter and architectural principles is also important to guide AI and integrate its results into the overall system. The developer becomes an architect and a conductor.

Will AI replace programmers in the future?

My personal experience shows that AI will not replace programmers in the foreseeable future, but it will significantly change their role. Routine tasks will be automated, and the focus will shift to high-level design, solving complex problems, project management, and a creative approach to product creation. Programmers who master working with AI will become significantly more productive and in demand.

How fast does AI generate code compared to a human?

AI can generate hundreds of lines of boilerplate code in seconds or minutes, which is tens or even hundreds of times faster than a human. However, the speed of generation is not equal to the speed of creating a finished, working, and error-free product. A human will still need time to review, debug, integrate, and adapt this code to the project's specifics.

What are the risks associated with using AI in development?

The main risks include generating suboptimal or erroneous code, introducing security vulnerabilities, and the difficulty of debugging code you didn't write yourself. There's also the risk of over-reliance on AI, which can diminish one's own problem-solving skills. It's important to always maintain critical thinking and deeply understand what AI is doing.

How does AI help in creating unique features for formspace.design?

In formspace.design, AI helps me not only with routine coding but also in forming concepts that then form the basis of unique modules, such as the AI-concept (2.1) or design by photo (3.1). It allows me to quickly prototype ideas, test hypotheses, and get drafts for further development, thereby accelerating the process of innovation and refining functionality to meet user needs. I use it as an intelligent assistant for generating ideas and initial implementations.

From Idea to Implementation with AI

My journey of creating formspace.design is a vivid example of how AI becomes an indispensable assistant, but not an omnipotent creator. It significantly speeds up the process, but the human is always responsible for the strategy, architecture, and final result. If you, like me, are full of ideas and want to bring them to life, don't be afraid to explore AI. It will open up new possibilities for you, but remember: the most valuable assets are your vision and experience.