This diary entry is a little different. Rather than focusing on a new feature or technical milestone, it’s about the process itself — specifically, what it’s been like building ClubAssemble with the help of an AI coding assistant.
From day one, ClubAssemble has been developed using AI assistance, primarily through Google’s Gemini service integrated within Firebase Studio, supported by ChatGPT for broader problem-solving, documentation, and design discussions.
The Upside — Speed, Insight, and Experimentation
Overall, using AI in development has been an overwhelmingly positive experience. It has allowed us to prototype new ideas incredibly quickly, explore different architectural options, and test functionality that would have taken days to build manually.
The assistant responds instantly, and it’s particularly good at helping identify bottlenecks or moments when our code has drifted away from best practice. In those moments, it’s like having an extra pair of eyes — ones that never get tired and can instantly reference an entire library of Firebase and React.js patterns.
It has also helped with experimentation. When we’ve wanted to explore a new design pattern or API structure, the AI can generate a baseline implementation in seconds, allowing us to test feasibility without a huge time investment.
In short, it’s made us faster, more flexible, and more creative.
The Downside — Context Blindness
However… AI coding assistance isn’t perfect, and learning to work with it has been a skill in itself. One of the main challenges is what we call context blindness — the AI’s tendency to make decisions that make sense within the scope of a single prompt but don’t align with the broader project architecture.
You can spend weeks refining your data and security models, carefully defining standards, validation, and naming conventions… only for the AI to “helpfully” rewrite entire sections of code, ignoring all of that groundwork.
This usually happens when you give what seems like a simple prompt, such as:
“The document update in form X is generating a validation error. Please take a look.”
Instead of just diagnosing the issue, the assistant might decide to “fix” it by editing multiple files, changing variable structures, and reworking components that were perfectly fine — leaving you with a project that no longer compiles or fits your design.
We’ve learned to write much more precise prompts. For example:
“The document update in form X is generating a validation error. Please identify the root cause only. Don’t make any code changes, but create a short report describing the issue. Ensure your analysis respects the existing data and security models and follows Firebase and React.js best practices.”
This kind of prompting feels overly formal at first, but it’s crucial. You’re not just asking for help — you’re managing an assistant that needs structure and context to behave predictably.
The Importance of Regular Backups
The second big lesson has been around version control. When the AI makes a breaking change, reverting it is not as simple as asking it to “undo”. Unlike a human developer, the assistant doesn’t revert from a previous version in your repository or roll back to a known state — it tries to reconstruct what it thinks it did and then “fix” it.
In our experience, this rarely works. Within minutes, you can go from a clean, working codebase to a completely inconsistent mess — missing functions, broken dependencies, invalid security rules, and models that no longer make sense.
So our rule is simple: Back up your code regularly. We now branch and push to our Git repository multiple times per day. It’s the safety net that keeps AI-assisted development sustainable.
Where AI Really Shines
Despite its quirks, AI has been a game changer in many areas — especially the smaller, repetitive or creative tasks. Simple code changes, refactoring, and debugging are far faster. And when it comes to visual creativity, it’s been a revelation.
The ClubAssemble logo, much of the landing page imagery, and even early colour palette concepts were developed through AI-generated ideas. With a few simple prompts, we were able to explore dozens of visual directions in a fraction of the time it would have taken manually.
It’s in these moments — when the AI becomes a collaborator rather than a co-author — that the technology feels truly empowering.
A Learning Curve Worth Taking
Working with AI has been a learning curve, no question. We’ve had to adjust our mindset, refine how we communicate, and learn to think more clearly about how we ask for help. Context really is everything.
But in the end, AI has made us better developers — faster, more deliberate, and more aware of our own design decisions.
It’s not a magic bullet, and it’s certainly not infallible, but as a creative and technical partner, it’s transformed the way we build.