Over the last 1+ year, I’ve been actively experimenting with GenAI tools for application and system development. It’s been exciting, humbling, and sometimes frustrating — and I’d like to share what I’ve learned.
At first glance, GenAI makes development look effortless. Building a simple to-do list app or a city weather widget is almost trivial: prompt → generate → run.
But the deeper I went, the clearer it became: once you move from toy apps to real business systems, the complexity rises dramatically.
The Plane vs. Car Analogy
Here’s how I think about it from my own journey:
A car is straightforward — most people with a license can drive it.
A plane is much more advanced, packed with features. But that same sophistication means not everyone can just take off and land it safely.
GenAI is like that plane. Powerful, multi-functional, but it demands skill and judgment from the pilot — in this case, the developer or architect.
And just like a real copilot, GenAI sometimes makes silly mistakes. The onus is on you, the pilot, to catch them and make sure your copilot adheres to the rules.
What GenAI Can Do (From My Experience)
In practice, I’ve seen GenAI tools step into multiple roles:
Acting like a business analyst clarifying requirements.
Suggesting UX/UI prototypes.
Generating backend and database code.
Even supporting Git workflows and deployment pipelines.
That’s an incredible co-pilot. But…
The Catch: It Depends on You
Whether a production-ready system emerges depends heavily on the user’s maturity. From my own work, I’ve found success when I:
Applied solid architecture and design patterns.
Periodically reviewed outputs to avoid hidden technical debt.
Fed the tool evolving context and updated its “do’s and don’ts.”
Never exposed production data directly.
Made incremental progress, with the discipline to revert when things broke.
GenAI will happily generate — but it’s up to us to steer it, supervise it, and correct its mistakes.
The Art of Delegation (AI Included)
Working with GenAI reminded me a lot of managing a team:
Delegate tasks, not the whole project.
Give detailed instructions when precision matters.
Rely on SOPs where repeatability works.
Do periodic checks to ensure things don’t drift.
Step in when the AI (like a team member) isn’t able to deliver.
It reinforced for me that AI is a partner, not a replacement.
My Takeaway After 1+ Year
Yes, I’ve been able to build production-grade systems with GenAI — but only when I approached it as a hands-on architect and mentor, not a passive user.
👉 GenAI doesn’t replace expertise. It amplifies it.
And just like flying a plane, the outcome depends on the skill of the pilot.
💡 I’d love to hear from others:
Have you tried building complex systems with GenAI?
Did it feel like driving a car, or flying a plane with a slightly clumsy copilot?
First published on LinkedIn.