Three months of building with AI every day gives you opinions. And they're wonderful. You don't need to fear them and I'll tell you about it in the end.
Here are mine.
What I Actually Used (And What Each One Is Good For)
Claude Code
The one I'd bring to a real job. When I needed to build actual features, authentication, QR scanning, Supabase wiring, security auditing, Claude Code delivered. It's not magic. It still needs a human to catch its blind spots and make the architecture calls it genuinely cannot make. But for shipping real things fast, nothing else I used came close. The MCP integrations with Supabase and Vercel were the specific parts that changed how I worked.
However, a single coding agent will never be fun. And this is where we introduce my Number 2.
Gemini / Gemini CLI
I treated this as a second brain where I need some prompting refinements when I wasn't sure about a direction. Second, it is better in understanding UI/UX design, especially tying it with Figma MCP versus Claude. Third, it is a capable fallback if you hit your limits on Claude Code. And finally, you can make them talk together. Good brain. Different muscles than Claude Code.
Google Flow / Veo 3.1
Genuinely impressive, within limits. The 8-second cap and visual consistency issues are real friction. But for AI-generated video on a free tier, the output quality surprised me every time it worked. Good for short-form and iterative content.
But the catch here is it's not for you if anything that needs a clean, continuous long take. Luckily if you have a Google AI Pro Subscription, you can perform multiple extensions as you like. But it will cause you inconsistent voices, artifacts, and unexpected elements beyond your prompt. Opting for the Veo 3.1 Pro can cost you many tokens but produces a higher-quality output.
NotebookLM
If you have scattered documents, event reports, and research spread across eleven regional contexts, NotebookLM turns that noise into something you can actually navigate. The auto-generated podcast and wiki features are better than they have any right to be. Tech communities doing knowledge management should pay more attention to this one.
This helps you worry less if your ideas are all over the place. Compile everything and NotebookLM helps you remember things if you need. It can even make you presentations, mind maps, quizzes, and even an interactive podcast! Your learning experience, maximized!
Google AI Studio
Very powerful for creating quick applications with GCP. Initially, I used it for prototyping DEVCON+ before fully transitioning to Claude Code. If you want instant applications, ready for deployment and can integrate fully with Google's AI suite, this platform is for you.
Overall among these AI-powered platforms I used, Claude Code is undisputed in terms of coding ability. Gemini is also smart in understanding UI and UX design. In fact, these AI-powered tools boosts my productivity and will serve practical purposes for everyone whether it could be coding, prototyping, or generating ideas.
Am I Cooked (By AI)?
No. But here's what I actually think after building with it for three months straight:
AI is the fastest, most knowledgeable, most tireless junior developer that has ever existed. It writes anything without complaining. It never needs a coffee break. It will do what you ask at a speed that genuinely changes what's possible in a workday.
But it has no taste. And they suck. No judgment about why you're building something. No sense of whether the thing being built should exist at all.
Every time I directed Claude Code to implement a feature, I still had to decide which feature was worth building and why. Every time it generated a security audit, I still had to decide what to prioritize. Every time Charity appeared on screen, I still had to decide whether that was even the right creative call.
The people who are going to struggle aren't bad programmers or bad designers. They're the people who built a career around being the person who executes tasks. The people who will be fine (and some) are the people who position themselves as the person who decides what tasks are worth doing and what "done" looks like.
My actual take: AI will get me to maximum output in less time, if I learn to use it well. And "learning to use it well" is a real, ongoing, full-time effort, one most people are underestimating right now, badly.
And of course, my training data is built up by real-world experience. Upgrading it with AI makes me an unexpected superhero, turning my thoughts into something tangible.