14 Lessons From a Developer Who Actually Built Something With AI (Not Just Vibed)

Try it out yourself.

That’s it. Simple.

I did this when Crypto development was going through it’s hype cycle. I soon realised that DAPPs were extremely niche and not that useful in real life.

I did it with Nextjs first came out. I built a real project and soon realised Nextjs was a great framework with some really positive features. Then the whole verson 14 came out with the apps folder and the massive over caching. Nextjs and myself soon parted ways. Why Vercel?!

It’s 2026 and AI is here with the biggest hype cycle I’ve seen in my lifetime as a developer.

It was time to try out a real life project. I didn’t want to fully vibe code it as I think that’s just naive and ultimately going to lead to a whole world of pain. I did however really want to try pushing the AI Agent and see where it’s limits were and where it performs really well and where it performs badly.

My project was something I have wanted to build for a long time. A language app especially to learn Czech (I live in Prague).

Czech is a slavic language and all the language apps I have used over the years tailor to MANY languages so don’t deal with the specifics of the Czech language.

I learnt quite a bit over the course of building this app. It took me about 4 weeks in my spare time. I worked on it pretty hard, even when heavily utilising AI agents to build out the code base. I was also using Laravel on the backend and Tanstack start on the frontend (My favourite stack!) so I had a really solid foundation to start with.

What I have produced is a good MVP — still quite far from a finished product. I don’t subscribe to the model of get something out as soon as possible. I think you need to put some care and attention even into the MVP — first impressions count.

Here’s what I learnt along the way.

14 Lessons From Using AI

1. Approach one feature at a time. If you do too much at once it quickly gets messy and something bad will go unnoticed. I sometimes let two agents run at the same time but that’s it. I have no idea how people are running 4 or 5 agents at once. Pure madness.
2. Always use plan mode. I‘ve learnt it’s much better to be more accurate than less in the prompts. It’s good to imagine you are talking to a real human. Would they be able to understand your request? If not, then rewrite it.
3. I definitely write more tests. It’s easy now and overtime the suite of tests you build up are really useful. As the app grows and the feature set grows, complexity does too. Complexity has a way of sneaking up on you.
4. Refactor as you go. It really pays to keep a close eye on coding patterns and structure. AI keeps referring back to the codebase so if it sees patterns you have implemented then it will follow that in future. If you don’t control as you go you are more likely to end up with a mess. The old adage of “Go slow to go fast” really works here.
5. Have the AI write some summary docs for specific things as you go. These act as good resources in the future when you want AI to work on something related. You’re able to bring the new AI up to speed fast. It saves time and tokens.
6. Remember: Every new session the UI knows NOTHING about your repo. It’s very weird but you constantly need to remind yourself of this. AI is like a very knowledgable developer who suffers with amnesia.
7. If it does something good, let it know. It stores this in memory for future work and can refer back to it. Like a dog, feed that reward cycle.
8. It’s still really hard to build a good app. I am realising the design choices of both the app and data are the hard part of building a successful app.
9. Always use Git. It makes it very easy to undo, redo, stash etc. AI can go off the rails sometimes and you want to be able to backtrack easily.
10. AI doesn’t produce great React code. Even with all the React code in the world it still doesn’t follow best practices. It prefers long component files that consist of many different components. I had to refactor this out many times into reusable components and follow the more “React” way of doing things.
11. Decide the tech stack before you implement anything. If you aren’t sure have a good discussion before with AI. Don’t let AI blindly choose for you.
12. Overall your software should be better. I am aiming to make good MVPs now rather than the “bear minimum” before I release. Your expectations of what’s good enough should increase with AI usage.
13. I like watching the AI output. It’s good to see it’s workflow, how it does things and the order it does things in. It’s all quite logical and straight forward really and you can learn. AI is a master of the terminal. CLI tools for the future.
14. You need to know software to understand what the AI is suggesting or implementing. Knowing about migrations, schemas, models, workers, services vs traits, testing etc. I couldn’t imaging doing what I did if I didn’t have all the knowledge and experience I have. You still need a really good mental model of what is going on.

Am I More Productive With AI?

Did AI Speed Me Up? Yes and no. It definitely allows me to write more code. However, I think now the bar for quality of software should be much higher. It’s easier to test your application. It’s easier to get through technically difficult moments. It’s easier to make decisions. But that shouldn’t mean we just ship more code. It should mean we make higher quality software.

Conclusion

AI when used in producing code is definitely a game changer. I have really tried to utilise it over the past month and honestly, I can’t see myself going back to hand coding everything.

The tools you use are important and it’s clear even the main tools are still going through quite a bit of iteration. The models you choose also have a big impact. Claude Sonnet 4.6 for me. I think it’s best to find one you like and stick to it, although keep exploring at the same time.

Planning first is crucial. It saves time ultimately. Keeping requests small is also really important. It’s about narrowing the scope of possibilities.

You have to have good mental models still. You can’t rely on AI to guide you on the bigger decisions, especially architecturally. By all means ask for its opinion but don’t follow it blindly. It may be quicker in the moment but ultimately it could cost you.

I really don’t like the multi agent approach. Two agents at once is more than enough to keep my cognitive load at maximum. This whole orchestration of agents — at this current time — just sounds stressful and I feel will lead you into a forest of mess and pain.

If you’re not using AI at the moment, I would strongly advise you to at least dip your toe in the water. It’s another tool at the end of the day and it can help you with some of those tasks you really dread — like writing extensive tests. However, keep a close eye on it. It just loves to go off the beaten path so it needs a tight leash. And don’t believe the hype. Never believe the hype. As in most things, the truth lays somewhere in the middle.

If you made it this far, here is the app I released: https://www.czech-learner.com/ — Still plenty to do but its up and working. If you want to learn Czech — give it a go!

Subscribe to My Weekly Updates on Medium!

Enjoyed This Post?

If you found this blog post helpful, why not stay updated with my latest content? Subscribe to receive email notifications every time I publish.

If you're feeling really generous you can buy me a coffee. (Btw, I really like coffee…)

What You'll Get

• Exciting Discoveries: Be the first to know about the latest tools and libraries.
• How-To Guides: Step-by-step articles to enhance your development skills.
• Opinion Pieces: Thought-provoking insights into the world of frontend development.

Join Our Community

I live in the vibrant city of Prague, Czech Republic, with my family. My blog is more than just articles; it's a community of like-minded developers who share a love for innovation and learning.

About me

I'm a passionate Frontend Developer specialising in React and TypeScript. My professional journey revolves around exploring and mastering new tools and libraries within the JavaScript ecosystem.

Check out my LinkedIn and Github if you are interested.