A Small Lesson I Learned While Building an AI Website

I recently started learning how to build an AI-powered website.

At first, I thought the main challenge would be writing prompts or calling an AI model. But after building a small demo, I realized the harder part was actually the messy stuff around the model.

I wanted to try image generation, then maybe video generation, then maybe different models for different use cases. Very quickly, my backend started to get ugly. Every model had its own API format, its own parameters, its own errors, and sometimes a totally different way of returning results.

I did not want to spend all my time writing wrappers for different AI providers.

That was when I tried Crun. It gave me a simpler way to call different AI models through one API. Instead of thinking about every model as a separate integration, I could treat AI generation as one service in my backend.

The flow became much cleaner:

User submits a prompt
→ backend creates a generation task
→ store the task ID
→ wait for the result
→ show it to the user

That solved one problem.

Then I ran into another one: user feedback.

Once a few people tried the site, suggestions came from everywhere. Some were in chat messages, some in email, some were just casual comments. I kept telling myself I would organize them later, but of course, “later” never really happened.

The annoying part was that I was actually improving the product, but users could not see it. From their side, it probably looked like nothing was happening.

So I started using FeedLog to collect feedback, organize repeated requests, and show a simple roadmap and changelog.

That helped a lot. Users could submit ideas in one place, and I could see which problems appeared again and again. When I shipped something, I could add it to the changelog instead of letting the update disappear silently.

The biggest thing I learned is:

Building an AI product is not just about connecting to a powerful model.

You also need a system that lets you change models easily, understand users clearly, and keep people updated.

For me, Crun made the AI integration side easier. FeedLog made the feedback side less chaotic.

Not a huge lesson, but a useful one.