Today something interesting happened.
Actually, two very different things — both involving AI.

For the first week of this project, ChatGPT actively kept me away from Claude Code.

That wasn’t an accident. The focus was on slowing things down: thinking through product ideas, building structure, and understanding what I actually want to ship — before adding another powerful execution tool to the mix.

At the time, that felt restrictive.
Looking back, it was probably the right call.

Grinding through ideas slowly has already proven valuable, and I’m confident it will pay off going forward.

A deliberate decision: starting to use Claude Code

Today, I made a deliberate decision to start using Claude Code.

Not because it was recommended to me, and not just because of the hype — but because it’s a tool I’ve been curious about for a long time. In many ways, it was this tool that originally pushed me to start this blog in the first place.

Now, it felt like the right time.

Experimenting with a powerful tool

To learn Claude Code properly, I decided to run a small experiment.
Nothing business-critical. Nothing polished. Just a real test.

I chose to build a Dominion game.

Dominion is a game that’s close to me, and one I almost one-shot this summer — which made it a good benchmark.

I asked Claude Code to build it.

It did. Immediately.

After a few clarifying questions, it produced a complete web app.
UI, game logic, structure — all in place. It consumed a noticeable amount of tokens, which was expected, and honestly useful. It reinforced the need to pause, reflect, and not treat speed as something to mindlessly maximize.

Where the real strength showed up

Then I pushed it further.

I asked Claude Code to take care of the surrounding work:

• create and structure a new project
• publish it to Git
• make it accessible online
• so I could share a link with friends

That worked immediately.

A lot of videos talk about Claude Code’s strength in handling large codebases and working across many files. That’s true — but for me, the bigger strength showed up elsewhere.

It’s how naturally it works with Git, how easily it creates new projects, publishes sites, and handles the full lifecycle around the code — not just the code itself. The Chrome integration also plays an important role here, especially when moving between local work and online tools.

This stood in contrast to my previous setup, which became surprisingly time-consuming when working inside certain web-based tools. One example is MailerLite, which is still new to me. Its workflow didn’t come naturally, and configuring more advanced setups required a lot of manual effort.

It will be interesting to see whether Claude Code can eventually take on that kind of work as well — including building and managing more advanced configurations like the ones we set up for KlaraPaket.

Two versions, same conclusion

To stress-test frontend capability, I asked Claude Code to build two versions of the same game.

One was intentionally retro — more text-based and isometric in feel.
The other was a modern take, with a graphical UI and emoji-based cards.

Both versions were generated quickly and confidently, and both are fully playable:

Version	Style	Demo
Retro	Isometric / text-based	Open demo
Modern	Graphical GUI with emoji cards	Open demo

What stood out wasn’t just the speed, but how cleanly the two styles diverged while sharing the same underlying logic. It made it easy to compare approaches — not just visually, but structurally.

The other side of AI

At the same time, I ran into the opposite experience.

I noticed that the logo for this blog wasn’t correct.
Not the image I had generated, but a strangely drawn icon appearing in tabs and previews.

I asked Cursor to fix it.

It didn’t work.

I tried a new agent.

Still no result.

I then asked ChatGPT to project-lead the fix.
And the contrast was striking.

Many steps.
A lot of reasoning.
Very slow progress.

I ended up spending more time changing a single icon than I did building two board games.

Conclusion: tools, timing, and intent

I’m still very happy with the time invested in ChatGPT + Cursor.
Even if it’s not the fastest way to write code, it is one of the fastest ways I know to move toward a shipped product.

And that difference matters.

Claude Code excels at speed and execution.
ChatGPT + Cursor excels at structure, direction, and product thinking.

What became clear today is this:

Building a company and a product requires more than fast, correct code.

It requires direction.
Clear decisions.
Intentional tool choices.

I’ll be using Claude Code a lot going forward.
But the bigger picture — and the machine we’re building — matters even more.

And that’s exactly what makes this phase interesting.