From free tier to dependency: how Claude got me

This isn't a tutorial. This is a confession.

Phase 1: "Hey, let me try it"

Claude free tier. Free. Why not. I opened the chat, typed something like "write me a function that sorts an array," and got a perfect answer. OK, fine. Then I tried something harder. And something harder still. And then —

Limit.

One hour of work. Three hours of waiting. Seriously. You're sitting at your computer, you're in flow, your brain is firing on all cylinders, and suddenly — "You've reached your message limit. Please try again later."

Insane. But also pretty decent marketing when you think about it. Because in that moment, you know what the tool can do. And you want more.

Phase 2: "$20? Fine, here you go"

Two days. I resisted for two days. Two days telling myself "I can do this without it, I've got GPT anyway." Two days waiting out limits.

On the third day, I bought Claude Pro.

And the world changed.

Suddenly no waiting. Suddenly Opus. Suddenly extended thinking, where you see the model reason. Suddenly Claude Code in the terminal. And suddenly — ideas.

Because when nothing is impossible, you start thinking differently.

Phase 3: Ducks on the Vltava

Yes. You read that right. One of my first "serious" projects with Claude was a system for monitoring the number of ducks on the Vltava river.

Why? Because it occurred to me, and I wanted to ask: "Is it doable?" And Claude said yes. Computer vision, object detection, stream from cameras, dashboard with real-time duck count...

Technically it worked. We built a prototype. The model recognized ducks with surprising accuracy.

The problem? Camera quality. The public cameras on the Vltava have 2008-era resolution. A duck on a 240p stream looks like a pixel. Can't be distinguished from a water reflection.

But you know what? Doesn't matter. The important thing is that "it can't be done" doesn't exist. The limits aren't in the technology — they're in your head. And occasionally in the resolution of city cameras.

Phase 4: Ideas stop being silly

After the ducks came more realistic things:

Publishing planner. An internal tool for content management — tracking posts, where and how they were published, which project they tie to. Nothing sexy, but it saves hours every month and I finally have an overview of what came out where.

Education portal. A full-blown system for online teaching — quiz series for students, learning progress tracking, exam mode, presentations for classes, a lecturer guide, cheat sheets and other nice touches. PHP, hosting on Vercel, Supabase database. This is where I first encountered this stack, and Claude walked me through the whole thing — "here's the database, here's the hosting, here's the CDN." Like having a mentor who knows everything and has infinite patience. Maybe it'll grow into a universal online learning system over time — we'll see.

Phase 5: Projects grow. And so do the demands.

The more I did, the more I needed. Not more AI — more of everything. Better hardware. Better tools. Better workflow.

I bought a new computer. Clean. Developer-focused. Almost no program I used to know. But now I know what I have and where:

• Docker — local dev, containers, isolation
• Warp — a terminal that looks like it's from the future
• Cursor — an editor with AI at its core
• Figma — design, prototypes, UI
• Raycast — a launcher that replaced Spotlight and five other apps

A year ago I might have recognized Docker from that list. Now they're my daily tools.

And that computer? It's been running non-stop for a week. Hasn't shut down once. Because there's always something to do, always something to try, always something to build.

Phase 6: carfast.cz

Then came a project that changed everything. carfast.cz — real business, real site, real users.

I asked Claude: "What can we improve?" And we were off.

Coding. Marketing. Design. All in one flow. One day we're writing code, the next day analyzing competitors, the day after that designing a landing page. Claude isn't just a programmer — it's a partner that understands the whole spectrum.

And then I realized I needed multiple Claude sessions at once. One Claude refactors the backend. Another writes copy. A third analyzes data. I've got three tabs in Warp, each running Claude Code.

I upgraded my subscription. Because it simply works and because time is money — and that cuts both ways.

Phase 7: Claude Remote

And as if that wasn't enough — Claude Remote. Claude running on a remote server. You assign a task, close the laptop, go for a beer. You come back and it's done.

Game over. This is what the future of development looks like. Not "AI will replace programmers" — AI gives programmers superpowers.

Reality

That's how it was. That's how it is. And I keep going.

It wasn't a linear "I tried AI and it was great" story. It was a story of:

• Frustration — limits, waiting, "why doesn't this work"
• Surprise — "this actually works?!"
• Learning — Docker, terminal, Git, Vercel, Supabase
• Obsession — a computer that doesn't shut down, ideas at 3 AM
• Results — real projects, real value

And yeah — it's a drug. You hit the limit and have to wait. But limits can be raised. Because time is money.

For those starting out

1. Start with the free tier. See what it can do.
2. Hit the limit. Know that you want more.
3. Buy Pro. $20 a month. Best investment.
4. Try something silly. Count ducks, generate cat names, whatever. Prove it's doable.
5. Build something real. A work utility, a site, an automation.
6. Get lost in it. And that's OK.

Because "it can't be done" doesn't exist. Only "I haven't tried it yet."

Specific tips and tricks I picked up in that month are in 10 things I learned in a month with Claude.

And now excuse me — I've got three Claude sessions open and carfast.cz isn't going to build itself.