How do you learn when Claude writes everything?

I build voice AI agents. One client, system has been live for about a year, handles around 700 calls a day. Python, LIvekit, Docker, Azure, websockets.

I started knowing only Python. Basic Python. No TypeScript, no web, no deployment knowledge, no backend design. Since then I've built a whole TypeScript SDK and a dashboard, and an infra that is deployed across 3 regions across a country and honestly Claude wrote almost all of it.

Here is my actual day. I describe the problem, Claude investigates, Claude gives me a plan. I read the plan. I don't understand half of it. It talks about functions and classes I've never opened, file paths I don't recognize, terms I've never heard. So I say yes. Then it says here's the next step, and I say yes. Approve, approve, approve. It works, the client is happy, and I close the laptop knowing nothing more than when I opened it.

I'm not lazy about it. When something new comes up I do go read about it. But it doesn't stick. Two days later it's gone, because I never had to use it, I just approved it.

So my question to people here:

1. How do you use Claude and still come out of it knowing your own system? Not "I built an app" knowing, but the kind where you could sit in a startup interview and explain your architecture, why it's built that way, what breaks it, how you'd scale it.

2. What actually works for you? Do you make yourself explain the change back before approving? Do you read the diff line by line? Do you learn separately, away from the work session? Do you ask Claude to teach instead of do, and does that survive a busy week?

3. And if you've been doing this a year or two, are you genuinely better than when you started, or is this just what the job is now? Honest answers appreciated, including the harsh ones.

4. Any particular skills, or workflow you are using?