There's one thing I didn't expect about Claude. It's not that it's good, I already knew that. It's that it gets better faster than I can adapt.
April started with me working more than ever. Several Claude Code sessions in parallel, projects I would never have bothered starting otherwise. I was hitting my limits almost daily and understood right away why people recommend the 100-dollar plan. It felt like the tool was pushing me to take on bigger things.
Then came the problems. Claude Code is not magic. A larger codebase or a task outside its direct reach, and it starts to crack. My concrete example was running Claude Code scripts on a schedule. No native solution, so I had to mess around with Windows Task Scheduler and NSSM. I got it working, but it took more time than it should have.
Five days later Anthropic shipped Routines. My workaround was suddenly unnecessary. This isn't a complaint. It's the observation. I solved a problem that wouldn't exist five days later, and had no way of knowing it.
It's this speed that forces you to change strategy. Later in the month I realized my limits weren't a blocker anymore. Not because they had been raised, but because I had gotten smarter. New sessions instead of bloated chats. Better prompts on the first try instead of iterating. That's the difference between fighting the tool and working with it.
And then Claude Design. It's a ridiculous way to quickly make complex and professional designs, and the best part is you can export straight to Claude Code. It's not just a new surface to learn. It's a surface that plugs directly into the one I already use the most. The tool is growing both in breadth and in how the pieces connect, and it's moving faster than you can map it.
That's what April taught me. You can't build a stable mental model of Claude as a tool, because the tool isn't stable. The only thing you can do is get better at adapting fast.