Not because I got faster at writing code. Because the tools got precise.
I tried a project like this 6 months ago and spent half my time iterating through garbage output and half my time rewriting what Claude spit out. This time? I specified what I needed, and got exactly that. No junk. Minimal handholding required.
My 9 year old is using it right now. That’s the bar that matters. I built it because she was in tears over spelling practice with other apps. They were all about catching mistakes. Mine separates practice from judgment.
Here’s what actually shifted: the bottleneck moved. Six months ago, you could ask an AI agent to code and spend more time manually debugging the output than I spent building this app. Now? If you know what you want to build, you get it. Which means the skill that matters isn’t writing code faster, it’s knowing what to build and why.
That’s architecture. That’s design thinking. That’s the intuition you develop from actually solving problems.
This is obviously a fairly simple project, but it’s my on-ramp to something bigger: multi-agent workflows in Claude Code, which require way more upfront design work, not less. The spelling app proved I could work with these tools at this level. Next is designing something more complex that needs agents to talk to each other and make real decisions.
This touches something else though: I built an app that delivers what the similar solutions behind a $20/month education SaaS subscription do in a weekend because I needed it for my kid. Why would I pay a subscription when I can build exactly what I need? Agent-assisted coding isn’t replacing software engineers. It’s giving us back the joy of building things we actually own.
The role is evolving. Fast. If you’re not experimenting with what these tools can actually do, not the hype, the real capabilities, the disruption is going to hit you different.
Here’s an overview of the app, which creates word lists with AI-generated context sentences: spelling-word-collector.