Using Claude Opus 4.5 to build Nova Music
AI isn’t the bottleneck anymore
I’ve had this idea since my wife started teaching piano when we moved to California. An app to help her with admin tasks, primarily billing, scheduling and lesson management.
I started by looking at competitors, there are a few, and trying to understand what the solutions to this problem looked like for them. Then I tried for months with various LLM coding models and vibe coding tools to build a version of this that works.
I began building with Bolt and Lovable, and with Sonnet 4.5 and Codex was able to make decent progress, but always bumped into a few of the same issues, namely creating a semi-separate system for student login that kept them connected to the teacher, basically relatively complicated database relationships, and then other small little features or bugs, usually inconsistent. But always present.
So I spent months in Bolt and Lovable trying different iterations of prompting to get around the problems, but could never get the LLMs to cross the finish line.
Enter Opus 4.5 + Bolt. Once Bolt had these enhanced capabilities, there were absolutely 0 problems. Tough database structure? No problem. Change the CSS from Tailwind to Shadcn? Barely 30 minutes of work. Want a Stripe integration? Easy, I’m the biggest bottleneck, not the AI.
It opened a whole new world of code to me.
So in about a week I went from this idea that I knew well, I knew the problem, the solution, what I wanted it to look like, how I wanted it to feel and work, and finally I could bring it all to life with minimal effort and cost.
The alpha is live now at trynovamusic.com if you’re interested and want to take a look, I’d love your feedback.
There are still plenty of features, bugs and integrations that need clean up and fixing, but I feel pretty comfortable launching this as an alpha.
Big takeaways:
Code isn’t the bottleneck anymore, I am.
Ideas are becoming more valuable than execution very quickly, execution is still a bottleneck, but it’s mostly about what I know or don’t know, AI can solve most of the problems I throw at it from a code perspective.
The barrier to entry for digital products is getting lower and lower, I’m not sure what that means going forward, but its great for someone like me who has lots of ideas on what I’d like to build. It’s also a bit of a forcing function, now that I can build anything, what should I actually build?
Great write-up on the shift Opus brings to buidling. The insight about code not being the bottleneck anymore really lands for anyone whos been stuck in endless iteration cycles. I've been in that exact loop before, trying toresolve database issues for weeks, and it's wild how a capability jump just dissolves those walls. The question of what to actually build now becomes way more urgent than how.