Startup Iteration Isn’t Sexy But It’ll Save Your Launch

Sometimes, the hardest part isn’t solving the problem, it’s realizing you’re solving the wrong one.

I’ll be honest. At ZillyPlanet, we hit a serious wall while building our MCP model. Our prompts had the right intent, but they just weren’t landing. The model couldn’t quite “get” the learner's context. And no matter how many tweaks we threw at it, the results stayed... meh.

Now, as both a CPO and founder, I’ve learned the hard way that stubbornly hanging on to broken systems is a one-way ticket to burnout, and a delayed launch. But still, this one stung. We had already invested weeks shaping this prompt flow. It wasn’t easy to say, “This isn’t working.”

So, we scrapped it. Fully.

You Know What Helped? Just Talking.

Not a fancy model. Not more code. Just…a real conversation.

We sat down, me, our math advisor, and a few teachers, and unpacked what wasn’t clicking. No slides. No dashboards. Just some good old-fashioned, "Hey, what’s actually going on here?"

And the clarity that came out of that meeting? It was a relief. Turns out, our prompt system wasn’t broken, it was just pointed in the wrong direction. We were trying to fine-tune something that didn’t need tuning. It needed rethinking.

So we rebuilt it from scratch. Different scaffolding. Different logic. Way better results.

Yes, It Slowed Us Down. But It Made the Product Worth It.

We’re a bit behind schedule now. That part is real. But what we’ve got? It’s smarter, faster, and way more fun for the kids who’ll use it.

ZillyPlanet’s Beta is nearly here. Grades 2–8. Adaptive math in a gamified world that actually feels like play, not homework dressed up in sparkles.

If you’re a parent, this is your shot to get early access. First year’s free for early adopters. And we’re opening just a few more teacher spots.

I’ve been in product for a long time. But I’ve never felt this excited about what we’re putting out into the world.

Kids deserve better tools to learn math. Not just “more engaging” worksheets. Tools that understand how they think.

We're getting close.

UPDATE: We were able to launch our Beta at the end of October and it is working much better. We have moved on to more competencies to get us to a full launch in November.

Q&A with Me, Founder and CPO of ZillyPlanet

Q: Why not just tweak the original prompt system? Because band-aids don’t fix the foundation. We needed context, not cleverness. Starting fresh saved us months in the long run.

Q: Isn’t it scary to start over mid-build? Yep. Every time. But what’s scarier? Shipping something half-baked. Parents and teachers deserve better. So do kids.

Q: What’s different now about the new MCP system? We added more context and got deeper into the details. Less guesswork. More precision.

Q: What’s something you’d tell other startup teams facing a similar wall? Don’t cling to sunk costs. If your gut says it’s off, pause and ask why. Sometimes the fastest path forward is a short walk backward.

Q: What’s next for ZillyPlanet?A: Launching Beta. Getting feedback from real classrooms. And growing our community of educators who want AI that actually helps, not hinders, learning.