Building in Public #18: Chasing the market leader
By Peter Coppinger
This week was a rollercoaster.
Last week I promised three things: get our new React builds back up to the quality we had before, get first builds well under 8 minutes, and move the last few kinds of project over. I'll tell you how I got on at the end.
We were ahead. Then I reset everything.
As a reminder, a few weeks ago we were actually ahead. In mid-October our websites beat the market leader in blind taste tests, all four test sites, and on a random app I built on both, Vorx walked all over them.
Then I decided we had to switch from Vue to React. They're two different ways of building web apps, and React is just far more popular, which matters for the people who'll use Vorx. But it meant redoing how we build almost everything, and it set us back. Speed, design and cost all went backwards for a while.
It was the right call, but it hurt. And the start of this week was one of the lowest points of the whole project. Everything seemed to be going wrong, builds were slow, and I couldn't see why.
Finding the slow bits
So I started digging. We've always logged what happens during a build, but this week we made that logging far more detailed, right down to every single AI call: when it started, when it finished and what it cost. So instead of guessing, we can see exactly where the time and money go.
And it found some real time-wasters. One planning step took 72 seconds while everything else sat waiting for it. Pages were waiting on images they didn't need yet. And we were paying the AI to write a whole plan on every build that nothing ever read. All fixed the same day.
Edits got the first big win. Describing a bug and getting it fixed now takes 47.5 seconds and costs 15 cents. Before, it was 137 seconds and 37 cents. I dared to hope I was on the right path. But something still felt wrong. The agents were busy, but we didn't seem to be making progress.
Tests: going nuclear
Last week I told you the agents were running every automated test on every change. This week I found more madness. I felt like I was fighting the agents. They kept adding tests that slowed everything down, even though I kept telling them we are pre-launch.
I literally told my COO agent that the only priority is getting the React build and update systems better than the market leader, and anything that slows that down needs to be turned off or removed. Here it is word for word:
The only priority is getting the React build and update systems better than [the market leader]. Anything that slows that down needs to be turned off or removed.
Beating the market leader is my obsession right now, and anything that slows us down has to wait. The one exception is anything that could delete data or spend money over and over. Those keep their tests, and another agent has to check them before they go in.
The scoreboard
By the end of the week, the numbers had turned. To monitor progress, I keep a simple Claude Code artefact that my COO agent updates every 30 minutes. It shows me the big picture and the momentum at a glance. Here it is on Sunday evening:

The good news first. An app's first screen now shows up in about 51 to 53 seconds, against their 55. On Thursday it was 97 to 117 seconds, so we've halved it in a few days. A follow-up edit now takes 67.5 seconds against their 113 to 136. And a full app build finished in 134 seconds against their 153, although to be fair that's only one run so far. By Sunday I was genuinely excited again.
Now the bad news. On design, judged blind, we haven't won any of the last three rounds. I won't launch this until we are beating the market leader consistently. It's a matter of pride.
I won't launch this until we are beating the market leader consistently. It's a matter of pride.
But I want to be ahead everywhere, not just on some apps. And design is where we're furthest behind.
So what's going on with design?
Good question. For apps, we're in good shape but for websites we're behind.
Just when I was feeling good, I opened one project and it was a disaster next to the market leader's version. And it wasn't a one-off: judged blind, our designs keep losing for the same reasons. Colours that don't come from the brief, too many fonts on one screen, and a first screen that isn't the actual app.
The flashcard app is a good example. Fake wood drawn in code, a "handwriting" font that was really just an ordinary system font, and a whole website header and footer wrapped around what should have been a simple one-screen app.
The good thing is we know exactly what's going wrong. So we're fixing it in the starting template every build begins from, which means every future build gets the fix too. Apps get a proper app layout, real photos instead of fake textures, and every project gets its own colours. That last one was me. I noticed 31 of our 35 test apps had come out in exactly the same colours.
The other work: betas and polish
While the agents chase the market leader, I've been working on everything else. The whole backlog is cleared: 9,187 tasks done.
The whole backlog is cleared: 9,187 tasks done.
There are 28 beta features built but switched off for now, and I'm switching them on one by one, only once I'm happy with each.
One I'm really enjoying is annotations in Visual Edits. You draw a box, an arrow or a pin on your app, say what should change there, and send it. I combined it with the Visual Edits feature we already had, so it all feels like one intuitive system for making visual changes.

I'm also going through every page, screen and dialog in the product to make sure I'm happy with it. The design system is getting a proper upgrade too. I first built it just to tick the box, and it was basic. I want it to be genuinely great.



How I got on with last week's promises
- React back to the quality we had before: Not yet. On speed we're there and pulling ahead. On design, judged blind, we haven't won any of the last three rounds.
- First builds well under 8 minutes: Yes. A full app build now takes 134 seconds, against the market leader's 153, and the first screen shows up in about 51 to 53 seconds.
- The last four kinds of project moved to React: Half done. The two kinds of landing page are switched over to React. Docs sites, both kinds, still build in Vue, so they're next.
We're nearly back to where we were before React, and we're starting to pull ahead of the market leader on many metrics now. It's exciting. Next week I'm hoping this post is called "Catching the market leader".
It's been some week: from one of the lowest points of the project to excited again. One big learning, which I've learned several times on this project: don't blindly trust the AI. Maybe this time it will stick.
So far we have just a handful of friends using Vorx to run simple websites. If you want a free website, or want to upgrade your existing one, let me know.
Stay updated: one founder, an army of AI agents, building this in public.
Get each new diary entry by email: the wins, the failures, and the one number we're chasing. No spam, unsubscribe any time.
Shipped this week 3 Nov – 9 Nov
- Platform: Form submissions read like an inbox
- Payments: Pay now on shop order confirmations
- Collaboration: Share a project as Can admin
Comments