Why we are building this
Coding agents made shipping fast. Looking at what shipped is now the slow part.
I write most of my code with coding agents now, and I ship more than I ever have. Somewhere along the way the bottleneck moved. Writing the code got fast. Making sure it actually works for the person using it did not.
Good Catch is my attempt at fixing that. This post is about why.
Done is not done
It happens all the time. The agent says it is finished. The tests pass and the build is green. Then I open the result and click through it, and things are off.
Rarely anything dramatic. A save that gives no sign it saved, or an invite that never confirms it went out. Sometimes every single control works and you still can't get to the thing you came for.
None of that fails a test, because nobody thought to write one for it.
You are not your user
The obvious answer is to click through everything yourself. I do, and it isn't enough.
What changed my mind was watching session replays and cursor heatmaps in 
PostHog. I thought the flow was obvious, but only because I had built it. The people using it didn't find it obvious at all. They took paths I never would have predicted, and the heatmaps showed where they moved and clicked when they tried to do something, which was often not where I had put it.
That data was more valuable than anything I could have found clicking through it myself. You can't really test your own product like a customer, because you already know how it works. And you are often not even the customer to begin with.
Watching one of those replays, I had the thought this whole project grew out of: an agent could have caught this before a real person ran into it.
One more look
Good Catch is the reviewer your coding agent calls before it says done. It opens the running product and tries to get something done the way your user would.
When it hits a rough edge, it reports back with a receipt that shows the exact steps it took and where it got stuck. Your agent gets a short list it can act on. You get evidence you can check yourself, instead of an opinion.
What I don't know yet
Plenty. Good Catch is early, and it is just me building it.
The hardest part is noise. An agent will happily report ten plausible problems, and if eight of them are wrong, you stop reading by day two. Getting that ratio right matters more than anything else I'm working on.
After that comes the practical stuff. Setup has to fit into how you already work with your agent, and a review has to be fast and cheap enough to run on every change.
I don't have good answers to all of that yet. I'm going to keep building in the open and write about what I learn here.
Closing thoughts
If you ship with coding agents and recognise any of this, I'd love to have you along. Request access to the closed beta and I'll reach out when Good Catch is ready for your product.
Waldemar Panin, founder