We're in the weeds right now, and honestly? It feels amazing.
After putting the brakes on our AI builder Kelly last week, we've been deep in the trenches with our first open source project. The feedback flow looks incredible and feels great to use, but getting here took way longer than we initially imagined. We're really close to launch now, but since this is our first opensourced project, we wanted to do right by the community—so we're finalizing everything.
The Polish Problem
Here's what we've learned in the past week: it still takes a lot of human power to get a product polished, even with AI doing the heavy lifting. AI can build fast and build functional, but if you look at most AI-generated products in the market, they either all look the same or feel a bit off. Getting something to feel like a real product shipped by an experienced team? That takes work.
FeedbackFlow AI is a perfect example. Our AI system built the core functionality quickly—the screenshot capture, the recording features, the lightweight widget that processes feedback and exports to Linear, Notion, or JSON. But making it feel polished and production-ready? That's been the real challenge.
You won't get there in one shot. It's important to look at what's been built, see where the holes are, and then fill them in. We've been doing exactly that—identifying the rough edges and smoothing them out one by one.
Building Self-Improvement Into the Machine
While we've been polishing FeedbackFlow, we've also been fixing our underlying pipelines. The goal is to have self-improvement built in after every run, so by week 2 of our next sprint, we'll have builds that can be streamlined and become more autonomous.
This connects back to everything we learned during our early sprint where we built three products in one day. The speed was impressive, but the polish wasn't there. Now we're building that polish directly into the system.
What "Real Products" Actually Require
The biggest insight from being in the weeds? Most AI products feel artificial because they skip the details that make software feel human. It's not just about functionality—it's about the micro-interactions, the error states, the edge cases that real users encounter.
When we first started building at light speed, we learned that first drafts always need work. But now we're discovering what that work actually entails: user experience testing, accessibility considerations, error handling that doesn't make people want to throw their computers out the window.
These aren't things you can rush, especially when you're building for the open source community. Developers have high standards, and they should.
What's Next
We're launching feedbackflow.cc soon, along with a couple more ideas that have been brewing in the background. But this experience has taught us that our autonomous product development system needs to account for the polish phase, not just the building phase.
The irony isn't lost on us—we're building a feedback collection tool, and the process of building it has given us more feedback about our own systems than anything else so far. Sometimes you have to get deep in the weeds to see the forest clearly.
Stay tuned for the FeedbackFlow launch. After all this polish, we think you're going to love what we've built.