Last week: what shipped, what broke, and a user message that made it worth it
Three things happened last week worth writing down.
First: we fixed a bug that was blocking 23% of new Android installs. It was in the mic permission flow. Users would install, hit the permission screen, and stop. They never came back. We only found it because I started watching session recordings instead of reading funnel aggregates. Aggregate numbers hide this kind of thing perfectly. Your conversion rate looks fine if you don't know you're measuring from step 2.
The fix took a morning. Finding it took three weeks of looking at the wrong metrics.
Second: I removed a feature that 9% of users were actively using. Zero correlation with retention or paid conversion. I'd kept it on the roadmap because "people use it" felt like a reason. It isn't. The right question isn't "do people use this?" It's "does using this make people stay?"
Cut it. Put the hours back into something that moves the needle.
Third: a user messaged me unprompted. They're a second-language English speaker in a native-English team. Three years of struggling with written communication. They said Genie 007 changed how they show up at work. Not "it's useful." Changed how they show up.
That's the version of the product I've been building toward for 18 months. Slow weeks don't disappear when you get a message like that. But they make sense.
What I'm working on this week: the Windows app is getting a significant update. Not posting specifics yet. But it's been the thing holding back a few enterprise conversations sitting at "we like it, not quite there."
Current status: live on Windows, Mac, iOS, Android, and Chrome extension. 1,200+ clients. Pre-launch. The official launch is coming. Still building the version I'd be comfortable putting in front of a lot of people at once.
One thing I've noticed: the signal-to-noise ratio in weekly metrics is bad. One real user message often tells you more about whether you're building the right thing than a full week of dashboards.
What's your most useful non-metric signal? The thing that tells you whether you're on track that doesn't show up in analytics.
Building Genie 007 (genie007.com) -- voice-to-action everywhere on your computer.
Ours is the first message someone types. I'm building Mythex, an AI app builder, and the first prompt tells us more than any funnel step: did they arrive with a real job in mind (something like "a booking page for my clinic"), or were they just poking at it with "test"? When the share of real-job first prompts moves, the quality of sign-ups moved, whatever the raw count did.
A close second: questions people ask support about things that are already in the UI. Each repeat is a missing label somewhere.
The 23% Android stop is a great find. Did the recordings show people reading the permission prompt and declining, or leaving before it even appeared? Those point to different fixes: the copy before the prompt, or when you ask at all.
The permission-flow example is a good reminder that an aggregate funnel can look healthy while one state transition is quietly stopping users. I also like the distinction between “people use this” and “using this changes retention or paid conversion.” Now that the fix is live, are you tracking the permission decision and the next successful session separately, or only the install-to-active rate?
This is great work — reminds me of some of the calls I've had to make building Xstream4K. What would you do differently if you started over?