Built in public
The journal
What I'm building, what broke, and what the numbers actually say — including the weeks that went badly. If you're deciding whether to trust a one-person product, this is more useful than a testimonial I could have written myself.
-
The bug that was quietly eating people's work
An adversarial audit of my own code found three ways the app could silently destroy something a user had written — press Enter twice, or edit while the AI was still thinking, and a version simply vanished with no error. Nothing crashed. Nothing logged. That is the worst kind of bug. Here is what it was and how it got fixed.
-
Why the AI now writes its own notes about you — for free
The Brain has an "AI learned" section, and for a while it was permanently empty because nothing ever wrote to it. Now the app works things out from what you actually did — how picky you are, whether you finish what you plan, which weekday your plans land on — using arithmetic instead of an AI call, so it costs nothing. And the moment you edit one of those notes, the machine never touches it again.
-
The number that decides whether this business lives
Five deep-model AI actions a day, given away free, would cost me about ₹528 per user per month. Twenty users and I am paying out more than a salary to give something away. That single line of arithmetic dictated the entire pricing ladder, and I would rather show it to you than pretend the plans were designed around your needs.
-
Google quietly blocks free Gemini keys from data centres
Four independent tests before I believed it. A free-tier Gemini key works from a laptop and refuses from a hosting server, which is exactly where an app runs. Cost me an evening. Writing it down in case it saves someone else one.
New entries land when something worth writing down happens, which is not on a schedule. Write to me if you want to argue with any of it.