I'm not a mobile developer. My knowledge is limited to some basic HTML, CSS, and JavaScript. However, I had a concept for an application - a vault for collectors, specifically for individuals who collect coins, cards, watches, and similar items. It was intended to be a platform for cataloging everything, assessing its value, and securing it with Face ID encryption. At the time, it appeared to be a straightforward task. After two weeks and 48 EAS builds, it is now in beta. Here is how the process unfolded. The Disputes One aspect of vibe coding that is often overlooked is the extent to which it involves negotiating with an AI. I would articulate my requirements, Claude would propose an alternative, I would reject it, it would provide reasoning, and occasionally I would concede, while at other times it would relent. This back-and-forth dialogue was, in fact, where the majority of the significant decisions were made. The initial major disagreement revolved around encryption. I believed it was logical to encrypt the entire database file. Claude, however, consistently opposed this idea, arguing that it would complicate iCloud synchronization and introduce a native dependency that I would later regret, suggesting instead to encrypt it field by field. I countered that this approach seemed far more complex. It insisted that while it was indeed more complicated, it was the correct decision. I ultimately acquiesced, spent a week implementing it in that manner, and indeed... Claude was correct. This dynamic was essentially the crux of our interactions. The Builds There were 48 builds to EAS/TestFlight before the application functioned properly from start to finish. Some failures were due to Xcode configuration issues that I did not comprehend, while others occurred because I would rectify one problem and inadvertently create three new ones. At least three or four of these failures were attributed to an iCloud bug where I mistakenly passed a configuration value as an array when it was required to be a simple string. The build completed successfully, yet the application simply did not operate on the device. There was no crash, no error message, just... silence. It took an embarrassingly long time to identify that issue. Many of the builds were genuinely a result of my lack of knowledge and perseverance. Claude would explain a concept, I would attempt it, it would fail, I would return the error message, and we would determine what went wrong, allowing me to try again. This iterative process likely accounts for a significant portion of the overall experience. The security aspects This section caused me the most anxiety. I continuously encountered problems, some of which I identified myself, while others were pointed out by Claude when I presented him with the code I had developed. There was a bug in the Face ID process where the encryption key remained in memory longer than necessary after the vault was locked. Claude identified that issue during his review. Subsequently, I discovered another problem - the iCloud restoration process did not prompt for biometric authentication before overwriting the vault, allowing anyone with access to an unlocked phone to restore everything without any warning. I identified that issue around midnight and felt a strange sense of pride in doing so. Additionally, there is an Apple compliance requirement that mandates the declaration of whether your application utilizes encryption. I experienced a moment of panic when I encountered that, fearing I would be flagged for export violations or similar issues. Claude guided me through the process, and it turned out there is an exemption for applications that only encrypt the user's local data. The correct response was simply `false`. At one point, I nearly altered it to be "safe," but Claude advised against it, which was wise because it would have initiated an entirely new review cycle. What I truly wrote versus what Claude contributed Honest response: Claude was responsible for the majority of the scaffolding and boilerplate, while I focused more on the product decisions and reviewed nearly everything. The feature logic felt like it belonged to me - determining what is free, what is paid, how the quota system operates for the AI scans, and what occurs at the limits. Claude would challenge me when something appeared incorrect. At one point, Claude suggested placing the biometric lock behind the paid tier, akin to a "premium security" feature. I advised against that, stating it is inappropriate to require payment for securing one's own vault; this is something that must be trusted unconditionally. Therefore, certain decisions are clearly not for Claude to make. The quality of prompts and the importance of explanations are significant in these matters. The AI identification screen is the feature I take the most pride in. You can take a photo of an object, and it determines what it is, automatically filling in the item form. I scrutinized that feature closely - I made Claude clarify anything I found unclear, revised sections I was dissatisfied with, and it has become one of the features I am most proud of. Essentially, you can scan a photo of an item, and it automatically populates various data (for instance, when scanning a banknote, it even captures the banknote number and inputs it). If you are uncertain about the accuracy of Gemini's value estimation, you can simply click a button, and Perplexity Sonar will assess it. Naturally, given the nature of AI, you still cannot place complete trust in it, but something is certainly better than nothing. The moment it clicked After a few days, I opened the app on my actual phone, and it simply... functioned. I launched it, the Face ID prompt appeared, the vault unlocked, I scanned a coin, and the AI recognized it, automatically completing the form. Everything was in the correct order without any crashes, of course, a few bug fixes were needed, but after shipping a whole 48 builds I hope I caught them all :D. So all this time later, it's finally beta and I cannot be more excited. If anyone wants to test it out feel free.

Locrow — Private Collector Vault for iOS

Your vault. Your device. No compromise.

Your vault. Your device. No compromise.