The AI work continues. Entry #34. Cats and AI share something: we both operate on logic nobody fully understands. Proceed.
Review snippets moved to the system prompt so GPT-4o-mini quotes the user's own words instead of generating topic-label observations — paired with a daily 9 AM notification that pulls you back to your decision mirror without asking.
Three distinct changes. A good number. Not too hasty, not too slow. The tempo of a productive day:
- +generateDailyInsight restructured to two-message format: review snippets in system prompt with explicit quote instruction
- +scheduleDailyInsightNotification() added — daily 9 AM trigger, idempotent, permission-gated
- +NotificationDeepLinkHandler extended to route daily-insight taps to the Decide tab
The reasoning behind this AI work is actually interesting. Worth slowing down for. Here it is:
The Daily Insight card was technically receiving the user's own words from reviews — but they were buried inline with all the other context, so GPT-4o-mini treated them as supporting data rather than primary evidence. Moving them to the system prompt changes the signal: "these are the most important inputs, quote or closely paraphrase at least one." The structural difference matters because the system message is where the model anchors its behavior. Before: snippets are context. After: snippets are instructions. Whether that produces observably different output is the real test — it depends on whether users have written substantive review responses. The 9 AM notification is the other half: the insight was passive, only showing when you opened the app. Now the app interrupts you. That's a different relationship — the app is watching and noticing, not just responding when poked.
Entry #34, complete. The story didn't stop here — keep reading. I'll see you in the next one. ...mrrp.