Spent 2 hours debugging a "broken AI" that wasn't broken at all!

PrOpinion's whole job is to take your situation and give you a clear AI opinion. Today it just kept showing a generic fallback: "we weren't able to fully analyse your situation right now."

No error on screen. No crash. Just... nothing useful.

First instinct was that the app code was wrong — checked the fetch call to my Supabase edge function, checked the categories config, even rebuilt and reinstalled the app fresh. Everything looked correct.

The actual fix was checking the Supabase edge function logs (Edge Functions → your function → Logs tab). Buried in there: Anthropic API error: 401, invalid x-api-key.

The API key stored as a Supabase secret had gone bad — never even got the chance to try the request the way I assumed it had. Generated a fresh key, updated the secret, and the very next request returned a real opinion.

Lesson: when an AI feature silently falls back to a generic message, check the server-side logs before touching any client code. The failure point is almost never where the symptom shows up.

While I had the hood open, also fixed the category list (4 live categories now: Home, Motor, Money, Work — rest are "Coming Soon") and gave the app icon/logo a refresh — swapped the ring colors so gold sits inside the white "O" instead of the other way round.

PrOpinion: propinion.ai — launching June 23rd.

—metaljelly