in an ideal world.
the ideal world however has a few assumptions baked in:
- you think of EVERY edge case a user can ever encounter before you start designing the system
- you think of EVERY design element a user can ever encounter before you start coding the system
- you think of EVERY bug the code can encounter before you start start implementing the feature.
- you think of EVERY test the code must pass before you pass it on to QA
well the ideal world didn't exist even before AI started coding and now it has gotten considerably worse.
there are bugs and then there are bugs which aren't recorded by Sentry. the hardest ones to find but also the costliest ones.
at @lucent_ai when we think of bug detection its not necessarily a bug in the code, we just want to make sure that your user's product experience is exceptional.
it doesn't matter if the error wasn't recorded in Sentry or it was a bug in your 3rd party tool or if it wasn't even an error(just a friction) or something completely unexpected.
the goal is "exceptional" and nothing less.
today i want to walk you through the 4 weirdest bugs Lucent found this week:
(i know you are gonna think man no way this bug exists but it does)
- a user hit a limit on free usage and clicked on "Unlock Premium" but instead of the upgrade flow, they got another "generation limit exceeded" popup.this is as far you can go from the concept of "conversion flow" as possible. this won't even be recorded as a sentry bug because the code works like how it should.
- on phone verification, the UI countdown finished, the user clicked resend code, and got "not enough time has elapsed." grontend timer and backend rate limits were out of sync - classic "worked in dev" bug.
- a user was stuck in a loop: authentication error → back to Login → welcome modal → get started → back to authentication error. no crash, no Sentry spike, just a user who can never finish onboarding.
- user clicked an upgrade link to billing. a loading skeleton layered on top of the real page and never went away even on page change. the app looked broken site-wide.
now you are gonna think, "no way the code i wrote is gonna cause such simple bugs" and you are right in the ideal world but let's be real on how far you can go reviewing a 5000 line PR.
does that mean you should ship anything?
absolutely not.
but your users will still encounter friction, bugs and unexpected things and there is no reason why you shouldn't put your all towards fixing them. leave the identification to us.
first 400 sessions (without a credit card) on us. if you encounter an issue, i'll reply < 2 mins and fix it < 2 hours. that's my promise.
we work with posthog, datadog, amplitude and also have our own SDK with privacy first defaulted i.e. every PII is obfuscated.
we ship every day and use Lucent to make Lucent incredible. come check us out :)