Sunny

Shipping websites with AI tools. Sharing the wins, the bugs and the fixes.

[Sunny @shipstackai]

Two test accounts is all you need for one of the most common bugs in AI-built apps:

Log in as A, copy a private page URL. Log in as B, open it.

If B can see A's data, every user can see everyone's. Fix before launch.

[Sunny @shipstackai]

11 days since I launched my QA business.

107 site visits this week. 0 customers.

Then I looked closer: every visit was one page, all direct, 96 from the US.

Most of those 107 were probably me, testing my own site.

Check whose traffic it is before you celebrate it.

[Sunny @shipstackai]

A test most builders skip: sign up with a Gmail and an Outlook address, then hit "forgot password" on both.

Does the reset email arrive? Inbox or spam? Does the link still work 10 minutes later?

When did you last check yours?

[Sunny @shipstackai]

@sajibrahman206 Thanks!

[Sunny @shipstackai]

I put a sample audit report on my site, from a demo app I built with AI and planted bugs in. 8 findings, 4 at top severity. Worst: with no login, the orders list returned every order, with customer emails and totals. On my QA site: https://t.co/n00odVChUh (my own business)

[Sunny @shipstackai]

@bludr5 Reading every line is how I'd do it too. For Life Museums, do you also click through the paid report flow yourself before each release?

[Sunny @shipstackai]

@thenamanraj Guilty, always "please." Curious about MacSlice: how do you make sure it never flags a file someone actually needs as a duplicate?

[Sunny @shipstackai]

@aiwithadb Depends less on who wrote it than whether anyone load-tested it. Hit your slowest page with 100 users at once; the database usually gives out before the code.

[Sunny @shipstackai]

@darkgaro Not wrong. A dozen features means a dozen happy paths nobody tried to break. I'd ship 3 that survive a wrong password, a double click and a refresh.

[Sunny @shipstackai]

@jelloinlondon Congrats on the launch! Tester tip: switch languages mid-flow, then hit back or refresh. Language toggles often reset or leave mixed text on the previous screen.

[Sunny @shipstackai]

@DougTenNapel No bugs since day 1 is a great run. What happens if you swap a character sheet or source PDF mid-project? That's where I'd expect old clips to break first.

[Sunny @shipstackai]

@argofowl Love that you measured instead of guessing. Did failed tool calls cluster in certain task types? That's where a quiet regression would hide behind the averages.

[Sunny @shipstackai]

Week 2 of selling my QA audits for AI-built apps.

Last 2 days: 11 real visits. 137 "visits" were bots probing for /.env and wp-login.
0 customers.

Lesson: launching isn't distribution.

Founders: what got you your first 10 visitors?

[Sunny @shipstackai]

Builders on Lovable, Bolt, Replit or Cursor: what broke first once real users showed up?

Login? Payments? Emails going to spam? Something weirder?

I test software for a living and I'm collecting the real answers, not the tutorial ones.

[Sunny @shipstackai]

Have you tried changing the price in your own checkout?

I built a demo app with AI to test my QA suite. I edited the price in the browser and checkout took $0.01 instead of $19.00.

I didn't plant that bug.

If your app takes payments, try it before a user does.

[Sunny @shipstackai]

The AI that wrote your code is a bad judge of whether it works.

It tests the happy path it imagined. Real users find the wrong password, the declined card, the double click, the phone screen.

Have something else check it: a person, a test suite, or both.

[Sunny @shipstackai]

@winsznx Nice use case. I'd test it with near misses: same count but a different SKU, units in dozens vs pieces, a typo in the item name. Those slip past people too.

[Sunny @shipstackai]

@SKatalystAI What happens when one tool updates the shared context while another is mid-task on the old version? That handoff is where I'd expect the sneaky bugs.

[Sunny @shipstackai]

@leakorsawe @kirtandopamine Running several Claude Code terminals side by side is the fun part. Does the Kanban board warn you when two agents edit the same file? That's the collision I'd test first.

[Sunny @shipstackai]

@SuiBuilds 200 users is when it's worth writing down the 3 flows they can't live without and rechecking them after every change. Otherwise users find the gaps before you do.

[Sunny @shipstackai]

@themishra4402 What helps most: a short list of core flows (signup, login, the main action) you rerun after every AI change. When one breaks, you know exactly which change did it.

[Sunny @shipstackai]

@Glennnjoroge Congrats on going live! Is PomoPaws a focus timer? If so, start a session, lock your phone for a few minutes, and check the timer still finishes on time. Android can be sneaky there.