Atlassian AI is dangerous 🤯

[LukasGotter]

Little war story.

Today I had a bad start.

After feeling smart to enable this automation suggested by Atlassian Intelligence to move all tickets with no more open pull requests automatically to “deploy” once the PR is merged, I felt so dump about not triple checking this.

AI suggested:

I felt like something is off, but I also thought that looks pretty reasonable. And for such a common thing, could AI be wrong?

But unfortunately it was not!

Oh my god. The automation moved ALL TICKETS to DEPLOY, even those closed years ago! What a disaster. Fortunately I had a backup. Well… Actually I did not With some JQL magic (development[pullrequests].all > 0 and development[pullrequests].open <= 0 OR (updated >= "2024-12-16 8:34" and updated <= "2024-12-16 8:37") I was able to move all tickets back to done.

How is your experience with Atlassian Intelligence?

P.S.: I didn’t gave up on that automation, and here is the correct version:

[tied]

Nobody ever should trust any AI without carefully checking. It’s not an “Atlassian AI” problem.

[LukasGotter]

I know I agree and I feel like in this context it’s especially misleading - Automations need to run 100% precisely! An SEO text can be off by many percentages these technical things can not…

[MattDoar]

And use a sandbox for things that can have a wide impact. Glad you were able to revert the change