IMO these are anti-patterns that makes code smelly.
A practice I like to follow is to document "hacks" and create tickets for TODO items. In my experience, this has scaled better with the growth of the team and also allows for knowledge sharing and brainstorming when planning.
For TODO items I used the "eisenhower matrix" to decide if I should create a ticket for it or not. Where TODO items that are not "urgent" nor "important" are simply ignored until they reemerge again.
There are few phrases that I like to mention when I see devs add TODO comments:
- if in doubt leave it out
- every line of code is a potential liability that needs to be maintain and tested
- in the future requirements may be different or no longer relevant
A very important topic. Specially if you are into MLOps!
If your product is mainly powered by AI, and you found "AI-product-fit", then this topic should be something you bring up to the AI team tomorrow or to the infra team.
AI models are powerful yet sensitive, and how new samples (out-of-dist) could affect them is hard to anticipate.
The other day I came across https://www.raymon.ai/, "Observability for AI systems
". I'vent tried Raymon yet, but looks promising.
It's pretty terrible if you compare it to CircleCI or GitLab from 4 years ago. I'm a big fan of GitLab, seems like the only company pushing things forward in an _elegant matter_, used it heavily in the startup world. Using GitHub again these days. I cry every time I need to do GHA stuff. Current setup of Github + CircleCI is miles more elegant as it was 6 years ago.