[atomicity]

Still think that tests have a role, but there's room to improve them. While human contact is nice, it's also costly, so improving testing would still be a great way to innovate.

For example, I received a great coding test from Vertica, a database company. Their test was a 2-hour timed exercise involving improving a ~200 LoC data querying program. This is nicer for the following reasons:

- It doesn't require me to bootcamp with LeetCode because you have a bunch of people out there who train and "overperform" on such coding Qs.

- It's specific enough to the domain but still fairly generic. The test covers techniques covered in a standard intro sequence.

- It's timed, so it doesn't waste time like most take-home assessments.

- It's predictive power of t skiechnical ability is better (my opinion). People like to say that such tests don't test enough for problem solving. However, if you are missing 6 months-1 year worth of domain knowledge, you won't be able to contribute for a while no matter how good your problem solving ability is. An average candidate stays for <<10 years so you want the person who can start contributing sooner even if they have a lower "peak" potential. (Also, I believe generic problem solving is kinda BS. I was good at math and coding contests years ago, but now I would consider myself to be an average interviewer. Even though I'm an good problem-solver for certain systems problems, I don't consider myself one for ML ones or people-heavy problems.)

To summarize, I would love to see better tests than the ones we have now and think there's clear room to improve.