> Writing new systems that don’t need to exist simply to scratch an itch to write (NIH syndrome, etc.) I've seen this so many times and it almost always ends i

> Writing new systems that don’t need to exist simply to scratch an itch to write (NIH syndrome, etc.) I've seen this so many times and it almost always ends in failure. Rewrites as well. Engineers can't resist improving a system while rewriting it (adding caching, redundancy, etc). They neglect proving core functionality is intact before making improvements. They also often neglect the thousands of man hours it took to find all the bugs in the system. I've been building ML systems for a while and I've even seen engineers coming up with their own search/optimization algorithms. This always fails. It's hard enough to get standard techniques to work, since your data usually differs (slightly or severely) from what the technique was validated on (standard, cleaned data sets used for journal papers).