[xkcd-sucks]

This is the sort of thing that gets written when well-meaning software engineers write about stuff they know little about.

Testing early drug candidates on living critters is already a thing, and has been a thing since forever. It's referred to in the industry as "phenotypic screening." In vitro testing emerged because it's expensive and possibly unethical to use huge numbers of mice/digs/people to test drugs in the pipeline. There's been a resurgence of phenotypic testing as the miracles promised of various screening technologies promised in the 1990s-2000s have failed to deliver.

Iterative refinement of drug structures has also been a thing since pretty much forever. Similar structures behave similarly. Changing stuff is somewhat predictable, and modifications are introduced at all stages of the pipeline. For example, a structure might first be tweaked for receptor binding, then for solubility, then for ease of.manufacture, etc.

[osharav]

The article is not about testing on living creatures or about iterative refinement - it is about having the right unit tests to make sure that the iterative refinement predicts better the effect on living creatures. I am indeed a well meaning software engineer who write stuff I know little about. That is because I don't think I can quit my day job and study biochemistry for a decade. What I can do is point out similarities between stuff I do know about and the field of drug development. If you don't find this useful or it doesn't trigger any new ideas (maybe totally different than the ones raised in my article) - then read on to the next one.