They probably aren’t experts at delivering Software as a Service (SaaS) or as a product to your customers
Don’t make this big machine learning mistake: research vs application
George Seif

Exactly. Most scientists I know are terrible programmers. I’m a chemical physicist in simulation science who has specialized in programming solutions in C/C++/MPI/OMP/NVCC/Go/Python/Haskell/assembly/Javascript/etc., including applications of deep learning to ab initio chemical simulation (quantum mechanics), one of the latest applications of DL. I’ve written finite element and density functional codes with proper style and implementation through an API for reuse. I know good software.

But I’m not a linux administrator. But the solution to that is just to have one on the team to provide deployment guidance to make the software features fit the hardware specifications and target audience size.