DON't disconsider this nightmare scenario too much - While engineering (as in classically designing, prototyping etc.) such a grey goo might be almost impossibly-hard because of chemical energetic constraint and the obvious fact that 3bn years of evolution were not enough to bring about such effective "grey goo" that could "consume the planet", there are some chilling new developments:

We're starting to be good at (1) using machine learning to "learn" designs for components much faster than naive-genetical-algorithms (aka "evolution"), and this will only get better as ML improves, and (2) we're getting better and better at simulating-in-software biological-like processes (and quantum computing will help with this). So we are pretty much in shape to vaaaaaastly outpower 3bn years of evolution.

A virtual + physical nanobots-evolution-gym where AI agents evolve and compete at creating better and better self-replicating nano-machines will give you the nightmare scenario easily...

Basically anything related to the combo of "ML + nanotech" should be carefully watched and monitored!

(And if you're willing to not start from scratch but instead use existing bio-building-blocks like viruses and bacteria, and don't have as high goals as "consuming all matter on earth" and instead you settle for "most multi-cellular living matter" then technically we already have the capabilities. You probably know what I mean.)