I think this assumes far too much passiveness in the human population and treats the problem of eliminating humans like any other passive engineering problem.
"Good enough" works for UI and database performance, or problems that aren't actively adapting to prevent the solution from being applied.
Nobody in security would apply "good enough" to well-known and high value attack vectors, you need to be constantly adapting and understanding your enemies next move. And in cybersecurity, you really have a limited surface area for attack.
Drones and robots need to apply their solution (getting rid of us) in the real physical world, not the digital one and each one of us is a supercomputer which can adapt to working with other supercomputers very quickly in the face of adversity.
I think they might make a good first push and catch us off guard, but some humans on defense will understand the systems that control the drones and how they collect/process the information. They do not have the same advantage with us.
Designing new camouflage to avoid detection, creating decoys which waste their ammunition/resources or spread their forces too thinly, and various other classic strategies that come naturally to any group of humans with a single strategist amongst them will be well outside of a simple "good enough" drone's ability to adapt and combat.
Anything but a perfect annihilation will only give us time to find exploits and counterattack. The only thing that should be a threat is an AI that has a similar capability to understand the physical universe and adapt faster than us. You might make a robot that understand a whole bunch of stuff about physics and biology and math... but if they don't understand chemistry and the many ways we can deploy it as a weapon against them (think acids and other corrosives), then they're going to be at a complete loss as to how to neutralize that threat.