Hey, if your point is that too many PHB's and programmers think "managed" is a cure-all, I won't stand in your way. What I'm saying is that managed is a huge win for security.
The hardest security problems to solve aren't the overflows, it's the features given to users.
By contrast, the most common security problems are any situation where you silently expect the programmer to manually preserve some invariant (e.g, never allocate memory without a plan to deallocate it, never deallocate if anything else holds a pointer to it, never write to a buffer without checking bounds, etc.). Managed languages eliminate C/C++'s largest (and most critical) attack surface.
Now sure, I agree that they don't eliminate all attack surfaces. Security is hard. Java/C# have their own "manual invariants", such as always escaping/parameterizing SQL. ASP.NET Forms have a nightmarish arrangement where some controls/properties auto-escape HTML and others don't. Crypto primitives are widely available but poorly explained. Multi-threading is a minefield. But even here, the industry can eliminate the widest number of security issues using secure-by-default design. In C# for instance, EF/Razor/TPL make it (1) easier to accomplish programmer intent while also (2) making it harder to break low-level invariants.
Think of VB macro viruses, that spread wildly in a managed language. Wordpress is another example of software written in a managed language with tons of exploits.
Office macros and PHP are some of the most hilariously bad designs in computing history. By necessity, any programming language worth its salt will let you make farcically bad decisions.
Notice (for example) his micro-agressions against people who understand garbage collection. The implication is you don't need to think about it, C# will take care of memory.......which if you take seriously, means you'll be leaking crap all over the place and someone like me will have to come clean it up for you.
As a Google developer, he can probably just throw clusters of auto-recycling web servers at the problem. Aside from opening avenues for DOS attacks, the consequences of this sort of problem (e.g., not knowing how your GC works) have more to do with performance/reliability than security (albeit the 3 are intimately linked).
Something we can probably both agrees on is that there's no substitute for knowing how things work. However, the reality is that most programmers don't care and even those who do have a limited mental budget for complexity. So there's also no substitute for being able to eliminate sources of complexity that are ancillary to the task at hand.