Lots of options for asserts other than show a dialog.

I don't think it's insane.

Historically, assertions tended to be genuinely expensive. However, this doesn't have to be the case.

1) optimizing compilers are *remarkably* better these days: chandlerc.blog/posts/2024/1...

2) you can engineer your assertions to be cheap (and you should)

We further *define* (b) to include any and all memory safety violations and undefined behavior.

The default is, throughout all of our C++ server code:

a) It isn't an *invariant*, it is just an error, and it is propagated explicitly to the caller

or

b) It is an *invariant* and the state is corrupt if not valid. These, when checked, are terminating by default.

Just to be clear, because it won't be from this poll:

All of Google's server C++ code uses terminating assertions by default.

There are debug-build-only terminating assertions, and non-terminating assertion tools available of course, but they are not the default.

We further *define* (b) to include any and all memory safety violations and undefined behavior.

The default is, throughout all of our C++ server code:

a) It isn't an *invariant*, it is just an error, and it is propagated explicitly to the caller

or

b) It is an *invariant* and the state is corrupt if not valid. These, when checked, are terminating by default.

Just to be clear, because it won't be from this poll:

All of Google's server C++ code uses terminating assertions by default.

There are debug-build-only terminating assertions, and non-terminating assertion tools available of course, but they are not the default.

I have heard it more recently casually (and imprecisely) used to refer to taking out a home equity loan (often a HELOC) in addition to a mortgage.

From what I can find, pre-1930s the loan-to-value was often <50%, and often under 60% through 1950s. Having trouble finding good data though.

Definitely was a phrase when I was growing up in the US.

My impression is that it mostly became uncommon when mortgages started commonly being for ~most of the value of the house and so you couldn't realistically take out a significant "second" mortgage. Guessing the term was more common pre-1980

We'd rather have racist billionaires, and coal mines, and cut foreign aid to Africa, and start wars with South America, and not sell soybeans to Asia, and let the 50 dudes that make up our entire our capital allocator class, pretend that monkey jpegs are money now.🤡

So here, take your monkey jpeg.

Not "gen-AI more broadly" -> outside of coding tooling. Outside of coding, completely different set of questions around generative AI for me.

My question is specifically around coding tools.

Regarding security - yes, I do understand. There are reasonable ways of running fully untrusted code, etc.

For example, Claude Code or Gemini CLI, etc. Largely the agent based coding tools.

I understand that you may feel these are strictly inferior to purpose-built classical tools, but there remains lots of interest in these approaches. My questions are about how to approach that interest.

So that makes me wonder: what reasons to folks see to view AI coding tools (note, NOT gen-AI more broadly) necessary to completely preclude engagement with? What goes wrong with allowing users to use these tools if they want, and trying to have the needed guard rails and other factors for safety?

Sometimes, I agree.

Other times, I think the best way to engage with users about how they are misusing tools is to integrate the tools and use that to drive a deeper discussion instead.

TBH, I'm not 100% sure where AI coding tools falls here yet, and that's the point of my questions.

TIA for genuine and engaging answers! Interested in broad thoughts on this space too.

Please, no taking the discussion into all of the non-coding use cases of gen-AI -- its a super interesting topic, but want to keep this thread focused on the narrow area of coding tools.