I would hazard because peoples change faster than we think.

There is no good drywall anchor. If it weighs a pound use double sided tape. Otherwise cut out the drywall, put blocking in, and patch it. It’s paper and mud.

FWIW the codex high model has been reliably outperforming claude code for me across several domains if I disable its sandbox (isolate it on its own machine instead). Worth a couple hours experimentation.

Filip's analysis is supported by the various attempts I've seen to compare RC generally as a GC algorithm: it uses more CPU time to manage the same amount of memory. It can be more predictable about when the CPU time is used.

Unless you are heavily concurrent, then RC is awful all the time.

tools are really bad at surfacing co-authored-by :(

brad+vibing@ nice

Small airline!

Android is a good chunk of the way there (and in fact is the pioneer, with MTE), so I expect it to quickly follow. The magic here is extending MTE to the point where we can rely on it, so we can get on with programming.

Until it hits non-mac laptops… dodge security teams? :)

Much more dramatically: I would shift some software from garbage collected languages to C++ (or some equivalent) if the safety concerns evaporated, because it becomes a pure productivity/performance tradeoff. Exciting times!

Nah. I'm not paying some huge productivity hit just for that. We avoid crashes by investing in testing.

C++ is good enough again? Will any arguments for productivity-costing modern language techniques like borrow checking survive if memory tagging actually works? Will security teams ban sensitive work happening on x86 laptops and desktops for 5+ years? Lots of change ahead!