Post

At the risk of starting the flame war to end all flame wars... Modern LLMs (GPT-5.1, Claude 4.5, Gemini 3) produce excellent code and can be a significant productivity boost to software engineers who take the time to learn how to effectively apply them - especially if used with coding agent tools

I still haven't worked out how to get LLMs to avoid adding increasingly more complexity to projects. The projects I tend to want them to work on are like multi threaded C++ projects, so maybe I should just stick to getting them to write little python tools to make my life simpler.

I don't have any relevant expertise with multi-threaded C++, I would be surprised to hear they're much less effective at that The most important trick I know is making sure they have a test suite they can run - most of my Claude Code sessions start with me saying "first run the tests like this:"

I asked Google (no quotes): "how good is ai at writing c++ multithread code". Google's reply included limitations like this: "Multithreaded programming [involves]... issues like data races, deadlocks, and race conditions. AI models may struggle to identify and prevent these subtle bugs..."

In theory, I can envision a "code-understanding" system that could prevent race conditions (which are notoriously hard for humans to see). But I'm skeptical we'll see that delivered, in reliable form, any time soon.

OK, hmm. I'd be interested to hear from people who have tried them for that fairly demanding task.

i feel like i've already diagnosed and addressed race conditions with claude? in fairness, those were likely in code it wrote, so it may not have been the pernicious edge cases people are thinking of that said, it also wasn't monumental enough difficulty to remember well ... just a passing blip

intuitively from other similar runs, if i lead the horse to water, it will probably drink if i tell claude the general shape of the problem i see, it's often very successful at drilling down it's not as great at recognizing problem shapes, i think? (but that may be outdated, or soon to be)