Working with unreliable tools is tiresome. I have this Chinese €28 cordless driver, and it stresses me out every time I use it. It works perfectly fine most of the time. But very occasionally it has a hickup and it messed up a screw. It’s tolerances are also not great.

My problem with the Chinese cordless driver is that I cannot trust it to do a good job. Those sporadic moments it doesn’t perform, takes away the trust I have in it. I guess it’s mostly a mental thing, as it’s issues are only very occasional.

This situation is comparable with LLMs. At my client we use Opus 5.5 which does it’s job most of time. Not always. Also, Opus is a black box that Anthropic controls. Anthropic can nerf models without telling me, or it can pull the plug whenever they like. I cannot trust Opus to do a good job. Because I cannot trust it, I don’t like working with it very much. I like reliable tools, ones I can trust.