I speak opus so I can help.. ahh I see, the classic fringe case blindspot..
To determine if you have the energy, ask Claude what is the likelihood percentage of this issue, and which suggestion provides more value to the user. From here it appears to be a case of omit and notify versus solve, and solve "appears to be more valuable", so I'd say the latter, but check with Claude to be sure. ๐คฃ
Workflows are one thing that can be explained away as the techs using the same gift. But there's a secret monster here I can yell at the top of rooftops that people won't accept until they see it and that's productivity become more granular.
It's a gold rush opportunity for all three sides: the developer, the business owner, and platforms that want to help with the impending modularity needs
Iโve run into the same thing. The density is real, but I donโt think thatโs the main issue. Weโre capable of absorbing dense information when we take it at our own pace.
What actually throws me is how the model compresses everything. It often skips the connective tissue, leans on pronouns and ambiguous nouns instead of clear subjects, and leaves almost no transitions. Even when you slow down and focus on a short stretch, itโs easy to lose track of what the its referring to. The occasional off-synonym that only makes sense with full context doesnโt help either.
It ends up feeling a bit like listening to someone tell a complicated story who keeps saying โsheโ and โheโ while jumping between multiple people. The individual words are fine, but the map never quite locks in. Scale that style up to topics that normally take years of context, and it starts feeling cryptic. Not because the model is necessarily smarter than you, more like itโs an intelligence that compresses differently: stronger in some places, thinner in others.
What helps me is steering the output toward explicit subjects and clearer structure. A few targeted instructions (or โskillsโ) that force it to name things cleanly go a long way.
@klewintsbuild It's like lifting weights. Keep at it and the pathways strengthen. It will come easier and stops feeling difficult. Turn that jelly-brain rock solid!
They're cheaper and faster, and you don't always require benchmark winning logic. It's the biggest bang for its buck.
Plus, Google is doing godlike work in the medical field for all you people that think they're sitting on their thumbs.
I don't love to hype conglomerates, but teams are doing amazing things and I give credit where it's due.
I have a diff script running that I read religiously, and don't push anything local without verifying, then I test and git push. It's good to be in that habit, and AI works so well at this point that I either generalize a fix or give it the proper code when needed.
But I weight quality over expediency, which I can for my use-case. And tbh, I'm sure more people would save time prioritizing quality, because when you don't, the fix takes longer than the build.
It prompted a review and a transparency disclaimer. Nothing odd.
But what I think we are seeing is what happens during an arms race where guards are hastily thrown together and the ends justify the means. I don't think it's that bad, but I am sure we'll see conspiracy theories and hyperbole, especially coming from competitor shills