Input quality * LLM quality = output quality
Assuming only few companies tackle LLM quality all other companies will have to focus on input quality and output tuning
AI generation doesn’t replace fundamentals, it makes them more valuable. Knowing the basics, having taste, and building real experience are what let you guide the model, judge the output, and push it toward something great. The tool can generate faster, but craft is what gives it direction.
Text generation doesn’t make you a great writer
Image generation doesn’t make you a great photographer
Video generation doesn’t make you a great filmmaker
Code generation doesn’t make you a great engineer
Great writers, photographers, filmmakers and engineers can move faster and do more using generations.
Do it yourself when it’s a one-off or when your unique taste, judgment, and touch are what make it genuinely rewarding. But if it’s repeatable and nobody can tell whether it was handcrafted or generated, delegate it or build a better tool. Save yourself for the part only you can do.
Your AI agent can answer questions about documents, but can it answer questions about what your team actually decided in email last week?
Threading, permissions, attachments, who said what. What was a quote vs. original text.
Email context is 10x harder than document RAG.
One API call: structured tasks, owners, decisions, citations.
https://t.co/U4T0FNKWHX
consistency ≠ uniformity:
for those who haven't lived through this era – we used to have beautiful, precision interfaces. now they're replaced by a design language that originated from the Apple Watch, with icons that only fit in squircles. but that's not even the point.
the point is we used to design the whole stack – the technology, the concepts, the interfaces. when designers only care about superficial consistency, platforms lose their uniqueness.
apple used to design systems. skeuomorphism wasn't just about leather textures – it was about teaching people new mental models. the trash can empties because you understand what a trash can does. aqua's lickable buttons and sheets had depth because the OS had layers you could understand. the old apple designed the whole stack – from metal to pixels to concepts. teams weren't just shipping features in the same box, they were building coherent platforms each with opinions about what computing should feel like for the medium.
liquid glass is fine on a phone. but on macOS it's unusable – lack of precision, visual noise everywhere. this is what happens when UI language designed for fingers bleed into macOS. we went from interfaces designed for a 27" cinema display with a precise cursor to interfaces designed for a 1.5" screen you tap with one finger.
the Mac is for creation and precision work. it needs information density. it needs chrome you can grab. it needs UI that gets out of your way but gives you power when you need it. instead we got padding and whitespace and translucent blurs optimized for touch targets nobody's touching.
the squircle icon mandate is a symptom. when you force every icon into the same shape, you're saying "brand consistency" matters more than "each app icon needs to communicate its function instantly." we traded clarity for uniformity. we traded precise design for cross-platform sameness.
consistency means your system has coherent rules within itself. uniformity means everything looks the same regardless of context.
the hardware team still gets it – that macbook pro + M-series chips, chef's kiss. but software design feels like it's chasing fat fingers instead of remembering what people do on a Mac.