It is made to perform much better than your typical electron app would. Saying electron-based == shitty is complete misunderstanding of the technology. Although i dislike Figma as much as the next guy, their app was in many ways very impressive. See Figma's cofounder old articles at https://madebyevan.com/figma/
Did I? Not only are you comparing apples to oranges, you even provide misleading numbers.
3090 gets 20-30 tokens a second for dense ~30B models (QwQ 32B, Gemma 3 27B Q4), similar to M3 ultra.
If you are talking about Qwen3-Coder 30B (MoE), then both 3090 and M3 Ultra are around ~70 tok/s.
But even if you were right about the speed - which you are not - speed is pointless if you need large model that wouldn't fit into your VRAM.
Are you aware that your 3090s have nowhere close to 256GB of VRAM?
Or maybe you are not aware that on macs you have unified memory (working both as RAM and VRAM).
I like Gemini CLI and I got a lot of value from their free tier, and I'm now on the $20 sub. But it is a level bellow the usefulness of Claude Opus 4.6.
What if I didn't know the words yet? Usage of totally unintuitive icons is obviously wrong. Colors can be helpful too. Colored text - not as concise as recognizable icons.
Another non-negligible advantage of icon is, that it is language agnostic. Not everyone is fluent in language they have to use.
There could be factories manufacturing your own design, just one piece. It won't be economical, but can be done. But parts are still the same - chunks and boards of wood joined together by the same few methods. Maybe some other materials thrown into the mix.
With software it is similar: Different products use (mostly) the same building blocks, functions, libraries, drivers, frameworks, design patterns, ux patterns.
That is not a technical constraint and may be automated if it made sense financially.
Same with software - for some time software won't be all designed, coded, tested, deployed to production without human supervision or approval. But the pieces in between are more and more filled by AI, as are the logistics of designing, manufacturing and distributing sofas.
I turned on this overspend and limited the spending to $20. A day later I checked my spending, I had used "295%" of my limit. Almost $60. No idea why it didn't respect my setting.
pro is the $20, right? It runs out quickly, especially using opus. But what do you expect for that kind of money? For serious work at least Max $100 is needed.
As an ADHD person, the landing page is absolutely anti-ADHD - a lot of stuff with basically no info about what it really does. It should have been all concise and tangible information, simple example, demo. Instead just a lot of marketing fluff. I spent all the focus budget there and I have no idea what it does.
I disagree with the vibecoding take. Its a new skill that absolutely has a place in developers skillset and it may be of great importance for some kinds of projects. You can learn so much by vibecoding little projects that otherwise would never see the light of day.
If you have inference running on this new 128GB RAM Mac, wouldn't you still need another separate machine to do the manual work (like running IDE, browsers, toolchains, builders/bundlers etc.)? I can not imagine you will have any meaningful RAM available after LLM models are running.
I'm on a team like that and I see it happening in more and more companies around. Maybe "many" does a heavy lifting in the quoted text, but it is definitely happening.
If you knew GraphQL, you may immediately see it - you ask for specific nested structure of the data, which can span many joins across different related collections. This is not the case with common REST API or CLI for example. And introspection is another good reason.
Using electron at least uses some UI primitives from chromium. Flutter has thrown away all the usability and robustness of existing components and just reimplemented everything. It absolutely is different from electron
Many of the bugs have very low severity or appear to small minority of users under very specific conditions. Fixing these first might be quite bad use of your capacities. Like misaligned UI elements, etc.
Critical bugs should be done immediately of course as a hotfix.
Regarding hyperrogue, I've seen it mentioned multiple times as a fun game. I tried to play it but I had found no fun in it at all. The non-euclidean take is interesting, but it felt just like a demo of the weird-geometry engine, I've found no enjoyment in that. Graphics is rough, I've found no interesting items, enemies, mechanics or puzzles. Not sure if I just played it wrong or why my experience was different.
I have the same experience as you. For me instructions in CLAUDE.md are followed almost always. On different projects, different CLAUDE.md files, some short, some long. No problem. When a specific instruction is skipped, I ask claude to emphasize it. It uses ALLCAPS, IMPORTANT!, etc., then it works 99% of the time. (Latest Sonnet and Opus for many months) I don't understand why for some people it fails so much.
This is quite a narrow view of how the generation works. AI can extrapolate from the training set and explore new directions. It's not just cutting pieces and gluing together.