Basically in any other country like the US or Japan I’d say nobody cares, but in the EU this shouldn’t be legal, we also need to have some benefits living here. We have all the costs (tax, products not available, Cookie Banner, plastic lids etc.) then they should at least give us something back (apart from forcing apple to use usb c 🤣).

idk man, it’s more nuanced than that and most of the OGs are just posting rage bait. It’s good for engagement. And it’s also not completely untrue. Software will get cheap af. Will everyone use personalized software? Def not. For some SaaS this will be the case for others not. Especially games. Ppl want to share experiences, they don’t want to play their own game nobody knows and they can’t talk about with their friends. Also at least in the next five years chatGPT on the fly generated software will be very rough around the edges. Ppl will pay money for a good piece of software but just not a subscription which costs 10 bucks a month. Those days are def over.

I still say Twitter so guess how renaming AI to SI is going to go 😅

Well, it's not like they already made the comeback, but interesting that a lot of ppl already think Google will have the best model by end of October. What ever *the best model* means these days. I'd say it's not only about the model anymore and it's pretty hard to tell how big of a part the model harness plays performance-wise. But def excited to see what's coming.

Wonder why they released Sol 6 in the first place, it’s just like a week ago.

Def remembering happily going to the computer first thing in the morning like "let’s see what my agent built over night", just to find the agent asking for permission after the first command 😂

The problem with plan mode is: If the feature is small, you don’t need it. Just tell the model to implement x. If it’s a bigger feature I usually start with a higher reasoning level, but later want to implement it with a lower tier or even a different model. The plan is buried somewhere in a codex or clause sub folder. This makes it hard to review and implement with other models. Instead I start a conversation about a feature in non-plan mode. Get feedback, let the high reasoning model write it as markdown and save it in plans folder. User other models to evaluate the plan. Go back and forth bewegen models and update the plan. Then finally implement it with a lower tier reasoning model. Then tell a different session to check the uncommitted changes against the original plan. So plan mode is not so useful for me.

If you have 10 gmail accounts, one for each project, then you'll probably haven't heart of migadu yet (no affiliation). Check it out, they have a website like the early 2000's, suck at marketing, but the product is great!

They want to compete with Apple so much they called it Googlebook which is really close to Google Books and doesn’t sound exciting at all. Microsoft has done a better job here with its Surface series.

Yes, and it’d be totally fine, if Astra delivered high quality output, but the thing is, it doesn’t and feels stupid at times. So we have a limit that doesn’t last long combined with outputs which are sometimes ok sometimes not, it’s not that great as a package tbh.

Doesn't look good tbh, Astra keeps finding bugs, which is a good thing, but it burns through the weekly limit pretty quickly.

Bc they’re waiting to have something to go against OpenAI. Fable 5.1 is too token hungrig, limits decreased by 17% and everybody is talking about Astra and hating Opus 5. They’ll probably try to be back with Opus 5.2 soon.

Already brewing

The last days bc I canceled. Comparing Opus 5 to Astra and tbh I‘m one of the few ppl on here actually liking Opus 5. it has strong beliefs and keeps challenging you. Astra is always like, yeah sure you‘re right.

Basically my workflow is: 1. Brainstorm a feature (no code edits here) 2. decide on a plan (the model writes this in a md and pus it in the plans folder) 3. discuss plan with other models (also in different apps, e.g. Claude). 4. give feedback back to original model and let it rewrite the plan from scratch 5. implement plan with a fitting model (depending on difficulty) It would be great to have something in codex to make this process smoother. And I know there is plan mode which I don’t like very much, bc the plan lives in some codex subfolder and it feels hard to iterate on. For starters a button like "brainstorm mode" would be great, just challenging ideas. @ajambrosino is always looking for feedback, hope this helps.

No I think it was the other way round, limits went back up for no reason. I was using Astra a lot, had it at 80% in the afternoon and later it was back at 92%. They fixed it and now it’s down to 70% 😅

wtf? this is greedy iphone indie dev style, lol!!!