Привіт,
Думаю, багато хто хоче піти з найму й працювати на себе.
Ті, хто таки мав сміливості та сил/фінансів і зробили це - чи справдилися ваші очікування, чи вдається вам отримувати саме те, що ви хотіли?
Які підводні камені та неочікувані речі ви виявили для себе?
Not defending Google here. Limits are annoying, despite that simple text messaging still does not consume too much. They also removed AI tokens.
But take a look:
GitHub Copilot tightened limits - now it is usage-based.
Claude team always speculates with limits too (especially after they cut programmatic claude -p usage).
Cursor - you can burn hundreds per day, but before you could do anything with cheap subscription.
This isn’t just 'Google being Google', but I think it’s the end of the subsidized AI era.
For the last two years, lowest subscription felt like an all-you-can-eat buffet for frontier models, but that OBVIOUSLY was never sustainable, they cant burn cash oceans.
This all comes to the usage-based payments in the future, which I don't like as well, so looking at running local language models on my machine.
Hey everyone,
I’m an Android-focused (Kotlin) React Native dev and I’m curious:
Are there indie devs here actually making a living from Google Play alone, without publishing on the Apple Store?
Would love to hear real experiences - whether Android-only is enough in practice this. I am for-life Android user myself and researching this area.
Thanks in advance!
Everyone's yapping about Codex lobotomization, but this dude has always been my last and solid gateway in cases when Opus starts doing weird things with wrong assumptions and overengineering.
Even today. React Native app. I had a problem with MSAL multi-accounting bug for native devices while web was ok. The fix appeared tricky one and I honestly had not time tracing it, so I handed it to Claude first (with CLAUDE[.]md, claude rules, etc...). This mf made +40 - 21 diff change, only masking the problem and causing regressions. Okay bud, I mean, I do git reset --hard HEAD.
Pasted same prompt to Codex with no rules for him - one turn, +3 -0 diff, right into the spot, and dude even checked via git blame who caused this bug (not me haha).
This AI race is 'Today X is better, tomorrow is Y', but this always seemed for me I come back to codex often after tried everything.
Anthropic just ripped off everyone and they still managed to make it sound deceptively friendly, beneficial even.
Anthropic just announced that the 200$ plan, which used to be okay to use in the SDK, is now "finally ok to use in the SDK" (except rare fraudulent cases it already was)
but for that you need to toggle an option that makes the SDK hit in an allowance of 200$ worth of credits, instead of the previously, opaque, yet, ultra subsidized limits
basically if you use claude code through the SDK or claude -p, for instance through Conductor, github Actions or anything else, because let's be honest TUI sucks sometimes (react in terminal cmon), their cloud coding agent offering also suck, and don't cover most usecases,
well, your plan that used to be worth 2000$ of tokens aproximatively depending on how much you flirted with the previous 5h windows limits, is now worth 200$
but they brand it as "oh we give you 200$ of additional credits for the SDK" so to the most unaware users it sounds like it's worth 2x more!
in reality it's now worth 10x less
*not surprised if they shut down this thread
That bud can drive me half day under high load (I am a Mobile Developer), from 10 am to 4 am and still some 20% left. Wire it up for half an hour while having lunch and you are good to go for the rest of the working day.
My 3 y.o. Lenovo with Windows 11 can do maximum of 3.5 hours with 3-5% battery left. How amazing is that battery optimization!
>DeepSeek TUI is a coding agent that runs in your terminal. It can read and edit files, run shell commands, search the web, manage git, and coordinate sub-agents from a keyboard-driven TUI.
>It is built around DeepSeek V4 (deepseek-v4-pro / deepseek-v4-flash), including 1M-token context windows, streaming reasoning blocks, and prefix-cache-aware cost reporting.
The repository: https://github.com/Hmbown/DeepSeek-TUI
Trending source: https://checkmygit.com/trending
Hello guys,
I'm handling OSS project with ~500 start on github and around 10k users. I do commits regularly. The stack is full on Cloudflare + SvelteKit.
Where can I see exact requirements for Project Alexandria application, if I am even eligible?
Thanks
Now 90% of trending repositories are AI-related: skills, agents, harnesses.
> Coding has evolved to organization, orchestration, problem-thinking
, - they say, yet Claude can't fix one-liner bug during peak hours. Tragicomedy.
Built this over a few weekends because I wanted a quick way to share my GitHub work without throwing together yet another portfolio site.
You drop in any GitHub username and it generates a portfolio page with different templates options (some more coming soon).
You can export the whole thing as a PNG or just share the URL like checkmygit.com/username?template=bento.
No login, no sign-up, fully open source. SvelteKit + Tailwind, hosted on Cloudflare.
Live: https://checkmygit.com
Repo: https://github.com/whoisyurii/checkmygit
Would love feedback, bug mentioning, feature requests!
Have a nice day yall
We have an extensively established CLAUDE md and rules in our project. Opus 4.6 was strict and compliant as hell, but 4.7 simply ignores all the rules most of the time, despite that I see it 'Loaded <rule 1>' etc etc. I see less sense to keep x20 subscription day-to-day lately.
... and their only option is Mistral. What are recommendations for coding with Mistral? Anything in AI plans, or shall we go with API? The product is not very big, we are 2 mobile devs and 4 backenders.
After 6 months of daily drive:
Love it. Learning Rust to become one of those golden contributors.