Zafar Abdurashidov

[Zafar Abdurashidov @zafar_io]

@PashaBorsai solid effort man. would love to test it

[Zafar Abdurashidov @zafar_io]

@alexcloudstar cool format, one question though - how are you going to handle distribution when you have 5-10 apps 10 days in?

also “I can’t stay on one app for long” made me smile 😁 - because once something takes off, you'll have to stay on it. the good kind of problem I guess))

[Zafar Abdurashidov @zafar_io]

latency based routing makes sense, but subscription tier in the gateway - that's a product decision, not an infra one

and the day you need to change smth, your business logic ends up in two places - half in code, half in someone else's dashboard

where would you draw the line - what's ok to move to the gateway and what's not?

[Zafar Abdurashidov @zafar_io]

@kunchenguid I’ve seen people flex those pretty $ numbers priced at api rates while clearly being on a subscription, but I haven’t seen anyone actually claim these are their real costs

[Zafar Abdurashidov @zafar_io]

@jscnjay where’s this screenshot from? TikTok?

[Zafar Abdurashidov @zafar_io]

@otherelias @alexii_9 what problem are you solving?

[Zafar Abdurashidov @zafar_io]

been doing basically this with Playwright MCP - ran it against my own app yesterday, 29 issues in ~40 min, 6 edge-cases worth actually fixing

biggest friction was stale element refs after re-renders. looks like browser-use avoids that by re-reading the page every step - how does that play out token-wise on longer runs?

[Zafar Abdurashidov @zafar_io]

@iamkarlstix @TTrimoreau 130.9k replies but 21k posts total on his profile. so that number is replies to him, not from him

still ~75 posts a day for 9 months. method’s still real

[Zafar Abdurashidov @zafar_io]

@shivanshco @levelsio congrats man, that's huge 🫡

[Zafar Abdurashidov @zafar_io]

but at the same time, back on SO you could immediately see what other people thought - votes, arguments in the comments

now that the data is private, the whole responsibility for the quality of the solution is on you. and the data itself is different, without that validation from actual people

[Zafar Abdurashidov @zafar_io]

feels like the content didn't disappear, it just went private. a bug used to turn into a stack overflow question that thousands of people read, now the same bug gets solved in a chat and one person sees it. the data is there, just at openai and anthropic instead of the open web

but that only works for stuff that lives in text. reviews, trips, "I used this for a year" - you have to actually go and do the thing, and that content is still made by people only

AI hasn't gotten there yet. at least not yet))

[Zafar Abdurashidov @zafar_io]

tried this on my own codebase. reads way better honestly, no argument there. but noticed one more thing - asked something like "should I do it this way here or is it fine as is". STE went straight into command mode, just "do not". the normal run said "yeah but not like that" and threw in a warning that wasn't in the STE version at all

same recommendation in the end, different delivery. makes sense though, a standard built for manuals behaves like a manual

[Zafar Abdurashidov @zafar_io]

@VadimStrizheus price point aside, what’s actually better about it?

[Zafar Abdurashidov @zafar_io]

i always write unit tests. the product itself i just click through by hand 😅

edge cases, regression - never get to it. checking the ui takes forever, and things ship too fast for that

so a client tells me something broke

no idea what the move is - what do you use?

[Zafar Abdurashidov @zafar_io]

@cryptopunk7213 everyone I know who tried claude code got hooked instantly

[Zafar Abdurashidov @zafar_io]

@tibo_maker AI isn't killing SaaS, the money isn't leaving the market. it's just moving - from the ones who didn't adapt to the ones building with AI. same market, different players

[Zafar Abdurashidov @zafar_io]

@jonathan_wilke @claudeai tried the desktop app for a while but it felt a bit raw. switched to CLI and once you get used to it, it just has everything you need

[Zafar Abdurashidov @zafar_io]

@GergelyOrosz soon "written by a human" will be a marketing label just like "organic" on food. and just as hard to verify

[Zafar Abdurashidov @zafar_io]

remote control in Claude Code is genuinely a great feature. I start a task on my laptop, go to the kitchen to eat, and I can see what's happening and jump in from my phone if needed

one thing that's missing though - voice-to-text. the regular Claude chat has it, but remote control doesn't

typing long prompts on a phone is painful, so right now I have to go to the regular chat, record voice-to-text there, copy it, go back to remote control and paste it in. kinda silly workaround honestly

@AnthropicAI would be great if voice-to-text just worked there too. or am I the only one doing this? 😅

[Zafar Abdurashidov @zafar_io]

@felixrieseberg love how AI agents are becoming something you just have on you at all times. less of a tool you sit down to use, more like an assistant you can ping from wherever