Meet Patel

If by next month (before June) Google releases the best version of Gemini 3.x Pro, then it would be the best. "Best" means properly priced, not token-hungry, and having proper intelligent and reliable tool calls. That would be it. People would happily pay even for API pricing.

The problem I feel with the Indian tech industry is, they focus too much on showmanship & are eager for output They try to sell mediocre stuff, claiming it's the best. On the other hand, Western/Chinese companies build incredibly good products & let the product speak for itself

@championswimmer I read your one & understand your direction you're trying to say. What i meant is this - factory.ai/news/compressi…

@___meetpatel___ Not a bad idea to read the entire thread and/or the docs in the repo :)

built a Pi extension called pi-context-prune because agent sessions are absolute context hoarders it: watches completed tool-call batches summarizes them drops raw tool spam from future context keeps originals queryable via context_tree_query less sludge, same memory

@neogoose_btw I used to think he was a genuine one to learn from. But he just over-explains stuff and over-engineers most things. (Only occasionally did he have proper content; the rest was just over-engineering.)

bro how do you know what skills needed for real engineers 😭

@DhravyaShah Capturing half backed & non accurate data is never usefull. No matter how much fast it's. Dynamic pages, All JS rendered important data & any non static dynamic content will be missed.

forgot to mention, it can pull any website (even larger ones like vercel) in about 15 seconds max!

@ajaypatel_aj Yes💯

@___meetpatel___ Yeah, it’s pretty solid. Gemini 3.1 Pro is doing some interesting things with design and UX.

Man, if you’re not using Gemini 3.1 Pro for design, then you have no idea how smart and unbeatable it is. It understands not just design but real UX, man. I’m just mind-blown !! look at this design and its UX. Not even in world's best designer could come up with this UX !!

MacBook Pro or MacBook Max (48GB+ only) will be the cheapest and best options now to get local inferences (not good speed of t/s but usable somewhat). Prices are going to shoot up like crazy soon. Buy before prices go uncontrolled!

Talking about qaulity of harness, you can now build sonnet 3.5/4.5 level workflow easily with local/cheap models. For imprtant & big task you can always pay few bucks & get your job done. Just like you don't need 1000 MBPS internet all time, pay for it only when you need it!

Never want to use token hungry harness like Copilot or Claude if I will be charged for unknown so many token they ponder it. If i have to pay for every token then i should have full control over every token as well.

There is something i remained calm for long. I knew someday usage/per token based price everyone has to face. Cursor/Copilot/Claude or anything else So owning harness & whole agent workflow will be most important, that's why i build private product to own entire flow & harness.

@staysaasy 😂😂

“Everyone can code now!” Dude, no one can code now.

If you have to pay for every AI-generated token, you should have full control over your agent harness and be able to build it yourself. Seeing a live editor and tester to control AI agents’ tools, those magical moments are coming soon. See you in blast mode.

@ChrisLaubAI How it compares with @lightpanda_io, which is also written in the very memory-efficient Zig language.

A Rust dev just killed Headless Chrome. It's called Obscura. The open-source headless browser purpose-built for AI agents and scrapers at scale. Chrome vs Obscura: - Memory: 200MB+ → 30MB - Binary: 300MB+ → 70MB - Page load: 500ms → 85ms - Startup: 2s → Instant - Anti-detect: None → Built-in Single binary. No Node, no Chrome, no dependencies. Stealth mode is brutal: → Per-session fingerprint randomization (GPU, canvas, audio, battery) → 3,520 tracker domains blocked by default → navigator.webdriver masked to match real Chrome → Native function masking so detectors can't sniff it out Drop-in replacement for Puppeteer and Playwright over CDP. Zero code changes. If you run agents or serious scraping at scale, this repo prints money. 100% Opensource.

You won't need to spend a crazy 500 to 1000 dollars, which can be done with 50 to 100 dollars if used wisely. Use local models (if you can) for the majority of Agent Building, or use the web version for the majority of testing of (as they are free) MVPs & Experiments.

After spending ton of money on diff providers, I realised best way to use AI is to not rely on any single one. Ex - use OpenCode Go (for daily routine & easy to medium task) Use Codex plus/pro for routine code cleanup task Use Opus/Sonner for UI task Use Gemini for data tasks

My thoughts on the future of IT. Everyone loves launching products as code gets cheaper (not code quality), so eventually, we'll have very few genuinely high-quality software left. Only the best ones will survive. Distribution will come from quality, not from followers or SEOs.

@paraschopra 😂😂

AI bois be like:

@OpenAI Hopefully for UI it will be as good as Opus or Sonnet. Then I am fully sold !

Introducing GPT-5.5 A new class of intelligence for real work and powering agents, built to understand complex goals, use tools, check its work, and carry more tasks through to completion. It marks a new way of getting computer work done. Now available in ChatGPT and Codex.