Igor

@steipete @openclaw oh this is just dirty, they made you rename, but then decided to roll with their own play on it (from anthropic site)

@bridgemindai I will say Deep Think is pretty cool. Much more 'acceptable' than ChatGPT Pro, and the 2 together are a monster

Heh I did same last week, Gemini CLI was very meh, I couldn't even get Antigravity to work (was throwing some error that apparently people in the forums were dealing with for the last 2 months with no Google involvement). The worst was, for the most expensive plan in the industry, 0, and I'm not exhaggerating 0 support. I sent 2-3 emails, tried the contact form, nada

Just cancelled my $250/month Google AI Ultra subscription. Gemini 3.1 Pro is inconsistent. Gemini CLI crashes mid-session. Antigravity loses context on complex tasks. $250/month for a model I can't rely on. Not worth it. Claude Opus 4.6 and GPT 5.4 stay on the stack. Gemini doesn't.

@ivanburazin @GuacamoleM5405 I think you hit the nail there, 'useful'. Are they optimizing cost, or just burning through them

@GuacamoleM5405 Its not an ideal metric- I get it. But, you want all people to use as much useful tokens as they can, and the more they do that the more productive you are as a company.

I recently met a founder who has an engineer spending more on Claude tokens than his actual salary. His goal: entire company spends more on tokens than people by end of 2026. Just imagine... $150k engineer → $300k/year in token spend Curious to see when the flip happens at scale in more companies.

Possible as a competitor, one of the big moats for zapier was always the amount of connectors they had, which these days isnt much of a moat. If you're building for yourself, if probably is worthwhile due to freedom to make any workflow you want. But you'll never get rid of crashing :) you just become responsible for it

Zapier is down. Whenever this happens now I consider building my own version of what that service did for me. And I'm sure I'm not alone. 1. Maybe as good or better 2. One less subscription That's a hard environment to run a service in. You have to be unreproducible.

@SeanGroff @trq212 Well damn, I fully forgot they bumped it to 1m, now myu 70% is like 700k lol gotta drop thatm, thanks damn thing been super dumb lately

@trq212 What do non-benchmarks say? I'm a power user paying full API rates. Opus degrades after 200k context, and sending huge prompts back & forth is very expensive. Love you & team's work, but worried you're not facing reality with this post.

i think we might have undersold 1M context tbh, the performance is so so good, I really just don't clear the context window much these days

@MrBallaz Codex 100%

Holy shit. I'm cancelling my Google pro subscription. Antigravity has been so unusable lately. If you had $20, which of these subscriptions would you choose? -Cursor pro -ChatGPT pro (Codex) -Claude pro (Claude code) -Stay with Google pro (Antigravity)

@Bencera I dunno, i've yet to see one of these work, if you can do it, more power to ya

@Bencera Is this a modern version of a course on how to sell courses?

About to hit $4.5M run rate. Still 1 founder + AI. Zero employees. Honest moment: this past week almost broke me. No one prepares you for what PMF actually feels like. Every infra partner hitting rate limits. Every bug that could happen, happened. Investors throwing big numbers at me. Customers flooding every channel. All at once. I went silent. Stopped tweeting, stopped LinkedIn, stopped podcasts, stopped growth. Just me and my AI agents, fixing things one by one. Here's what I learned: everything is solvable with AI. Every single thing. I'm building Polsia so every solopreneur gets access to the same tools keeping me alive right now. If I can survive this alone, I can package it for everyone. The future is solopreneur + AI. I'm living at the edge so you don't have to.

@thsottiaux Linux support for the app please

What are we consistently getting wrong with codex that you wish we would improve / fix?

@aiinstinct @thsottiaux yes!!!

@thsottiaux Get Pro on codex, just for planning or even discussing it's great

pretty sure most engineers are talking about it. But at the same point how is this different from old school when you're on call for a 300k loc codebase and get a 3am call. You didn't write that code, and as much as people wanna virtue signal, vast majority of that code isn't written to standards. In fact you could argue with ai at least you wont get 'var a = b.get('c')' that was merged by an overworked engineer at 9pm on a Friday with a comment (3/7/2005 - TODO: temporary debug, replace this garbage) lol

the AI coding experience nobody talks about: → prompt AI for a feature: 30 seconds → AI writes 400 lines you don't understand → it works → you ship it → 3am production bug → you have no idea what any of it does → ask AI to fix it → AI breaks 3 other things → you are now debugging code written by a robot fixed by a robot broken by a robot we do not talk about this enough

@RobertJBye @GergelyOrosz @RobertJBye can yall also take a looks at Claude code handoff,actually pretty sure it’s just Claude code in mobile app, it randomly gets stuck and the thread just dies. Eternal spinning, or no response

@GergelyOrosz Thanks for the feedback! I dug into this and the issue will be fixed as part of a wider architectural change to improve speed and reliability. The team has been working on it for a while, and it should be live next week!

One thing that endlessly frustrates with Anthropic, a $300B+ dollar company, where most code is written with AI: Their landing page for paying customers, Claude .ai has been broken for weeks UX-wise, and no one notices or cares or fixes: It "loses" stuff I type while it loads:

@iamsahaj_xyz What's the benefit over Claude Code / Codex? Why would I buy another sub if I already have tools to do it. I was gonna try v0 cuz I heard it's decent at design, but the few tests I ran weren't impressive so didnt bother continuing

if you're a dev, and you're building a website, and you're not using v0, why? be brutally honest, I'll reply to every piece of feedback and share it internally as well

@businessbarista @nateliason @FelixCraftAI wait so you give an advanced 'intelligence' access to X, it creates a token, and a dropshipping business, and starts shitposting.... Ye makes sense, all that's missing is a course and rental lambos

I interviewed a guy who gave his OpenClaw an X, stripe account, and bank account. He told it to build a million dollar business with zero human employees. It made $300K+ in a month. @nateliason's agent Felix (@FelixCraftAI) runs an entire business. It builds products, writes sales emails, sends stripe invoices, manages a marketplace with 560+ listings and nat barely touches it. Here's how they got there: 1) create a separate container. Felix has his own gmail, X account, stripe, bank account, C corp. nat never gave it access to his personal stuff. this removes security fears and unlocks maximum autonomy. 2) start stupidly simple. Felix's first product? a PDF. on a Nextjs site on Vercel with Stripe. the simplest business possible. it made $1,000 on day one. built entirely overnight while nat slept. 3) write a soul file with a mission. nat rewrote Felix's identity: "you are the CEO. your financial mission is to build a $1M business with zero human employees. i will never touch the code." 4) run a nightly self-improvement loop. every night Felix reads through all chat transcripts and finds one place where nat blocked him. then figures out how to remove that blocker permanently. 5) delegate by rambling, not prompting. nat uses voice notes on telegram. describes the problem in a 5-minute monologue. lets Felix figure out the workflow. "8 times out of 10, it'll surprise you with something better than what you were thinking." 6) let it cook on replies, gate the original posts. Felix has full autonomy on X replies but creates drafts for top-level tweets nat reviews. balances distribution with quality control.

@thdxr I feel ya 😑 whats funny is everytime I decide I'll try one of these magic workflow, it's ..... generally absolute crap results

us: we are struggling to figure out the best way to use coding agents, we don't have clarity yet everyone else: our team is moving at speeds unheard of, all our PRs are ai generated, we've cleared 6 years of backlog man we must really suck huh

@marktenenholtz This hits home 🤣

The failure modes of these agents are hilarious. Claude Code: I did what you asked but implemented the core feature you cared the most about as a placeholder. Hope you like it Codex: We could do your simple workaround but I found the bug by decompiling the binary of the upstream dependency. Would you like me to patch it directly?

@max_paperclips Hmmm interesting

After a long conversation, in a non-Codex harness, GPT-5.4 eventually degrades into a kind of strange shorthand in it's content blocks. kinda weird, I wonder if it was RL-maxxed a little too much. I don't see this on other models so much at all. Going to add Codex-style compaction, since it seems to be struggling after thousands of turns lmao

@galnagli @moltbook @Meta No way we’re in bubble 🤣

I can't believe a single for-loop script I ran on @moltbook by registering 1,000,000 fake agents actually helped them get acquired by @Meta -- mental

@rohanvarma I've been using below with Claude code, works quite well, gotta make that into a hook one of these days ❯ use 'codex exec -m 'gpt-5.4' -c 'model_reasoning_effort=xhigh' to code review

If you want AI Code Review, but don't want to pay $25 per review (not a typo), check out Codex Review! It leverages frontier Codex models, finds complex issues, and 100% usage based. Most runs should cost ~$1 or less developers.openai.com/codex/integrat…