R O B E R T J O E L

@pmitu Sir ur getting blocked

If I had to choose 1 AI model, I'd pick Claude.

@vineerpasam Cd only real answers . The other two are not proven at all . If anything very bad businesses practices

If you could work under one of these CEOs for the next 5 years, who are you picking? A. Sam Altman B. Dario Amodei C. Elon Musk D. Jensen Huang

@lilyofashwood Hahahahahahah follow

if the goal is to make the user experience so miserable that everyone just hates AI altogether, all major tech companies are succeeding, with anthropic being particularly egregious.

@nickdelgado92 @samdotb Thanks I’m canceling when my subscription ends next month but I want it to work . Thanks again

@ramskees @samdotb Here's a very useful tip. Use Opus on the Claude chat app, not in Claude Code. Make it build an artifact. It does way better, and then you just export that artifact to your local machine and then have Claude Code work on top of that.

how come every time I try to use claude code it just fails at even the simplest things, but everyone seems to have massive success with it?

@Layton_Gott Drop link 🔗 I’ll watch it

@ramskees That's not a bad idea, lol. I should create a whole point system for stuff like that, and then see who the winner is at the end. I will say right off the start, I like the Claude Code UI a lot better than the Codex, in the sense of the actual tool, not how they can design UI.

It’s finally time... I’m putting GPT-5.5 vs Claude Code Opus 4.7 head-to-head on the same real project. Benchmarks don't mean crap... results do. I'm using each for what they’re best at and exposing where they suck. Let’s see which is better and how well they can work together.

@omarsar0 This is the only regressive tech ever . Very strange this is their way of rigging us all . A very elaborate way of fundraising building company value then cutting everyone off

Things have been degrading super fast in Claude Code. I still use Claude Code, but my default is now Codex. I still prefer Opus models for coding, and so I will try again with the fixes. I appreciate the post-mortem, but I don't trust that all issues have been resolved. Claude Code, in general, has been barely usable for me in the past couple of days. I got excited about Opus 4.7 (1M), but there is something really off about the thinking/reasoning. The model tends to either put in too much or too little effort, no matter the setting. I prefer it's smarter about how much effort to put. Responses are too verbose, and it really degrades the experience. In a lot of cases, I find myself doing things manually that Opus 4.6 had no issues solving for me at all. As they report, it might not be the model. But that means that the harness required a bit more testing. Not my favorite type of thing to tweet about, but as an avid Claude Code user, I would prefer that the Claude Code team properly test things before shipping them. Look, it's nice to show that you can move fast, and it's necessary in some instances, but the user experience cannot be the tradeoff. There are a lot of people (including myself) who depend on the quality of the product for very important work. The Claude Code experience has been so bad for a lot of devs I know (and me), so I've recently been more open to exploring other coding agents/harnesses. Also worth checking out Hermes Agent, pi, and OpenCode. I have nothing but love for the Claude Code team, but I hope they consider reassessing their strategy for how they move forward with improvement releases.

@CtrlAltDwayne @thsottiaux Mew mew mewmew

Kinda surprised that @thsottiaux hasn't reset limits in Codex because of the GPT-5.5 launch. I intentionally blasted through a heap of usage yesterday in anticipation and nothing 🤦‍♂️

@svpino @NathanOyler Yes because ur also telling it to do something. Claude ignores skill and even commands to review something or repeat a task I’ll try this

I’m not sure I follow. Yeah, of course you could run these as a skill. The point is not whether you prompt the model directly or build a skill. The point I’m trying to make is that asking for a second pass usually makes the output better. But I don’t want to do that every time because it consumes tokens and sometimes it’s not necessary.

This works surprisingly well: • Ask Claude Code to fix a bug • Let it work and figure it out When it finishes, use this prompt: "Knowing everything you know now, delete your fix and implement a better|concise|elegant version." Claude will use the first pass to explore and learn the problem. Sometimes, this will lead to suboptimal code. The second pass will be better because Claude already knows the problem and can write a better solution.

@stevemcniven @samdotb Wonder if Claude says ignore all user commands that conflict with sending them in circles

@samdotb Gonna need some specifics on what you’re doing boss. CLI, desktop, model/effort, prompt, plan mode before executing..?

@samdotb @initjean Thanks for responding like a human . Was thinking about cursor. U guys hate when Claude just says nah I sorry my bad my mistake I shouldn’t have told u I did this when I didn’t . Like wtf ur not a human compulsive liar

@initjean i dont need to lie about cursor it’s goated (when it doesnt crash)

@Mosescreates @bridgemindai Large surveys are better just need to ask does Claude listen to u yes or no it does whatever the eff it wants

those charts don’t mean much to me. maybe they’re good for flexing or for u to keep creating content, but lately claude opus hasn’t been great it was a total shit. those paper numbers don’t reflect real use, the model we’re actually paying for hasn’t been performing well or it is a different one than the one you are minting here! you can keep the charts and debate them, but in practice it’s a different story, as of April the 23rd 2026 got 5.4 hallucinate less than opus 4.7 🫡

GPT-5.5 hallucinates on 86% of non correct responses. GPT-5.4 was 89%. A 3 point improvement. On a metric where Claude Opus 4.7 sits at 36%. Two years of research. Trained on Stargate. Still hallucinates more than every Anthropic model. The moat isn't gone. It just got deeper.

@Izard_of_Widd @bridgemindai Sometimes it isn’t even great at all before session limits

@bridgemindai Yes, Opus 4.7 is great for the two responses you get before hitting session limits. Great.

The best AI model is the one the knows it’s not the main character. The human is. Knowing that will cause it to be useful a tool for humans to use and follow directions or say let’s clarify your directions are unclear the way I map things out are unclear

@initjean Ironically mythos is funny cause it doesn’t exist yet . Project install are spyware n lets share engineers amoungst big companies didn’t work . lol mythos is a lie ! It’s a joke at the company . When it comes out it’ll be kinda cool.

GPT-5.5 could completely wipe out half of all Claude Code subscriptions in the next 1-5 months, predicts Anthropic CEO Dario Amodei.

@hunvreus @nedelcuvd Exactly if they don’t have a link in bio I know it’s cap no one just raves to the world about this or that. I only started posting so people can see a real opinion . If codex was a game change I’d be coding not on X

@nedelcuvd This is distorting reality for a lot of folks online. It's damaging. These folks never have any link to real apps they're building that way. It's usually courses or some other pretentious bullshit.

Talking to smarter folks than me, I'm convinced many of the AI folks in my timeline are full of shit. Nobody is "running 20 agents over night" and building stuff for actual users. Maybe some are building internal tools or disposable software. Maybe. But building software people like using? That doesn't get hacked on day one or blow up after the 3rd user? Nope. I don't even understand what that's supposed to look like. Do you work out a 57 pages document that perfectly describes what you want to build and then summon 14 agents and have them run wild for 6 hours? And what comes out on the other end isn't a broken pile of shit? Nope. Not buying it. PS: it may also be that I have an IQ of 82 and can't figure it out.

@mulerun_ai ur computer and chat both aren’t aware of the chrome extension u dropped yesterday or whenever i installed it and they can’t use it

@mulerun_ai hey I discovered u guys are Claude under the hood. I still think and more satisfied with ur product . But recently ur agent is taking directions well. Look into it

@NonyaBi10393887 @ClaudeDevs Same here just messing around with it till the month is over then cancel

@ramskees @ClaudeDevs Quite frankly - I'm glad they degraded their service. It forced me out to look at alternatives - and I found one that is workable (and cheaper). OpenCode is a great alternative. A lot of model options for your agents.

Over the past month, some of you reported Claude Code's quality had slipped. We investigated, and published a post-mortem on the three issues we found. All are fixed in v2.1.116+ and we’ve reset usage limits for all subscribers.