Ray Amjad

@jxnlco @thsottiaux Sent you a DM and uploaded a neat Codex video too :)

@theramjad @thsottiaux Let’s reverse this dm me

What are we obviously not getting right with Codex?

@reach_vb Excited for how this changes when /goal is implemented on the Codex app

I love that Codex App now shows the Progress of your task in an easily parseable UI right in the chat! ✨

@jxnlco @thsottiaux Yo Jason. I can’t upgrade my account to 20x Max and OpenAI support was useless. I’m just trying to make videos about Codex. Help a bro out 🥲

@thsottiaux Text editor

I wish Codex had intended sub-subagents further when you set agents.max_depth = 2 Even though the bottom 3 are sub-subagents, the UI makes them appear as sub-agents

Thinking of switching over to Codex soon because the Claude app is now unusable Kills my memory a few minutes after opening. Every time

I wish Codex supported image inputs when it comes to generating images rather than text-only prompts. Kinda like the ChatGPT web

I can feel the vibe coding

@donk8r Claude says it's because of the system reminders. I think this is new. I don't remember the skill catalogue being injected after each turn.

@theramjad New Claude versions have different context window behavior. Check your session logs for what the model is auto-reading — it might be pulling in files you didn't explicitly request.

Is it just me, or are the tokens in Claude Code sessions rising really fast today? I went from 0 to 330,000 tokens in under 3 minutes during pretty standard work. This would've been like 70,000 tokens a few days ago.

CC @trq212 Feedback ID: 39e4c68d-f74e-4445-b832-de24aa2373d9

@patio11 I made this mistake once before 2 months ago. It took me about 20 seconds of brushing with shaving cream to realise

The UX of everyday things, hotel toiletries edition

@mmabas77 I’ve also noticed this! Things seem better after the recent OpenClaw ban they did

@theramjad have you ?

Have noticed that opus is doing much much better today, idk if it's a new update or what but it pays more attention to details and is better in following instructions in a very noticeable way.

@Plamen__Andonov @jescalan @OpenAI Glad that I’m not the only one who noticed this.

@jescalan That's nothing dude, check this out @OpenAI completely inventing new language in the middle of a cron report...

Transition to gpt in openclaw not going so great this far 😂

@bcherny This is like physicists getting emails from randos saying they discovered a new theory 😆

A weird part of working at Anthropic: getting a few of these each day

@levelsio @nikitabier @photomatt Wow, reading through the replies is actually bearable now

So @nikitabier implemented @photomatt's idea to stop AI bots from destroying the reply section on here You can set it to only allow people you follow and the people in turn they follow to reply, nobody else If on average ppl follow 500 people that means still 500*500=250,000 possible repliers But all the spammers are isolated out 👏

@noahzweben Can you make it so subagents can spawn other subagents like the max_depth feature of Codex’s subagents? I have a really interesting use case.

Zoom zoom

@bcherny Claude Code 3.0 ?? 👀

Today was a good day

@TahaTesser I integrated it into HyperWhisper.com and after some daily use, I find it to be worse than Parakeet. But the speed is great

Apple blew my mind with SpeechAnalyzer API. Highly underrated. developer.apple.com/documentation/…

@jessethanley @openclaw @Bento For many apps (like PostHog), I don't even open the dashboard anymore. I just have Claude Code to gather the data via an MCP server and send me a graph on Telegram each morning (alongside a bunch of other things for my morning report)

People are now sending @openclaw agents into @Bento to do work when the API cannot do the task and I really feel like I wish I spent more time early on accessibility. Like all the best practices apply v-well here. Need to spend more time and tokens on it.