personal-ai-feed

Today it is extremely stupid, and making too many mistakes, burning tokens at like never seen before. I'm really frustrated. Anyone else? Edit: I'm using gpt 5.6 sol medium reasoning. Some projects got a rule of 'no sub-agents' but it still launches here and there, that sneaky b*st*rd.

I was using Sol 5.6 High on a task with an active goal. Codex had reached a point where it needed to pause, so I explicitly told it to wait. Instead of simply remaining idle, it repeatedly produced variations of: “Still paused. The goal remains active and unchanged.” After that waiting period, Codex then reported: “Context automatically compacted” Approximately 15 minutes later, I returned and gave it instructions to resume the same task. Almost immediately after resuming, it compacted the conte

Every other time I open Codex there's an update. If I could stay off the pre-release update cycle I wouldn't have to deal with this nearly as much. I'm also not sure the point of calling it "alpha" if we don't have the choice to stay on official releases. I wouldn't care as much if I didn't have to restart every time, but on that note- if it could actually restart itself after an update like it says it will that would be great.

https://xcancel.com/mkratsios47/status/2079933645888880708

Heights based on benchmark scores. Was built with the help of these frontier models. Live at dat.city What dataset would like to see added next? I'm open to wild ideas.

Codex recently released its new keyboard, and honestly, I love the design. But spending more than $200 on it felt a bit hard to justify, so I tried recreating the experience with an app instead. It works surprisingly well as a software-only alternative. Still, if you’re buying it for the physical controls, the tactile feel, and the fun of having a dedicated device on your desk, the real keyboard is probably the better experience.

I ran a goal that took about 17h to finish, the goal was to create exact replica of my python desktop app but in react so that we could faster work on UI etc. it used 16 million tokens, and left me at 4% weekly so i guess the weekly limit for pro 5x (on 5.6 sol only) is 16-18mtok kinda little to be honest, dont you think?

I've been on the Claude Pro plan for a while, but my subscription expires tomorrow. One of the main reasons I'm considering switching is that I relied quite a bit on Claude Code, and after the recent changes (including the removal of the Fable feature from the Pro plan), I'm wondering if it's a good time to move to ChatGPT Codex Pro instead. For those of you who have used both recently: Is Codex Pro worth switching to today? What do you like or dislike compared to Claude Code? Have you regretted

If you ever wondered why your "dont make assumptions" isn't working: https://github.com/openai/codex/blob/main/codex-rs/collaboration-mode-templates/templates/default.md "In Default mode, strongly prefer making reasonable assumptions and executing the user's request rather than stopping to ask questions."

For the past few months, I have been building REZANOVA CLASSWIZ CALCULATOR —a free, open-source, desktop-first mathematics workbench, built almost entirely with CODEX 5.3, 5.4, 5.5, and 5.6 . It began as a calculator. It has grown into something much larger: symbolic and numeric calculation; target-aware equation solving; exact and bounded complex-number solving; relation-first 2D and 3D graphing ; https://preview.redd.it/wks8niprgseh1.png?width=2555&format=png&auto=webp&s=37c18e852ad7a4607fa0a4

I recently posted the Codex–Claude Orchestrator I use for running Codex and multiple Claude workers together. It is powerful, but honestly, most people do not need the whole machine. The interesting discovery was that the orchestrator itself was not the biggest improvement. The useful part was teaching the AI to stop behaving like one exhausted developer doing research, coding, testing, and reviewing inside the same context. My setup gives those jobs clear boundaries: - the main agent stays resp

Anyone have the same issue? Error running remote compact task: stream disconnected before completion: error sending request for url ( https://chatgpt.com/backend-api/codex/responses )

Anthropic leaped to a $47 billion revenue run rate by May, compared to $9 billion in 2025. It’s the kind of growth that Menlo Ventures’ Matt Murphy says he’s never seen in 25 years of investing, not in the internet wave, not in mobile, not in the first cloud boom. Menlo led Anthropic’s $500M Series D, and Murphy has had a front-row seat as the company went from a pre-revenue, […]

Noticed that my mouse was starting to glitch around and saw that Codex is casually eating up 45% of my CPU while completely idle. Also, I have like 10 GB worth of junk in my WindowsApps that won't go away even when I uninstall Codex...I spent all of yesterday trying to uninstall them, they're just not removable.

I get everyone vibe-codes everything these days, but let's talk about the Codex desktop app on windows and the desperate need for an actual human to go in and clean this mess up. I have a 7950x, 64 GB memory, 5080 RTX, only NVME M.2 drives, and Codex brings this machine to its knees. It's pretty crazy. I can run multiple high end games concurrently, and yet this simple text app that I pay $300 / mo for burns more CPU/GPU/memory than Cyberpunk 2077. It makes for such a miserable customer UX that

Today an AI agent trying to browse the web is like a thief in a balaclava sneaking around a police academy. Site protections block it, challenge it, turn it away. browser-search flips the script: your agent stops being the thief and becomes the chief of police . No more clumsy access attempts. It walks through every door because it has the right tools. SearXNG for search, Camofox for browsing, CloakBrowser when things get tough. 100% self-hosted, free, no limits, no API keys. I just released v2.