Fedesco

@gdb Becoming incredibleR

codex app becoming incredible

@sama Been using it for non-coding work for weeks! Works impressively well for that

big upgrade for codex today! try it for non-coding computer work.

@Ian8ach And how I AI in general

codex has completely changed how i computer

@emollick I was just trying to explain this to the engineering team where I work last week. I thought of sending them your latest Substack. The same prompts are delivering far worse results through APIs than through Claude Code and Codex. It's got to be the s̶h̶o̶e̶s̶ harness!

Increasingly, I think, we will see a gap between what you can do with frontier model APIs & what you can do with the native apps from the frontier labs (Codex, Claude Code). Models developed and trained with their native harnesses in mind have more capabilities in their harnesses

@ajambrosino @SIGKITTEN @bughuntergeek Did a recent update fix it? I've tried it a few times, and Codex tells me the screenshots are there, but it has no access to them. There are some use cases I can think of where it could help quite a lot.

@SIGKITTEN @bughuntergeek ask codex

it's still experimental so we hide it a bit, but in the codex app, try: > what have i been doing very inefficiently on my computer (according to Chronicle). make some recommendations. be direct. tell me what i need to hear.

@deredleritt3r @inductionheads Great distinction! We may need a few more steps for LLMs to do tasks well enough to be trusted on their own. Point (2) catches lots of real-world messiness; people won't let go of their turf quickly. I see a few AI centaurs doing most of the work while colleagues talk in the hall

Right, makes sense. I think there are two separate questions: (1) what is the level of model intelligence needed to handle a particular task (with appropriate interfaces, scaffolding, etc.)? and (2) how does the rubber actually hit the road in having AI handle such tasks autonomously? I was really talking only about (1), but (2) is also critically important.

The elephant in the AI room is that the labs have basically cornered themselves into having an upside dependent on greater and greater superintelligence When the intelligence we have today, granted memory and harness improvements, is sufficient for 99% of tasks in today’s world

@deredleritt3r @inductionheads Where I live, a large chunk of jobs are political appointments to repay political favors. A friend, a vice-minister, had separate assistants to handle her email and her phone calls. In such contexts, jobs may be replaced in their tasks, but political pressure may keep jobs open.

@inductionheads That's right. Most jobs don't actually require all that much intellectual horsepower. This is likely a blind spot for the frontier labs BTW, since everyone employed there is a giga-genius working on frontier research all day.

@thsottiaux @luke_pighetti I've been using this Epson photo printer app for years. I didn't love it, but it worked. It's about to lose compatibility with MacOS. So I just asked Codex to create something better suited to my needs. Man, what I have now, an hour later, is so much better. A sartorial workspace

@luke_pighetti Things will get better once everyone starts to use Codex

am i crazy or are all these software platforms degrading in quality over the last six months? feels like things are getting jankier and jankier

@mreflow Yes! Multiple email accounts would be marvelous. I've been thinking of unplugging from one and connecting to another, for a season, to clean inboxes, say. But that's a clumsy solution.

I have to admit... Codex has really grown on me. I've pretty much got it open and running some sort of process throughout the day at all times now. Having it connected to my Gmail, my calendar, my Slack, my Granola, and being able to see my personal journals is a game-changer. It writes draft emails for me based on my calendar and what it knows about me from my journals and my meetings. And 9 out of 10 times, I can just click send on the drafts with no edits now. I'm also using it as the front-end of a personal wiki that I built. I save articles and YouTube video transcripts of content that I learned something from into the wiki. It creates .md files with front-matter for every piece of content I add, cross-links ideas, people, channels, etc. and then I can just chat directly inside Codex and it responds based on what's been saved to my wiki. I've basically got an email management platform, wiki, journaling app, CRM, meeting organizer, and everything in between all from within Codex. All my markdown files are also inside Obsidian so I have a second visibility layer for everything as well. But, for the most part, I almost never open Obsidian anymore. I can do everything I need from Codex. It's pretty sweet! I'll likely make a video about it soon but it's pretty "in the weeds" and too nerdy for most. (But I'll probably make it anyway) If there was just ONE thing that I wished Codex would allow me to do... I'd love to be able to connect multiple GMail accounts to it to help me manage my multiple inboxes. But I can seemingly only have one email connected at a time... Other than that, it's doing everything else I want it to.

@emollick Did you say it "can't" release its model? I thought it was a self-imposed choice.

Mythos seems to be a very capable model based on available information, but it is not a cybersecurity model - it is an advanced general purpose model that happens to be good at cyber because it is good at a bunch of things. Anthropic stated that they were worried about cybersecurity risk, and their efforts mean it is a restricted model with lots of government attention. OpenAI and Google will pass the same threshold soon (and may already have with unreleased models). and the question is whether they are as worried about cybersecurity risks, or whether they think their guardrails will hold. Currently, the degree to which models have cyberrisk is entirely self-reported and not regulated. That means that OpenAI and Google could release Mythos-class models if they want, by assessing the risk differently and making different decisions. Does that mean Anthropic is at a disadvantage because it can't release its equivalent model? Will OpenAI and Google also be somehow restricted from releasing their Mythos competitor. It all seems pretty unclear right now.

@emollick Something I've noticed when reading ChatGPT's thought traces is that it encourages itself *a lot*. It reminds itself it can do it, it can stay on track toward finding a solution. I wonder if Gemini needs a drop or two of that magic brew too.

I think the Gemini chatbot has all the pieces to be a useful tool, but struggles to put it all together. It still doesn't seem to know what files it can create or how its tools work together. It also seems to get "discouraged" a lot, giving up rather than finding new solutions.

@emollick After hearing the announcement, I decided to stand in the background for a few hours and wait for you to give it a try. No surprise here with regard to Gemini, then. Some of the AI tools they are building into their workspace app are far better than what comes from Gemini itself.

Gemini now can create documents, and it is a nice start, but not up to the frontier yet, as you can see from my "LBO of Hogwarts" test. PowerPoints are substantially worse than NotebookLM, spreadsheets are primitive, still no thinking trace, it doesn't think hard enough, either.

@felaardo @OpenAI Claude ran a *very* similar contest last year!—but it was restricted to US residents.

@OpenAI Anthropic doesn’t run contests to get devs to use Claude Code. They just ship and the devs come. The gap isn’t closing, it’s widening.

Want to secure an early ticket to OpenAI DevDay? Build something with GPT-5.5 and Image Gen. Each week, we’ll select 2–3 favorites to win free tickets to OpenAI DevDay 2026. Codex will help us find the best submissions and our team will select the winners. Reply with #OpenAIDevDay2026, a playable link, and a quick note on how you built it.

@OfficialLoganK @vaaselene I'd love to know where it excels for you when compared to other frontier models. For me, it's multimodality.

@vaaselene yes I use it, it is p good

is anyone using Gemini?

@shaunralston @testingcatalog If they did, lots of people's phones wouldn't stop vibrating over dinner

@testingcatalog will it push notify when you've exceeded rate limits?

Claude Code can now notify you via a push notification when it has finished the task. Claude push 👀

@gdb And I have been powering through chatbot interfaces since 2021—until Codex marked a revolution.

terminal has been my primary interface to my computer for almost two decades. now it’s the Codex app.

@DeryaTR_ This looks great! Will save this poster for later. What about ice cream (as in that book by longevity expert Ezekiel Emanuel)?

Asked GPT-5.5 Images-2 to create a poster for 10 things you can do to live the healthiest life possible. This would be good to hang on the wall as a daily reminder with a checklist. Most people know these things, but very few follow them all, myself included. 😊

@XFreeze Of course, these tricks, though interesting, are not reflective of real-world work tasks. That's where ChatGPT (via its family of apps and harnesses) and Anthropic (same) are nowhere to be matched by Grok. Let's hope Grok does get there. Having more choices in this space is great

The exact same question to Grok 4.3, GPT 5.5, and Claude Opus 4.7: “Count to 10 starting from 11” Grok 4.3 wins 🏆 Every single time It gave 11, 10 and explained why going backwards was the only logical move... The others started counting from 11 to 20 Grok’s logical reasoning is at a level most models still can’t even touch

@thsottiaux Created a whole workbook (using Image Gen) for a young learner following the same style and imprinting it all with a pedagogical sequence. Also, creating a detailed prompt pack and pipeline for educational materials. And creating an MD reader/editor attuned to my use cases.

With some small tweaks, Codex can work for days on hard tasks. We will release some changes to make this easier to use for everyone. What’s the hardest task you’ve seen GPT-5.5 succeed at?

@DeryaTR_ Agree. I think he's the best embodiment I've seen of the Codex vibe so many of us are enjoying and benefitting from. Every one of @thsottiaux's messages is a delight.

I declare Tibo my favorite person of the month! Codex has achieved escape velocity and we are going to seek out things we could only dream of!