It will have strong privacy and security options so you can trust it to handle all of your personal content knowing that no one else can access your information, similar to how encryption works on WhatsApp
List of interesting links I found today:
Read HN twice a day for the last decade. Here's my list of S-Tier HN links https://news.ycombinator.com/item?id=49183198
Why Estonians invite strangers into their back gardens each summer https://www.bbc.com/travel/article/20260731-why-estonians-invite-strangers-into-their-backyards-each-summer
My emergent thoughts on ads https://x.com/TnvMadhav/status/2085657823099433121?s=20
The Friendship That Made Google Huge https://archive.is/w4oWL -- 2026-08-07T09:34:43.157Z
- One super app for both chatGPT and Codex
- You can create automations from chat itself
https://x.com/OpenAI/status/2075274271845404744?s=20
It’s codex but it’s re-introduced for non-coders to start work from ChatGPT work desktop!
OpenAI chatGPT Twitter account also shared a look and feel, a dip toes in the water sort of dem video. https://x.com/ChatGPTapp/status/2075336584451539353?s=20
While my work is intact on the Codex app, the icon itself has changed a bit, as in the name has changed to chatGPT instead.
The icon on the status bar, however, is still the chatGPT icon and not the Codex icon. I'm not sure if this is a bug but when I enforced custom icons to be the same as Codex itself, it didn't change on the status bar. 
…and then I opened the app and noticed 3 new models available for use:
- GPT 5.6 Sol
- GPT 5.6 Terra
- GPT 5.6 Luna at this point I just don't know what they are but I want to find out quickly.
Most of the marketing content on my Twitter timeline doesn't talk about soul, luna, or terror. It just talks about GPT 5.6 so clearly you might have to do things and figure out by yourselves.
What if I asked ChatGPT itself?
ChatGPT also stated or cited the blog post from openAI.
We’re launching the GPT‑5.6 family of models for general availability following our limited preview: our new flagship, Sol, alongside Terra, a balanced model for everyday work, and Luna, our most cost-efficient model.
GPT‑5.6 Sol sets a new high of 53.6, eclipsing Claude Fable 5 (adaptive reasoning) by 13.1 points. Even at medium reasoning, it beats Fable 5 by 11.4 points at roughly one-quarter the estimated cost.
Results of a certain few benchmarks have been shared in this blog post. https://openai.com/index/gpt-5-6/
For my automations that are setup, I asked chatGPT to compare the prices of GPT 4o mini and GPT 5.6 Luna and it looks like luna is 6.7x more expensive than the former.
If you meant GPT‑4o mini, the gap is larger:
- GPT‑4o mini: $0.15 input / $0.60 output
- Luna: $1 input / $6 output
- Luna is approximately 6.7× more expensive on input and 10× on output. GPT‑4o mini pricing
ChatGPT iOS app has replaced codex tab with remote tab
The slider feature for reasoning levels for the GPT 5.6 Sol is sort of interesting. Is this feature here to stay?
It's also interesting that in the chatGPT app, without the remote connections, GPT 5.6 Terra and Luna aren't available and only Sol is.
Another update then I noticed is that there are more options for selected text on the on the ChatGPT Codex app on the desktop.
More details seems to be slightly similar to ‘Ask in Side Chat’ (I think the naming wasn’t changed to Side Task or whatever)?
I asked GPT 5.6 Sol about this and learned the following:
- The ask ‘for more details’ basically means explain this sentence or thing in more words without worrying about follow up questions (you still can if you want to)
- The ‘ask in side chat’ is a more generic one where you can take things (out of context?) and ask another agent to do something with it. So basically ask for more details is a more specific implementation of side-chat.
There should be an /action in a /sidechat that essentially tells codex to update findings in the main thread -- 🏞️ Context #1 🏞️ Context #2 🏞️ Context #3 🏞️ Context #4 🏞️ Context #5 🏞️ Context #6 🏞️ Context #7 🏞️ Context #8 🏞️ Context #9 -- 2026-07-10T10:36:13.114Z
The more I work with A.I. for engineering , the more I got back to fundamental principles of a good system.
Now is the best time to celebrate past efforts to get cleaner architecture.
LLMs are truly autocomplete engines.
Clean architecture yields clean architecture.
Clean architecture helps human in loop comprehend autocompleted changes easily.
Thus resulting in faster deployment cycles.
Clean architecture won’t matter if there are no humans in the loop.
But, if automated agents of all kinds are the workers of the future, clean architecture helps here as well.
It’s a win-win strategy.
Long term benefits outweigh the initial learning curves.
Things can easily go wrong at the start with all the psychosis at first.
Firm hand to get things put into place is a necessity. -- 2026-07-08T09:17:18.852Z
This is correct, I think a number of people on the tl didn’t read past the title and made inferences and comparisons that are just wrong and then use it as an opportunity to take cheap shots. This is not a “feature” like some crappy Slack bot and it’s certainly not a Claw, though it has aspects of it. It is an org-level harness. The difference will become clearer over time.
The idea of Org Level Harness is the key subject that caught my eye.
Claude Code was a butterfly effect.
i'm referencing a post that can be found here: https://x.com/karpathy/status/2069822834160124091?s=12 -- 2026-06-25T06:17:26.861Z
The best mobile software does not shrink a desktop interface. It identifies the decisions that matter when you are away from the desk and makes those decisions fast, legible, and safe.
Optionality of convenience makes your software workflows anti-fragile
i'm referencing a post that can be found here: https://developers.openai.com/blog/mastering-codex-remote-for-engineering -- 2026-06-23T18:08:52.417Z
The hardest part of any task is overcoming the activation energy. The rest is just riding the momentum.
Been thinking about this too https://x.com/tnvmadhav/status/1760205214555291714?s=46
i'm referencing a post that can be found here: https://x.com/fchollet/status/2068771319341367390?s=12 -- 2026-06-22T05:03:33.931Z
it’s June 2026. Stop hand editing your prompts. Hold down the dictation button and ramble for 10 minutes. Give the model every fragment, caveat, example, and vibe in your head. It is literally a large language model. If it’s superhuman at anything, it’s reconstructing latent intent from language.
I agree with this, i have been using codex’s dictation to paste my ramblings and it’s decent enough for small tasks that I give it.
i'm referencing a post that can be found here: https://x.com/guinnesschen/status/2068744472528314811?s=12 -- 2026-06-22T04:58:00.436Z
In the age of A.I. based lightning fast execution, being good at finding problems just by thinking about it is a sought out skill.
Idea guys with agency are surfing Shai-Hulud at this point. -- 2026-06-21T12:29:50.484Z
I completed Dune: Prophecy last night.
An important disclaimer is that I have been a naive fan of Dune and Dune Part 2 (coming in just from my appreciation of Dennis Villeneuve), which means I haven't read Frank Herbert's books.
I must say Dune:Prophecy hadn't succeeded in keeping me interested at first, however it's probably not a debate when I say the production quality is top notch.
Needless to say, I do feel it's a good one. I am captivated by the lore of this story (especially when it comes to Shai-Hulud).
Also, Thinking Machines?
That's the name of Mira Murati's A.I. company.
Not sure if that was the inspiration but it loved the acknowledgement.
It's strange that I didn't come across this nod/acknowledgement on my twitter timeline.
I liked the cameo(ish) from Mark Addy. I just love the way he speaks. -- 2026-06-21T12:06:03.177Z
NOTE: This feed is a sliding window. One can find a significant portion of a feed archive on my website.
All credits of this idea to Simon Willison.