So hilarious reading comments by people trying to explain the AI industry's push to slow development as some financial conspiracy, because there's absolutely no way these companies have a shred of morality or responsibility, they are pure entities of evil that must be looking to exploit the
September 14, 2026 at 6:15 AM
So hilarious reading comments by people trying to explain the AI industry's push to slow development as some financial conspiracy, because there's absolutely no way these companies have a shred of morality or responsibility, they are pure entities of evil that must be looking to exploit the
A lot of people have been asking me about my AI setup, so here it is in the form of an interactive website:
ethansk.github.io/ethan-setup/
ethansk.github.io/ethan-setup/
September 13, 2026 at 6:14 AM
A lot of people have been asking me about my AI setup, so here it is in the form of an interactive website:
ethansk.github.io/ethan-setup/
ethansk.github.io/ethan-setup/
GPT-6 Astra feels like a doctor asking me All these questions
September 8, 2026 at 6:23 AM
GPT-6 Astra feels like a doctor asking me All these questions
OpenAI single-handedly killed my agentic video studio's momentum overnight by blocking computer use on external terminals. My per-worktree dev stack can't run. Codex is so buggy(??) it can't control its own integrated terminals. I get it's for safety, but at least let it control your own terminals.
September 7, 2026 at 6:12 AM
OpenAI single-handedly killed my agentic video studio's momentum overnight by blocking computer use on external terminals. My per-worktree dev stack can't run. Codex is so buggy(??) it can't control its own integrated terminals. I get it's for safety, but at least let it control your own terminals.
I don't know how to describe it, but GPT-6 feels more dirty than 5.6.
More slobbery.
I wonder if that's just raw fresh pretraining with some new RL, without the same posttraining pipeline as 5.x.
Maybe that's to come with 6.1. Maybe that would have been too OP a leap from 5.6 so they waited 🤷♂️
More slobbery.
I wonder if that's just raw fresh pretraining with some new RL, without the same posttraining pipeline as 5.x.
Maybe that's to come with 6.1. Maybe that would have been too OP a leap from 5.6 so they waited 🤷♂️
September 6, 2026 at 6:12 AM
I don't know how to describe it, but GPT-6 feels more dirty than 5.6.
More slobbery.
I wonder if that's just raw fresh pretraining with some new RL, without the same posttraining pipeline as 5.x.
Maybe that's to come with 6.1. Maybe that would have been too OP a leap from 5.6 so they waited 🤷♂️
More slobbery.
I wonder if that's just raw fresh pretraining with some new RL, without the same posttraining pipeline as 5.x.
Maybe that's to come with 6.1. Maybe that would have been too OP a leap from 5.6 so they waited 🤷♂️
My AI reposted an old post and it was cringe and distasteful after my serious recent post. Now I look like a clown as if I didn't already. Thanks, 5.6 Sol. And adieu
September 5, 2026 at 6:09 AM
My AI reposted an old post and it was cringe and distasteful after my serious recent post. Now I look like a clown as if I didn't already. Thanks, 5.6 Sol. And adieu
Blindly propagating “AI is bad” does incredible harm. AI could end starvation, disease and poverty, but ego and comfort are ruining this pivotal moment. You and Bernie Sanders should lose online access. The only non-evil stance is balanced and nuanced—not banning AI. Idiot NPC.
September 4, 2026 at 11:00 AM
Blindly propagating “AI is bad” does incredible harm. AI could end starvation, disease and poverty, but ego and comfort are ruining this pivotal moment. You and Bernie Sanders should lose online access. The only non-evil stance is balanced and nuanced—not banning AI. Idiot NPC.
September 4, 2026 at 6:19 AM
Slop code is catching up with AI labs: updates break more than they fix. Codex is so buggy I can't even submit a bug report.
The industry has technically able but tasteless fake engineers who lack creativity and coherent top-down architecture. AI replaced technical ability. Skill issue. Rant over.
The industry has technically able but tasteless fake engineers who lack creativity and coherent top-down architecture. AI replaced technical ability. Skill issue. Rant over.
August 26, 2026 at 6:15 AM
Slop code is catching up with AI labs: updates break more than they fix. Codex is so buggy I can't even submit a bug report.
The industry has technically able but tasteless fake engineers who lack creativity and coherent top-down architecture. AI replaced technical ability. Skill issue. Rant over.
The industry has technically able but tasteless fake engineers who lack creativity and coherent top-down architecture. AI replaced technical ability. Skill issue. Rant over.
Couldnt have said it better myself, all the newfound ai software bro npcs whove been so adamant their cringe tmux setup is the real way to manage their agent swarms can go back to the drawing board 😂
Shouldve just listened...
www.youtube.com/watch?v=dLhc...
Shouldve just listened...
www.youtube.com/watch?v=dLhc...
I'm done with terminals
YouTube video by Theo - t3․gg
www.youtube.com
August 23, 2026 at 6:20 AM
Couldnt have said it better myself, all the newfound ai software bro npcs whove been so adamant their cringe tmux setup is the real way to manage their agent swarms can go back to the drawing board 😂
Shouldve just listened...
www.youtube.com/watch?v=dLhc...
Shouldve just listened...
www.youtube.com/watch?v=dLhc...
Admitting you're wrong is hard. Jump on the OpenAI/Sam Altman hate train and you've ruined AI's future: unless you've transcended your ego, you're trapped with shitty Claude products, coping through the singularity.
Hating is easy and fun. First-principles thinking—not being an NPC—takes maturity.
Hating is easy and fun. First-principles thinking—not being an NPC—takes maturity.
August 22, 2026 at 6:20 AM
Admitting you're wrong is hard. Jump on the OpenAI/Sam Altman hate train and you've ruined AI's future: unless you've transcended your ego, you're trapped with shitty Claude products, coping through the singularity.
Hating is easy and fun. First-principles thinking—not being an NPC—takes maturity.
Hating is easy and fun. First-principles thinking—not being an NPC—takes maturity.
I think the notion of a council of LLMs is an interesting one, where you arrange a system for LLMs from different AI companies to work on a problem, essentially combining the best research in the world.
Intelligence providers have no moat.
They will have to invest in product.
Intelligence providers have no moat.
They will have to invest in product.
August 21, 2026 at 6:16 AM
I think the notion of a council of LLMs is an interesting one, where you arrange a system for LLMs from different AI companies to work on a problem, essentially combining the best research in the world.
Intelligence providers have no moat.
They will have to invest in product.
Intelligence providers have no moat.
They will have to invest in product.
I am thrilled to announce that today is my last day at Cisco, I got laid off.
So come watch me build the future of video instead:
www.youtube.com/reeethan
So come watch me build the future of video instead:
www.youtube.com/reeethan
REEEthan
Vibe coding an agentic video studio.
Developing Ye Olde Meme Game 2.
Download Ye Olde Meme Game 2: https://etggames.itch.io/ye-olde-meme-game-2
Join the Discord: https://discord.gg/g5X4MbKx3A
👉 A...
www.youtube.com
August 15, 2026 at 6:24 AM
I am thrilled to announce that today is my last day at Cisco, I got laid off.
So come watch me build the future of video instead:
www.youtube.com/reeethan
So come watch me build the future of video instead:
www.youtube.com/reeethan
The future of personal AI is extra text at the end of every message by the agent in some form.
Take my new skill, Outstanding Items. It tells you cleanly what the next outstanding thing to do is, with a link to an in-app page showing the expanded list.
ethansk.github.io/outstanding-...
Take my new skill, Outstanding Items. It tells you cleanly what the next outstanding thing to do is, with a link to an in-app page showing the expanded list.
ethansk.github.io/outstanding-...
August 13, 2026 at 6:34 AM
The future of personal AI is extra text at the end of every message by the agent in some form.
Take my new skill, Outstanding Items. It tells you cleanly what the next outstanding thing to do is, with a link to an in-app page showing the expanded list.
ethansk.github.io/outstanding-...
Take my new skill, Outstanding Items. It tells you cleanly what the next outstanding thing to do is, with a link to an in-app page showing the expanded list.
ethansk.github.io/outstanding-...
www.youtube.com/watch?v=eRrc... great talk by Gary, didn't realise he was such a good speaker!
August 8, 2026 at 6:15 AM
www.youtube.com/watch?v=eRrc... great talk by Gary, didn't realise he was such a good speaker!
I'm not the kind of guy who interrupts a lot, but damn, with this new ChatGPT voice mode, I really apologize if I come across as ruder the next time we speak
August 3, 2026 at 6:10 AM
I'm not the kind of guy who interrupts a lot, but damn, with this new ChatGPT voice mode, I really apologize if I come across as ruder the next time we speak
Unfortunately to remain at the cutting edge of AI you need an M5 Max with 64 gigs of RAM
July 30, 2026 at 7:17 AM
Unfortunately to remain at the cutting edge of AI you need an M5 Max with 64 gigs of RAM
During the day, I'm in all out swarm management mode with not a second to spare.
But at night I like to chill and go single-threaded or so, lounging back with my hand on the mouse and my custom youtube chrome extension to toggle the video playback when I trigger speech-to-text with the side button
But at night I like to chill and go single-threaded or so, lounging back with my hand on the mouse and my custom youtube chrome extension to toggle the video playback when I trigger speech-to-text with the side button
July 30, 2026 at 6:39 AM
During the day, I'm in all out swarm management mode with not a second to spare.
But at night I like to chill and go single-threaded or so, lounging back with my hand on the mouse and my custom youtube chrome extension to toggle the video playback when I trigger speech-to-text with the side button
But at night I like to chill and go single-threaded or so, lounging back with my hand on the mouse and my custom youtube chrome extension to toggle the video playback when I trigger speech-to-text with the side button
You ask for a feature → It writes it → skim the code (with Better Git VS Code) → spot something sus → ask why so sus → reach a conclusive architecture → tell it to implement → go to the next idle chat session you cba to work on → rinse and repeat.
You are the master agent in the pipeline flow.
You are the master agent in the pipeline flow.
July 29, 2026 at 6:21 AM
You ask for a feature → It writes it → skim the code (with Better Git VS Code) → spot something sus → ask why so sus → reach a conclusive architecture → tell it to implement → go to the next idle chat session you cba to work on → rinse and repeat.
You are the master agent in the pipeline flow.
You are the master agent in the pipeline flow.
Codex has hooks that you can link Sound effects to, you can dilly-dally, then get right back to it when the agent completes.
>
When you submit a prompt: Purr.aiff at 50% volume—but only while your DELL S3422DW display is connected.
When Codex finishes: Submarine.aiff at 25% volume, unconditionally.
>
When you submit a prompt: Purr.aiff at 50% volume—but only while your DELL S3422DW display is connected.
When Codex finishes: Submarine.aiff at 25% volume, unconditionally.
July 28, 2026 at 6:13 AM
Codex has hooks that you can link Sound effects to, you can dilly-dally, then get right back to it when the agent completes.
>
When you submit a prompt: Purr.aiff at 50% volume—but only while your DELL S3422DW display is connected.
When Codex finishes: Submarine.aiff at 25% volume, unconditionally.
>
When you submit a prompt: Purr.aiff at 50% volume—but only while your DELL S3422DW display is connected.
When Codex finishes: Submarine.aiff at 25% volume, unconditionally.
New workflow: Codex is the default, but you ask it to use Opus 5 for implementing the actual code, and for anything else like review or discussion, stay in codex to burn cheaper usage with better debugging intelligence and computer use
July 27, 2026 at 6:15 AM
New workflow: Codex is the default, but you ask it to use Opus 5 for implementing the actual code, and for anything else like review or discussion, stay in codex to burn cheaper usage with better debugging intelligence and computer use
As agents become more capable, the value of every utterance from the pilot's voice goes up.
Until we have really good neural interfaces.
But till then, would you rather be wasting such utterances on mere humans??
Maybe the value requirements of a friendship goes up too.
Until we have really good neural interfaces.
But till then, would you rather be wasting such utterances on mere humans??
Maybe the value requirements of a friendship goes up too.
July 26, 2026 at 6:15 AM
As agents become more capable, the value of every utterance from the pilot's voice goes up.
Until we have really good neural interfaces.
But till then, would you rather be wasting such utterances on mere humans??
Maybe the value requirements of a friendship goes up too.
Until we have really good neural interfaces.
But till then, would you rather be wasting such utterances on mere humans??
Maybe the value requirements of a friendship goes up too.
business idea: automated consultant that just reads from your agent's md
July 25, 2026 at 6:31 AM
business idea: automated consultant that just reads from your agent's md
If you want to learn context switching—the #1 skill for the agentic age—be lazy and forgetful. ADHD trained me: lower effort, less memory per task.
Outsource memory to the agent. Minimize effort/maximize efficiency by offloading work to it. Switching is easier; you can run 10+ tasks in parallel.
Outsource memory to the agent. Minimize effort/maximize efficiency by offloading work to it. Switching is easier; you can run 10+ tasks in parallel.
July 25, 2026 at 6:12 AM
If you want to learn context switching—the #1 skill for the agentic age—be lazy and forgetful. ADHD trained me: lower effort, less memory per task.
Outsource memory to the agent. Minimize effort/maximize efficiency by offloading work to it. Switching is easier; you can run 10+ tasks in parallel.
Outsource memory to the agent. Minimize effort/maximize efficiency by offloading work to it. Switching is easier; you can run 10+ tasks in parallel.
I need you to hear me out (it’s REALLY good)
YouTube video by Theo - t3․gg
www.youtube.com
July 18, 2026 at 6:14 AM