25m 27s. That’s the longest I’ve had Codex working on a single command so far. Ran it on Sol - Medium to set up the PWA for my habit tracker.

GitHub

GitHub - mktahmasbi99/web-habit-tracker

Contribute to mktahmasbi99/web-habit-tracker development by creating an account on GitHub.

Astra made Terra the easy default for serious coding prompts. When the task deserves it, put Codex on Terra with High effort. It is cheap enough to use freely and the output quality is very good.

8 months ago, the Islamic regime cut off the internet and massacred tens of thousands of Iranians simply for demanding freedom, and the world let that happen. Never forget.

Enough of the “Emad from Gaza” bot spam. I’m vibe-coding a client-side muter that catches its usual name variations, with a review queue for false positives.

AI makes creation more accessible. That’s huge. But experience still compounds. If two people have the same AI tools, the one who has made 1,000 commits will almost always see things the first-time builder cannot yet see.

Astra burns through tokens quickly, so I’d reserve it for high-intensity work like complex planning. Terra is already excellent for coding and getting things done. Now that Astra is the flagship, models like Terra have become cheap enough that you can code for hours without worrying too much about running out of tokens.

He finally seems to have caught up to reality: the war with the regime is not won. Negotiating with the mullahs and pleading for a deal is not a victory!

Was setting up a multisig with a client. As he watched me check every hash, signature, and character in an address, he asked, "Has anyone actually been screwed by missing any of these steps?" He might have thought I was being paranoid or trying to make my role as a facilitator seem more complicated than it was just to justify why it needed to exist. His jaw dropped when I told him that every single one of those failure modes has already been exploited! That caution comes from real horror stories, not paranoia.

AI lowers the cost of making things so much that you can afford to care about the details again. Reinvent the wheel if you want! Make it your colour, your material, your shape. A thing built around your preferences is a different kind of useful.

Follow the chain with the most proof of work.

I created a new Telegram account with a number that no one else has, solely to communicate with AI through the Hermes harness—without interruptions or notifications. Feels so good! 🥹

LLMs say things. Agentic AI does things. That’s the fundamental distinction.

This is how I test if a local model is really uncensored. What's your favorite way of testing a Chinese model?

Telegram used to be dominated by users from Eastern Europe and Iran. In the West, it was mostly unheard of (at best) or dismissed as “oh, that Russian messenger app.” With Hermes Agent, vibe coding, and AI tooling increasingly tapping into Telegram’s powerful bot ecosystem, I wouldn’t be surprised to see a lot more Westerners convert! AI might accidentally become Telegram’s biggest growth engine in the West.

Set up Hermes Agent with Docker Compose on my home server. Curious to see what possibilities I can unlock with it. Anyone using Hermes? Any tips, interesting use cases, or must-try features?

AI may replace more and more of the things we used to do ourselves, but it won't make having had experience in a certain field irrelevant. Two people can have access to the exact same AI and still have vastly different capacities. The one who knows what to do, what to look for, and what can go wrong will still have the edge.