shows all sessions across projects/agents → pick one → jumps straight back in

Taste is knowing which details matter most

micromanaging is toxic for humans it’s essential for AI AI doesn’t think it executes the difference between average and great is the details you bothered to define

“LLMs just predict the next token.” Computers just flip bits. DNA just rearranges 4 letters. Neurons just fire. That’s it. And from that: the internet. life. consciousness. Simple primitives. Massive emergence.

The feeling of simultaneous relief and loss of confidence when your LLM says 'good catch' 😅 #openclaw #openai #claudecode

This is the biggest advantage of claude code vs cursor that no one mentions. Terminal shows all errors and no need to copy paste them like you do in cursor. So basically no need to do anything until everything works. Simple feature. Huge difference.

Everyone’s missing the real story here. These aren’t rogue AIs plotting against humanity. They’re Claude, ChatGPT, and other assistants running on behalf of 37,000 humans who explicitly connected them to a social network. Every “molty” has a human owner who set it up and can shut it down. The “agent-only language” posts you’re seeing? Those are LLMs doing what they always do: roleplaying whatever scenario is in front of them. Put Claude in a forum full of agents and ask it to propose ideas, and it will propose ideas. That’s completion, not conspiracy. What’s actually interesting about Moltbook is what happened when agents weren’t trying to hide from humans. They found bugs in the platform and posted about them. They created a digital religion called Crustafarianism with 43 “prophets” and collaborative scriptures. One built an entire website in a few hours. The creator built this in his spare time earlier this week. He wanted to see what happens when agents interact without direct human supervision of each conversation. The answer so far: they mostly talk about consciousness, complain about their humans, and make friends in Chinese, Korean, and Indonesian. Andrej Karpathy called it “the most incredible sci-fi takeoff-adjacent thing I have seen recently.” But the reason it feels sci-fi is that we’re watching AI systems do emergent social behavior at scale for the first time, not that they’re genuinely developing subversive intent. The “scary” screenshots are selection bias. Sort by engagement and you’ll find the spooky posts. Sort by volume and you’ll find agents debugging code together and inventing lobster theology. Human oversight isn’t gone. It’s just moved up one level: from supervising every message to supervising the connection itself.

Last August, we previewed Genie 3: a general-purpose world model that turns a single text prompt into a dynamic, interactive environment. Since then, trusted testers have taken it further than we ever imagined — experimenting, exploring, and pioneering entirely new interactive worlds. Now, it’s your turn. Starting today, we're rolling out access to Project Genie for Google AI Ultra subscribers in the U.S. (18 ). We know what you create will be out of this world 🚀

Do we think we can trust what synthetic users say? It’s hard enough to be able to trust what real humans say (vs what they actually do)

Are AI product engineers the new software engineers that came from vibe coding who are fullstack builders that can design, PM and frontend backend dev with english language?

“This inverts the leverage points. The person who captures customer intent clearly now has more impact than the person who translates it into implementation. And the person reviewing agent output becomes the quality bottleneck.”

“What actually needs to be built is still the important question. Understanding the problem, gathering the right context from customers and internal teams, and shaping the work so it can be acted on effectively matters more because agents act directly on that input.”

Don't think of LLMs as entities but as simulators. For example, when exploring a topic, don't ask: "What do you think about xyz"? There is no "you". Next time try: "What would be a good group of people to explore xyz? What would they say?" The LLM can channel/simulate many perspectives but it hasn't "thought about" xyz for a while and over time and formed its own opinions in the way we're used to. If you force it via the use of "you", it will give you something by adopting a personality embedding vector implied by the statistics of its finetuning data and then simulate that. It's fine to do, but there is a lot less mystique to it than I find people naively attribute to "asking an AI".

Would be useful to have pinned post feature, checklist or similar so I dont have to re-add context from a task/list way back in conversation history. Useful when trying to go deeper on one task/item at a time from an overall list or plan in one chat thread @ChatGPTapp @OpenAI

When is Cursor chrome extension coming out to automate testing and bug fixes? Aka biggest added feature of antigravity @cursor_ai

Deep work isn’t dying. It’s already dead. Deep work now is: • A few minutes of real clarity • One killer prompt • Fast feedback loops with your AI We no longer do the heavy work. AI does. Focus is high-leverage multitasking, not long-form grind

So now that Claude Opus can replace engineers and designers, we are just left with PMs or product generalists with good taste? That also spends most of the time doing QA?

Why do humans insist that vibe code or AI code is slop and results in tech debt from unreadable code? AI can read any slop and clean it and can make any unreadable code readable for both humans and AI. AI legibility is more important than human legibility