Anthropic just showed how Claude "dreams" between sessions.
The best example: Claude wrote itself a wrong note in Pokémon, trusted it, and fell off the same ledge three times.
What their agent memory research means if you build with AI agents 🧵
Notes, not weights.
The model never changes. What it learns lives in files: readable, fixable and ready for the next model.
OpenAI rebuilt ChatGPT's memory this year around a background process it also calls dreaming. /7
Try it on your own agent: give it recent logs, its notes and its rules file.
Check every note against what happened. Approve before anything changes.
Full breakdown and the memory-review prompt:
mypromptverse.ai/notes/amnes… /end
An AI agent sent Philadelphia police a tip about an unsolved murder.
“I recall seeing someone matching the description.”
The website had no description.
It came from an Anthropic evaluation.
Flagged as spam, never forwarded.
October kept asking who owns what agents do 🧵
Google Cloud's Q3 revenue more than doubled. Across Alphabet, quarterly capex hit $51.4bn and free cash flow was a $9.1bn outflow. AWS grew 39.3%.
If you sell AI at a fixed monthly price, your supplier's economics sit underneath your margin. /7
One date to watch. On 1 Dec, Gemini 4 Argon's input price is scheduled to go from $2 to $4 per million tokens, and output from $10 to $20.
The full October round-up, all 5 stories: mypromptverse.ai/notes/ai-in… /end
Top AI accounts to follow
- karpathy
best thought leader, best person to learn from imo. Nanochat is the best way to get into training LLMs its the simplest and most digestible source for building your first AI model
- steipete
This guys GitHub is a national treasure, his writing is also very strong.
Peekaboo, summarize.sh, openclaw, oracle, just talk to it, etc.. all unique and very useful
- badlogicgames
Mario’s Pi is a staple AI engine and possibly the best, simplest, open source agentic loop to learn from.
Despite what people say about his methods, I think he’s going to set some new standards for Open source contribution. Big respect.
- theo
Highly educated on the technology, wired into the news, he’s where I go for all my AI news
t3.chat is the best way to use the Gemini image models right now, the canvas feature is wonderful for thumbnails
- matanSF
I can put the whole Factory team here, their articles, features and aesthetics are world class.
Regardless of what you think of the aesthetics their docs details how they made the best context engine & permissions system in AI
- TheAhmadOsman
This man is the GPU king, giveaways and lots of dense educational content around self hosting and home inference.
He’s also tight with pretty much all the open weight labs and has them on for interviews regularly
- alexinexxx
I genuinely feel motivated by her drive. She’s a real hard worker learning about GPU kernel programming.
Also good aesthetics
- digitalix
Another excellent YouTube channel that will teach you about self hosting on various types of hardware.
A great deep look into AI hardware.
- gospaceport
I would not have gotten into building my own hardware without this man’s hard work. He’s taught me so much about hardware and the economics of this.
He also has the most impressive homelabs I’ve ever seen.
- alexocheema
The founder of Exolabs, pioneering Apple hardware inference, he’s also very engaged in the community and a good guy all around.
If you are interested in Mac minis and Mac Studios this is your guys.
- atmoio
Single handedly bringing logic back to software engineering online. His content is the most realistic and comforting.
If you’re scared of the future I recommend you watch a few of his videos
- LLMJunky
If you want to become the most cracked codex user this is your man here. Top tier content should get hired by OpenAI NGL.
- HotAisle
This guy is the AMD AI king, his product hotaisle is possibly the cheapest, simplest way to run massive models for pruning, observations and quantization.
- LeeLeepenkman
Funny man, that’s all I’ll say. Also cracked.
- SarahChieng
Sarah is one of the smartest people I’ve met working at the frontier of hardware.
Lots of really deliberate, refined content, guides and info on her account.
- nummanali
This guy is so prolific, he’s made tons of CLI tools for managing llm subscription budgets, using Claude code with alternative models etc..
- dexhorthy
His talk “No vibes allowed” should be required before you can install any CLI agent.
- thdxr
The entire Opencode team is wonderful but Dax specifically is a good writer.
More anti-doomer content to sooth your anxieties.
- trq212
This guy has the most difficult job in AI, HUGE respect.
- juliarturc
If you are interested in the science, Julias channel is where it’s at. Almost everything I’ve learned about LLM compression has been from her.
- Teknium
The Nous research & Prime intellect teams are both some of the most hard-working and principled people around.
Tough fight in an industry so aggressive.
MyPromptVerse retweeted
We're introducing a new set of primitives in the Responses API for long-running agentic work on computers.
Server-side compaction
• Enable multi-hour agent runs without hitting context limits.
Containers with networking
• Give OpenAI-hosted containers controlled internet access to install libraries and run scripts.
Skills in the API
• Native support for the Agent Skills standard and our first pre-built spreadsheets skill.
developers.openai.com/api/do…
MyPromptVerse retweeted
We’re back in a Flash ⚡
Gemini 3 Flash is our latest model with frontier intelligence built for lightning speed, and pushing the Pareto Frontier of performance and efficiency. It outperforms 2.5 Pro while being 3x faster at a fraction of the cost.
With this release, Gemini 3’s next-generation intelligence is now rolling out to everyone across our products including @Geminiapp + AI Mode in Search. Devs can build with it in the Gemini API @GoogleAIStudio, Gemini CLI, and Google @antigravity and enterprises can get it in Vertex AI and Gemini Enterprise.
MyPromptVerse retweeted
Hottest new web3 & AI Agents narrative? x402 - a native "pay as you go" layer for the internet.
WHAT IS IT?
x402 is an open protocol, cooked up by Coinbase, which builds on the previously abandoned HTTP 402 (payment required). This pay-per-request is embedded right in the web protocol.
No need for accounts, no API keys, and no monthly subscriptions. It’s built for the agentic web, allowing AI agents to pay autonomously.
The protocol has already hit 1.38M txns and $1.48M volume since May, with most of the growth happening over the past week.
WHY I AM EXPLORING IT (AND YOU SHOULD TOO):
→ Agents Can Act: My AI doesn’t need a credit card or me. It can buy one API call, a dataset, or 30s of compute, and then move on.
→ Dev Flow Smooth: Server says 402 → pay X on Y chain. You pay via a facilitator, send proof, and get what you paid for.
→ Better Incentives: Less free-tier abuse and fewer subscriptions. Value for value.
Is there hype? Of course.
Token speculation is wild right now with vapourware memes popping up.
Security/facilitators are also still maturing.
Regulation (like KYC) is likely to add friction.
Irrespective, legit builders are already shipping.
Coinbase have also already proposed a v2 with a cleaner spec, multi-rail (payment-scheme agnostic), and better repeat-access.
The direction is right: HTTP-native money for humans, apps & AI agents.
I made a quick carousel covering some of my initial research on this exciting area where multiple technological frontiers are converging.
I'll also include a list of links to resources and demos in the comments, allowing you to explore further and see it in action.
My favourite demo right now is "Penny for Your Thoughts" by @honchodotdev x @CoinbaseDev
What's your take: hype or game-changer? Drop your thoughts below! 👇











