Rohme

Writer, self-proclaimed nerd, gin buff, subtly opinionated, serious case of wanderlust, and admirer of beauty and reason

[Skynet Insights @Skynet_insights]

Jamie Dimon runs $700 billion today as CEO of the biggest bank in America

Years before that, he warned his own bank it was heading straight for bankruptcy

They fired him for saying it out loud

Then the bank collapsed anyway, exactly like he called it

JPMorgan saw all of that and hired him

He just spent 35 minutes at Harvard being more honest than he has ever been in public

He breaks down how he actually thinks, how he sizes risk and why most banks keep making the same mistake

The framework he lays out is the exact one in my article

Find the independent signal nobody else is running

Size it correctly

And never let conviction override the model when the data turns against you

That is how a neural network quietly beats a room full of overconfident traders

Save the video, then read the article below

[NeilXbt @neil_xbt]

A dedicated market research analyst charges $80,000 a year to do what this Hermes and Claude system does overnight for the cost of 30 minutes of setup!

→ overnight research: loops across news, competitor activity, regulatory shifts, funding announcements, and customer sentiment
→ episodic memory: stores everything across weeks and months, connecting today's signal to last month's automatically
→ synthesis: Claude reads new information against everything in memory and produces implications, not headlines
→ delivery: Telegram briefing every morning before you sit down, five minutes to read

The memory architecture is what makes this different from every previous automated approach.

A standard AI assistant starts over every session with no knowledge of what it told you yesterday. Hermes builds episodic memory in as the core architecture, so the agent remembers everything it has ever researched and connects today's signal to last month's.

With just thirty minutes of setup, you have a system that runs while you sleep.

Bookmark so you do not lose it!

Follow @neil_xbt for more Hermes Agent and Claude builds.

[Dep @0xDepressionn]

Anthropic engineer, applied AI team:

"nobody babysits Claude anymore. the new skill is setting it loose and letting it run"

30 minutes from inside Anthropic on exactly how the Claude Code team does it: auto mode on, agent running on its own, review what it shipped.

worth more than any $500 agentic course.

watch it, then read the exact setup that runs Claude unattended below.

[darkzodchi @zodchiii]

https://t.co/vf5HUE5T5k

[Elon Musk @elonmusk]

🤣🤣

[Charlie Hills @charliejhills]

I compiled everything you need to master Claude Skills.

17 free Claude Skills resources (links below):

𝗢𝗙𝗙𝗜𝗖𝗜𝗔𝗟 𝗗𝗢𝗖𝗦

↳ The Unofficial Guide:

https://t.co/IgzBEz4jeU

↳ Best Practices Guide:

https://t.co/SX6Y3itSze

↳ Skills Documentation:

https://t.co/Pozs7Zw2ZE

↳ API Reference:

https://t.co/pNBTaV8Nm0

↳ MCP Documentation:

https://t.co/wbg0e68wBm

𝗕𝗟𝗢𝗚 𝗣𝗢𝗦𝗧𝗦

↳ Introducing Agent Skills:

https://t.co/TeBnaMVxni

↳ Engineering Blog:

https://t.co/dLBolEqfYk

↳ Skills Explained:

https://t.co/PksGsVBpyV

↳ How to Create Skills:

https://t.co/Qg1ppCkYwA

↳ Skills for Claude Code:

https://t.co/F664hheB0y

↳ Frontend Design Skills:

https://t.co/ydbeQt4VS0

𝗘𝗫𝗔𝗠𝗣𝗟𝗘 𝗦𝗞𝗜𝗟𝗟𝗦

↳ Anthropic's Official Library:

https://t.co/iR9fyxUFvL

↳ Partner Skills Directory:

https://t.co/kyp4MEMwXA

𝗖𝗢𝗠𝗠𝗨𝗡𝗜𝗧𝗬 𝗟𝗜𝗕𝗥𝗔𝗥𝗜𝗘𝗦
↳ https://t.co/Apk1gSY0Zp: https://t.co/d6KCvDRVP2
↳ SkillsMP: https://t.co/6SAHQrillL
↳ Smithery: https://t.co/zVeBczxFb0
↳ SkillHub: https://t.co/dbCbgQcWov

Repost ♻️ to help someone in your network.

P.S. Which Skill are you building first?

[Rahul @sairahul1]

Everyone is building agents.

Nobody agrees on what an agent is.

"It's an LLM with tools."
"It's an autonomous system."
"It's a multi-step pipeline."
"It's just a for loop with an API call."

They're all partially right.

An agent is actually simple: an LLM + a tool call + a loop that takes feedback from the environment until it arrives at a solution.

That's it.

No framework required.
No PhD required.
No 6-hour YouTube course required.

Once you understand these 30 core concepts behind that loop — you can build anything.

Here they are ↓

[Akshay 🚀 @akshay_pachaar]

the four pillars of loop engineering.

the loop itself is six lines, and nobody competes on it. every serious agent framework lands on the same tiny while-loop. model reads context, calls a tool, you feed the result back, repeat until it stops asking.

so if that part is solved, what is everyone actually engineering?

the answer is everything around the model. Boris Cherny, who built Claude Code, put it plainly. he doesn't prompt Claude anymore, he writes loops and lets them run.

that shift has a name now, and it rests on four pillars that are harder than the six lines make them look. these are the parts that actually break:

→ knowing when to stop. a terminal message ends the turn, not the task. an agent will write failing code, glance around, and declare victory. "done" has to mean the tests pass, not the agent feeling good about its work.

→ keeping the context clean. long loops rot from the inside as old outputs and dead ends pile up. a worse context produces a worse decision, which adds more noise, and the agent gets dumber the longer it runs. you fight it by treating context as a budget, not a bucket.

→ tools the agent can actually use. pile on a hundred tools and it loses track of which one to reach for. writes have to be safe to repeat, because loops retry, and a retried "create customer" call leaves you with duplicate records.

→ something that can say no. left alone, an agent agrees with itself. the fix is to separate the maker from the checker so the worker never grades its own homework.

put those four together and your job changes. you stop steering the agent move by move and start designing the system that steers it.

Karpathy runs research loops overnight that tweak a script, test it, keep what works, and throw away what doesn't, with himself nowhere in the loop. he arranges it once and hits go.

the model is becoming a commodity. the loop around it is where the real engineering lives now.

the best builders stopped asking what they should tell the agent to do. they started asking what system would do this without them.

I wrote the full breakdown. the article is quoted below.

stay tuned for more on this!

[Harry Tandy @HarryTandy]

Paul Graham, co-founder of Y Combinator:

"Do things that don’t scale"

That advice works for AI systems too

Before you automate your life with Claude, run the loop by hand

Here are 8 rules for turning one useful prompt into a daily machine:

1. Start manual
- open Claude
- paste one lecture, article, or note
- ask for the exact output you want

2. Define the output
- sections
- format
- file name
- minimum quality bar

3. Add success criteria
- has 5 key ideas
- has 3 quotes
- links to 2 existing notes
- has 5 review questions

4. Force a verify pass
- Claude checks the result
- scores each rule
- fixes the weakest section first

5. Run it 3 times
- one easy input
- one messy input
- one long input

6. Turn the prompt into a script
- Python reads the input folder
- Claude processes the text
- output lands in Obsidian

7. Wrap it in a loop
- trigger
- do
- verify
- retry
- stop

8. Schedule only after it works
- nightly article scan
- morning review brief
- lecture notes when a new URL appears

Most people jump to automation too early

Prove the loop by hand, harden it, then let the machine run it while you sleep

[Noisy @noisyb0y1]

My 16-year-old brother gave his parents back $80,000 using Claude Code

He’s not a programmer. A year ago he didn’t know what a terminal or AI was

Watched one 30-minute video by Andrew Ng on YouTube. And overnight automated a process that had been costing the family a week of manual work for years

The next morning his parents woke up and saw that the process that used to take them a week every month was now running on its own

McKinsey charges $2,000,000 to do the same thing for large companies
He did it for $20

The video he watched that night - below

[Nyx @Nyx_Tech_AI]

How To Use Claude To Start a Business📍
Follow me (@ShadinRaj231681)

[Nainsi Dwivedi @NainsiDwiv50980]

Sam Altman just made the boldest prediction about AI:

"We're going to see 10-person companies with billion-dollar valuations pretty soon."

Those companies won't be hiring 1,000 engineers. They'll be running on AI agents.

Here are 8 AI tools changing how software gets built:

[Khairallah AL-Awady @eng_khairallah1]

I genuinely don't understand why everyone isn't using this yet

Web scraping is dead.

PixelRAG is a retrieval system that skips HTML parsing completely.

100% open-source. Coes with a Claude Code plugin that gives Claude eyes.

PixelRAG skips HTML parsing completely. It screenshots the page and a vision model reads the answer straight off the pixels.

instead of converting a webpage to text, chunking it and embedding it, it screenshots the actual rendered page. a vision-language model reads the answer directly from the pixels.

this sounds like a gimmick until you see what HTML parsing actually costs you:

- a single HTML-to-text converter silently drops 40%+ of a page
- tables, charts, and visual layout get flattened into meaningless strings
- just swapping your parser can shift accuracy by ~10 points on the same documents

the parser was the problem the whole time.

the Berkeley team indexed all of Wikipedia as screenshots. 30M+ images. 8.28M articles. and it still outperforms the best text RAG baseline by 18.1% on pure text QA.

not visual QA. text QA. the format it wasn't even designed for.

it also ships a Claude Code plugin called pixelbrowse that gives Claude actual eyes.

instead of scraping the DOM, Claude screenshots any URL and reads the rendered page as a human would see it. hand it a live site, an arXiv paper, or your localhost and ask what's on the screen.

one pip install. no MCP server. no backend.

how the pipeline works:

- renders every document (web pages, PDFs, images) into image tiles
- embeds them with Qwen3-VL-Embedding, LoRA fine-tuned specifically on screenshot data
- builds a FAISS index and exposes a search API

the best part: upgrading your reader model improves accuracy instantly with zero re-indexing. because the index is just pixels, not text that needs to be re-chunked and re-embedded every time a better model drops.

a stronger VLM plugs in and everything gets better for free.

(Link in the comments)

[Anatoli Kopadze @AnatoliKopadze]

an OpenAI engineer just showed how he gets agents to do his whole job: code, debug and more, using loops

29 minutes from the engineer who coined "harness engineering"

he writes the rules, agents write the code, a reviewer agent loops until it's right

the winners won't have the smartest model, they'll have the best loop around it

watch it, then read the full guide on loops below

[Anatoli Kopadze @AnatoliKopadze]

Google Brain founder, Andrew Ng:

"100% of my tasks are done by ai agents, self-improving loops are next.

Give it 3-6 months and prompting is gone."

31 minutes of clear explanation on building self-improving agents from scratch.

Worth more than any $500 agentic course.

Watch it, then read the full guide on loops below.

[Nainsi Dwivedi @NainsiDwiv50980]

Anthropic just accidentally made every AI course on the internet worthless.

A free 24-minute video. No signup. No paywall.

Taught by the people who literally wrote the code Claude runs on.

I watched it twice.

The part at 8:12 alone is worth more than any $300 course I've bought.

Most people will scroll past this. The ones who don't will have an unfair advantage for the next 2 years.

Bookmark before it disappears 👇

[spect @spectnfa]

Co-Founder of a16z (the biggest VC fund in the world) Marc Andreessen:

"People who want to build their careers should be spending every spare hour talking to AI: alright, train me up"

He says this is the habit that separates who gets paid and who gets replaced

In 1 hour 44 minutes, he explains why the people winning right now are training themselves on AI every spare hour

Open the model Ask it to teach you Practice Ask again Repeat daily

That is the new career ladder

Bookmark and watch the interview

[MIKE @mikenevermiss]

THIS SINGLE CLAUDE.md FILE HIT #1 ON GITHUB TRENDING.

90k+ stars. 7.8k forks. zero dependencies.

save this before the feed buries it.

it’s not another prompt pack.
it’s a simple CLAUDE.md file built around 4 principles inspired by Karpathy’s coding habits:

/ think before coding
/ keep it simple
/ make surgical edits only
/ define the target before you start

drop it into your project today and Claude Code starts behaving very differently.

read it today. link below.

[NeilXbt @neil_xbt]

Spent too long thinking better prompts were the answer!

They're not. Better prompts are optimizing one component of a system while leaving the architecture broken.

The systems that actually run unattended for hours, fix their own mistakes, and only surface to a human when something genuinely needs judgment are not built on better prompts.

They are built on loops with four parts that most people never build completely.

Most failed automation attempts are missing one of those four entirely.

Karpathy noted that 90% of AI mistakes trace back to missing context, not model weakness. Loops fix this by accumulating and re-injecting context across every cycle instead of starting from zero each time.

The model is commoditizing fast. GLM 5.2 is within 1% of Claude Opus on the hardest benchmarks. What doesn't commoditize is the loop design around it.

That's the skill gap. It's quietly opening while everyone argues about benchmarks.

Bookmark so you do not lose it!

Follow @neil_xbt for more AI engineering intelligence that shows you where the real gap actually is in 2026.

[Movez @0xMovez]

Andrew Ng:

"100% of my tasks are now done by AI agents - hype has exceeded my expectations. Loops is next step.

in 3-6 months, everyone will be using self-improving loops. No more prompting."

In a 30-minute talk, Andrew Ng explains how to build self-improving agentic systems from scratch.

Worth more than a $500 agentic course.