killaz

learning, researching & building in ai

dms open

[killaz @0xkillaz]

This is f*cking insane. Andrej Karpathy (Anthropic, ex-OpenAI):

"The biggest mistake people make with AI right now is making agents do the work before they've mastered the harness around the model

We made exactly that mistake at OpenAI in 2016. It cost us five whole years"

Opus 5.5 → Agents → Loops → Graphs → Harness

What Karpathy really means:

step 1 → quit forcing one agent to do everything, learn the model under it first

step 2 → a demo takes a weekend, a real product takes ten years. Self-driving cars showed it: skip the foundation and nothing holds

step 3 → the agent isn't what you sell, the foundation is. Get that right and the agents show up on their own

"If you're building agents today, you're at the frontier. Not OpenAI. Not DeepMind. You"

Bookmark it and watch it today, before your feed buries it

Then read the harness engineering guide that fixes this exact mistake, in the article below ↓

[Arcos @0xarcos]

Companies pay $300K+ a year for engineers who can run AI agents without babysitting them

Last week 20 Claude Code agents shipped 143 changes while I slept. I approved 6 merges

Here is graph engineering in Claude Code, file by file. Free ↓ https://t.co/RF9Pw3N4Yc

[killaz @0xkillaz]

This is f*cking insane. Demis Hassabis built DeepMind from zero, sold it to Google and took home a Nobel Prize

Now he went back to Cambridge and gave one free 1-hour lecture on what the next 10 years look like

Code → Natural Language → 10x Engineers → AGI

His advice: get good at the tools now, engineers are about to do 10x what they do today

Most $5,000 AI bootcamps teach less than this

You probably don't have an hour right now

Bookmark it before it disappears from your feed

Then read the harness, loop and graph breakdown below ↓

[rari🈀 @0xwhrrari]

Your agent passed every test

The issue is still open

A better prompt won't fix it, because the model isn't the broken part

I broke down the harness, loop and graph around it: what each one owns, how they nest, and which one to fix first ↓ https://t.co/rcmwM1jrfB

[killaz @0xkillaz]

This is f*cking insane. Anthropic CEO Dario Amodei just said something most people aren't ready to hear:

"Most of them are unaware that this is about to happen. It sounds crazy, and people just don't believe it"

He's talking about 50% of all entry-level white-collar jobs. Disappearing within the next 1 to 5 years

4 minutes on live TV. Watch this before making your next career decision

Lawyers → Consultants → Finance → Tech → Anyone Still Ignoring AI Bots

Surviving the AI shift isn't about being the smartest person in the room

It's about becoming the person who knows how to put bots to work

Most people will sit back and watch. The ones deploying bots today could be the ones doing the hiring tomorrow

More valuable than any $3,500 AI career course

Bookmark this and watch it today

Then explore the 25 bot setups the people building Grok Bot actually use in the article below

[rari🈀 @0xwhrrari]

Your agent passed every test

The issue is still open

A better prompt won't fix it, because the model isn't the broken part

I broke down the harness, loop and graph around it: what each one owns, how they nest, and which one to fix first ↓ https://t.co/rcmwM1jrfB

[killaz @0xkillaz]

This is f*cking insane. Ex-Google CEO Eric Schmidt:

"The programmers will go away. Mathematicians will go away.

The future is hyper-intelligence"

"In the next 5-10 years we'll see superintelor this"

In 54 minutes, the man who ran Google for 10s coming

AI programmers → AI mathematicians → AI thatigence

Once AI can replace programmers, it can impr everything speeds up

Most people are still arguing about chatbotsing

This podcast is worth more than most $1,000

Watch it today

Bookmark this. Future you will thank you ↓

[Arcos @0xarcos]

https://t.co/0Cs6YR6cf5

[killaz @0xkillaz]

Anthropic engineer:

“At Anthropic, we have moved beyond writing prompts.

Now we engineer harnesses.”

In this 42-minute session, she explains how the Claude team creates harnesses that generate and improve their own prompts.

Agents → Harness → Loops → Graphs → Self-Improving Systems

If someone sold this as a $500 course, people would call it one of the strongest agent engineering courses of the year.

Instead, it is completely free.

And more useful than most paid programs.

Watch it today.

Then follow the step-by-step harness engineering guide below

[Arcos @0xarcos]

https://t.co/0Cs6YR6cf5

[killaz @0xkillaz]

SpaceXAI engineer Lauren Tan:

“250 applications. Two years. Not a single offer.

Then he stopped presenting himself as one engineer and showed up as one engineer operating 30 agents.

We offered him $850,000.

I’m currently running 30+ agents through a loop with /pstack. Last month I shipped 1,000 PRs, and this month I’m on track to double that.

I woke up this morning and 20 PRs had already landed.

My job is no longer writing the code. My job is building the team that writes it.

99% of engineers still work from a single chat window.

That isn’t a skill gap. It’s a stack gap, and you can close it in one evening.”

/pstack → Loop → 30+ Agents → Overnight PRs → Autonomous Workflows

In this one-hour session, Lauren shows how to build an effective team of AI agents that keeps working 24/7.

Research → Build → Launch → Improve

More valuable than most $1,000 agentic engineering courses.

[Arcos @0xarcos]

https://t.co/0Cs6YR6cf5

[killaz @0xkillaz]

SpaceXAI engineer Lauren Tan just revealed why 99% of people are using GrokBot completely wrong.

And what she's building with 20+ autonomous AI agents is insane.

Lauren Tan:

"99% of users are barely tapping into 1% of GrokBot's true potential.

They run a single agent with no loops, no graphs, and no real coordination."

"Meanwhile, I'm running a fully autonomous team of 20+ GrokBot agents.

One Chief of Staff. One PM. 20+ specialized workers.

This is the next generation of engineering."

Here's the architecture:

GrokBot → Chief of Staff → PM → Workers → Autonomous Fleet

In this 1-hour session, Lauren Tan breaks down exactly how to build an entire AI agent team from scratch.

The kind of knowledge most people pay $1,000+ to learn.

And 99% of developers haven't even started exploring it.

Bookmark this before everyone catches on.

Watch the full workshop today.

Then read the complete breakdown below ↓

[Arcos @0xarcos]

https://t.co/0Cs6YR6cf5

[killaz @0xkillaz]

This is f*cking unreal. One GrokBot is already the old stack...

SpaceXAI senior engineer Lauren Tan, formerly at Cursor:

“99% of people open one GrokBot, ask it something and close the tab.

I operate 50+ GrokBot agents like a real company. One Chief of Staff manages my day, three PMs distribute the tasks, and 40+ workers ship everything overnight.

I just review the results over coffee.

That team does the work of 20 engineers.”

In this one-hour workshop, Lauren Tan shows how to build the entire GrokBot team from zero, step by step.

GrokBot → Chief of Staff → PMs → 40+ Workers → Autonomous Company

This is where it gets f*cking scary:

97% of engineers at SpaceXAI already operate teams like this, and the ones doing it are receiving $750K+ offers.

Everyone else is about to compete against them.

More valuable than most $5,000 agentic engineering courses.

Bookmark it and watch it today.

Then read how to build your own GrokBot team in the complete harness engineering guide below ↓

[rari🈀 @0xwhrrari]

https://t.co/3ANIWs48lJ

[killaz @0xkillaz]

This is absolutely f*cking wild. Andrej Karpathy spent 8 years working at OpenAI and Tesla.

Last week, he packed everything he learned into a free 2-hour masterclass.

Agents → Harness → Loops → Graphs

Some bootcamps charge $15K and still deliver less value than this.

Don't have 2 hours to spare right now?

Save it and watch before it disappears into your feed.

Then check out the guide below and start building your first agent harness ↓

[Arcos @0xarcos]

https://t.co/0Cs6YR6cf5

[killaz @0xkillaz]

@0xarcos agents can tighten the harness but only humans can loosen it is the whole article in one line

[killaz @0xkillaz]

@0xarcos so claude.md tells it what not to do and https://t.co/xZWTZApqlU actually stops it?

[killaz @0xkillaz]

Andrej Karpathy just broke down the 5 major shifts transforming LLMs into powerful AI agents.

Here's what the next generation of AI looks like:

→ Memory - AI that remembers and adapts to you
→ Multimodality - AI that sees and understands the world
→ Reasoning - AI that tackles complex problems
→ Search - AI connected to real-time information
→ Tools - AI that takes action and gets things done

Most people are still using LLMs as simple chatbots.

But Karpathy reveals the bigger picture:

Memory → Vision → Reasoning → Search → Tools

The future isn't just about writing better prompts.

It's about building AI systems that can think, learn, and act independently.

This 40-minute breakdown delivers more practical insight than many expensive AI agent courses.

Save it and watch before this becomes common knowledge.

Then explore how to build self-improving AI agent loops below

[Lunar @LunarResearcher]

https://t.co/X1bBruM26X

[killaz @0xkillaz]

@0xtombine giving ai access without boundaries is asking for trouble

[killaz @0xkillaz]

Holy sht. Claude Code just turned one reverse engineer into a full fcking team...

A professional reverse engineer with 12 years of experience tested REA and explained the part everyone misses.

The breakthrough is not decompilation.

It is the investigation loop:

Inspect the binary → capture evidence → ask the next question → repeat until the behavior is proven

Now split that loop across four isolated roles:

Analyst → Spec Writer → Rebuilder → Verifier

One extracts evidence.

One writes a behavior only spec.

One rebuilds without seeing the original source.

One tries to prove the rebuild wrong.

This 78 second clip explains the engine behind the entire system.

Watch it, then read the full 7 step Claude Code + REA setup below ↓

[Gyomei @Gyome1_]

REA hit #1 on GitHub trending, and most people still run it inside a single chat

Split it into 4 subagents and one command takes a program you own from sealed binary to a tested rebuild with a match rate

Set it up once, aim it at the next one: https://t.co/MxHPDVOZO2

[killaz @0xkillaz]

Anthropic just made “smart agents” look f*cking primitive...

Anthropic engineer:

“Anthropic hired me at $350K–$950K because I know how to engineer harnesses.

99% of people still hand build multi agent systems and wonder why everything falls apart.

I create systems that run loops, use sandboxed tools and convert yesterday’s logs into new skills.

Same model. 10x more reliable agents that keep improving overnight.”

Opus 5.5 → Agents → Harness → Loops → Graphs

In this short workshop, he reveals how to build the entire system from scratch.

step 1 → the harness manages loops, context and sandboxed tools

step 2 → the brain and hands work separately, cutting time to first token by 60%

step 3 → everything runs server side and resumes from logs after a crash

step 4 → failures become cheap, retry and replay instead of restarting everything

step 5 → yesterday’s logs become new memory and skills, so the system wakes up smarter

Anthropic calls this “dreaming”.

Stop collecting prompts.

Build the machine around the model.

Bookmark and watch it today.

Then read the complete Opus 5.5 harness engineering guide below ↓

[Movez @0xMovez]

https://t.co/4qTQfdoerp

[killaz @0xkillaz]

@henkjan that's right

[killaz @0xkillaz]

SpaceXAI engineer (ex-Cursor):

“I don’t prompt my GrokBots one by one. One Chief of Staff coordinates the team. This system creates and launches new cloud agents on any AI model.

You can get very far with one GrokBot. But scaling across multiple projects, codebases and tasks requires a full team of GrokBot agents.”

This 10 minute AMA with Lauren Tan reveals her exact SpaceXAI workflow and shows how she runs agent teams 24/7.

It replaces dozens of paid courses on agent engineering.

Watch it today and apply the workflow to your own system. Then read the article below for the complete architecture ↓

[rari🈀 @0xwhrrari]

Grok Bot can learn from your whole team

One bad memory can mislead everyone

What to save, what to keep private, and when to check the source

A practical guide to Grok Bot memory engineering

Bookmark and read below ↓ https://t.co/3SD7Ey9Q9N

[killaz @0xkillaz]

@shariar_design i think claude

[killaz @0xkillaz]

@Whail100x gmgm

[killaz @0xkillaz]

@Tesla I'd like to see some more tests with this Tesla