No f*cking way, SpaceXAI engineer Lauren Tan:
"You don't need a 10,000-person engineering org like Meta anymore, you need a codebase your agents can't break
Someone drops a screenshot in Slack with just '???', and my agent finds the broken feature and ships the fix itself"
53 minutes on going from reading every line her agents wrote to a team that merges its own PRs while she sleeps
Verification → Feature Map → Evals → Hard Rules In CI → One Engineer Shipping Like A Whole Team
Her PMs have an agent that tells them what she shipped last night
Hire one more engineer or spend tokens so the cheapest model writes good code, she picks tokens
Most people still believe their agent the 100th time it says "I found the bug"
Worth more than any $3,500 AI agent course
Save it, watch it tonight
Grok Bot can now make your calls, sort your inbox and run your week while you sleep
Most people still use it like a chatbot
Everything it can do in one guide: first bot, routines, a full team, calls, and 5 rules that keep it off your money ↓ https://t.co/G8aN1eOOcF
This is f*cking insane, an Anthropic engineer just explained why your Claude Code usage limit is gone by Wednesday:
"The bottom session uses twice the amount of tokens, just because task two and three were still carrying task one around the whole time"
"When a team spend is unexpectedly high, this usually traces back to just the biggest model being left on as a default"
In a 27-minute session, the Claude Code team shows where every token actually goes
Model → Effort → Context → Cache → Subagents
Clear between tasks, compact before you step away, never switch models in the middle of one
You don't need to use Claude less, you need to stop paying for task one during task three
Worth more than any $2,000 AI course
Bookmark it and watch it tonight
This is f*cking insane, an Anthropic engineer just explained why your Claude Code usage limit is gone by Wednesday:
"The bottom session uses twice the amount of tokens, just because task two and three were still carrying task one around the whole time"
"When a team spend is unexpectedly high, this usually traces back to just the biggest model being left on as a default"
In a 27-minute session, the Claude Code team shows where every token actually goes
Model → Effort → Context → Cache → Subagents
Clear between tasks, compact before you step away, never switch models in the middle of one
You don't need to use Claude less, you need to stop paying for task one during task three
Worth more than any $2,000 AI course
Bookmark it and watch it tonight
Anthropic measured it: 80% of an agent's results come down to how it spends tokens, not the model you pay for
LangChain kept the model and changed only the setup: 52.8% → 66.5%
Prompt engineering is 1 of 9 layers, here's the full map + an AGENT.md ↓ https://t.co/IIPAQScnZ7
What the f*ck, SpaceXAI engineer Lauren Tan:
"One lazy workaround in your codebase spreads like a virus, a few days later every agent on the team has copied it
So I stopped fixing my agents, I fix the kitchen they cook in"
38 minutes on her jump from 5 agents she had to watch to 2,000 PRs a month, a number she never even set as a goal
Babysit Every Chat → Agents Verify Themselves → Lint Rules Stop The Bleeding → A Gardener On The Team → A Michelin Kitchen
Now designers, PMs and even CEOs ship features into the Grok Bot codebase, and the agents keep it clean for them
Your codebase is your agent's real memory, whatever's in it gets copied
Most people still fix the same agent mistake by hand, again and again
Worth more than any $9,000 AI engineering program
Bookmark it and watch it tonight
What the f*ck, SpaceXAI engineer Lauren Tan:
"One lazy workaround in your codebase spreads like a virus, a few days later every agent on the team has copied it
So I stopped fixing my agents, I fix the kitchen they cook in"
38 minutes on her jump from 5 agents she had to watch to 2,000 PRs a month, a number she never even set as a goal
Babysit Every Chat → Agents Verify Themselves → Lint Rules Stop The Bleeding → A Gardener On The Team → A Michelin Kitchen
Now designers, PMs and even CEOs ship features into the Grok Bot codebase, and the agents keep it clean for them
Your codebase is your agent's real memory, whatever's in it gets copied
Most people still fix the same agent mistake by hand, again and again
Worth more than any $9,000 AI engineering program
Bookmark it and watch it tonight
Claude's smallest model just beat Claude Opus with one text file
84 real tasks: Haiku with skills 27.7%, Opus without them 22.0%
The difference wasn't the model, it was writing the work down once
6 parts, 6 broken skills and a SKILL.md you can copy ↓ https://t.co/AfkuzTIpei
Holy f*ck, Andrej Karpathy, the OpenAI co-founder who coined vibe coding:
"Start a new chat every time you switch topics, a long chat makes the model dumber and slower
80% of what I ask AI, I just say out loud into my phone, I'm too lazy to type"
In this 2-hour lecture, he shows how he went from one chat window to a $200 a month AI stack that reads books with him, codes his apps and makes his podcasts
One Chat → New Chat Per Topic → Thinking Models → Deep Research → Vibe Coding → An AI That Knows You
He builds a working game in Cursor by typing 5 sentences, then adds confetti and sound without touching a line of code
The chat is the model's working memory, everything you leave in it quietly changes the answer
Most people still run their whole life through one endless chat
Worth more than any $3,000 AI course
Bookmark it and watch it this week
Holy f*ck, Andrej Karpathy, the OpenAI co-founder who coined vibe coding:
"Start a new chat every time you switch topics, a long chat makes the model dumber and slower
80% of what I ask AI, I just say out loud into my phone, I'm too lazy to type"
In this 2-hour lecture, he shows how he went from one chat window to a $200 a month AI stack that reads books with him, codes his apps and makes his podcasts
One Chat → New Chat Per Topic → Thinking Models → Deep Research → Vibe Coding → An AI That Knows You
He builds a working game in Cursor by typing 5 sentences, then adds confetti and sound without touching a line of code
The chat is the model's working memory, everything you leave in it quietly changes the answer
Most people still run their whole life through one endless chat
Worth more than any $3,000 AI course
Bookmark it and watch it this week
Same model, same prompt, and their agents ship while yours say "done" on broken work
The difference is the harness: what it reads, what it can touch, who checks it, when it stops
6 parts + a 3-agent setup you can copy tonight ↓ https://t.co/vQYjiFqn6O