ELON MUSK SAYS MONEY WILL STOP MATTERING, AND HE ALREADY KNOWS WHAT REPLACES IT.
In his sit-down with Walter Isaacson, the man who wrote his biography, Musk laid out where he thinks the economy ends up once AI and robots do the work.
His math:
"If you have robots and you get to space… you can get to, I think, potentially a million times the size of the Earth economy."
Then the twist:
"Money becomes irrelevant. I mean, there will be still a currency, but I think currency will be measured in fundamental physics metrics like wattage and tonnage, mass and energy."
It sounds like sci-fi until you look at where the AI money is going right now. A huge share of it goes into power plants, data centers and chips, which is exactly the kind of wealth he's describing.
That's how I run my trading agents too: follow the watts, not the hype.
If money really stopped mattering, what would you still want to own?
Save this before you lose it 🔖
STANFORD STUDENTS PAY $22,577 A QUARTER TO SIT IN THIS ROOM, AND YOU CAN WATCH THEIR AI AGENTS CLASS FOR FREE.
This is lecture 6 of CME295, Stanford's course on Transformers and LLMs. It's taught by Afshine and Shervine Amidi, the twins behind the Stanford cheatsheets a whole generation of ML students learned from.
In 100 minutes they go from the basics to the exact stuff people are building with today.
Here's what's inside:
0:00 Intro
8:53 RAG
44:31 Tool calling
1:12:42 MCP
1:22:53 Harness engineering
1:24:26 ReAct
1:29:02 Context compaction
1:31:00 Memory
1:33:37 Coding agents
1:38:53 Discussion
Start at 1:22:53. Harness engineering, ReAct, context compaction and memory are all packed into the last 20 minutes, and that's the part most paid agent courses never even get to.
You're not going to watch 100 minutes right now, so don't let this one disappear from your feed.
This is the same stack every trading agent I run sits on, so it's going straight into my notes.
Which part of the agent stack still feels like magic to you?
Save this before you lose it 🔖
THE GUY BUILDING GROK BOT AT SPACEX EXPLAINED EXACTLY HOW IT WORKS IN A FREE STANFORD LECTURE
Lee Robinson helped build Vercel and Cursor, and now he works on the team behind Grok Bot. At Stanford's CS146S he walked through the product, the infrastructure and the harness behind an agent that keeps working while you sleep.
Most AI tools sit and wait for your prompt. Grok Bot doesn't. It runs in the background, decides when it's time to act and finishes the job on its own, and this is the clearest look I've seen at how something like that is actually built.
Here's what's inside:
0:00 How we got here
3:38 What changed in the models
7:31 Inside an always-on agent
15:32 The harness
24:51 Context engineering
31:51 Where this is going
Start with the harness at 15:32. It's the part agent courses charge thousands for, and he explains it for free in under ten minutes.
You probably don't have 43 minutes right now, so don't let this one disappear from your feed.
Watch the lecture first, then read X's own guide to setting up Grok Bot below ↓ Together they give you the whole picture: how it works under the hood and how to put it to work for you.
I build my trading agents from the same pieces he walks through, so this one is going straight into my notes.
What's the first job you'd hand to an agent that never sleeps?
Save this before you lose it 🔖
SKIP NETFLIX TONIGHT AND WATCH STANFORD’S FREE 70-MINUTE CLASS ON HOW AI AGENTS ACTUALLY WORK
Stanford just started a new course on building AI agents, and the first lecture is free on YouTube. My favorite part is the story of how agents went mainstream. In March 2023 OpenAI released GPT-4, the first model that could follow long instructions and plan with some reliability. Less than two weeks later, Yohei Nakajima put out BabyAGI. You gave it a goal, and it wrote its own to-do list, worked through it and kept adding new tasks as it went.
Two days after that came AutoGPT. It hooked GPT-4 up to web search and your files and let it keep going on its own, without anyone typing the next prompt.
Diyi Yang, who teaches the course, calls this the moment the word “agent” probably entered everyday use. Then she gives her honest verdict on how well these two projects actually worked. It’s at 1:12 in the video, and it’s not what the hype threads from 2023 would make you expect.
Here’s what I took from it after running my own setups since then:
Every agent since is the same loop this lecture describes: think about the next step, act, look at what happened, think again. If your setup never looks at the result before its next move, it isn’t an agent yet.
The model underneath sets the ceiling. BabyAGI and AutoGPT only became possible because GPT-4 could plan, and most of what agents can do today comes from the models getting that much better.
“Runs on its own” was the exciting part, and it’s also the part I trust least. Mine still stop and ask me before anything touches money.
Here’s what’s inside the full lecture:
0:00 What the course covers
5:54 What AI does to homework scores vs exam scores
7:44 A short history of agents, from 1973 to AlphaGo
14:08 ReAct, the loop every modern agent runs on
14:50 GPT-4, BabyAGI and AutoGPT
16:31 How much models improved from GPT-3 to GPT-5
17:52 How fast coding agents got better
22:05 The anatomy of an agent
53:31 Workflows vs agents
58:41 Tool calls and MCP
Save this before you lose it 🔖