@BlockTheMiddle
Account based inUnited States
About this account
- Account based in
- United States
- Connected via
- United States App Store
Account-level information from X, not a live location or the device used for a specific post.
Own your data. Know your word. Move your body. Father of 3. ⬇️ Community ⬇️
Texas
Joined April 2021
- Tweets1.8K
- Following452
- Followers617
- Likes545
finance.yahoo.com/sectors/te…
as we've spoken about repeatedly on my channel...verifiable intelligence by way of verifiable compute will lift AI and the overall digital economy into a new realm much like what happened when we created new markets by measuring electricity with the kilowatt-hour etc. exciting times!
He's right about the asset class. But futures markets need a measurable underlying. Otherwise you just get 2008 all over again. Right now compute is priced on time and usage, not physical work. Nobody can tell you what an hour of AWS actually produced in energy, efficiency, or output quality. You can't build a legitimate derivatives market on a parking meter. The bottleneck isn't just power and chips. It's that nobody knows if anyone has built the unit of measurement yet. 👀 Every trillion-dollar market in history started with one.
LARRY FINK JUST PREDICTED A NEW ASSET CLASS
The BlackRock $BLK CEO laid out where he sees the bottleneck of the AI buildout at the Milken Institute.
His framing of the problem:
"The United States is short power, short compute, short chips. There are going to be shortages in all three. We just don't have enough compute power right now."
His prediction for what comes next:
"I actually believe a new asset class will be buying futures of compute."
Compute futures do not exist as a standardized tradable contract today. Fink is calling out the emergence of one.
Digital Oil Hunter retweeted
we just helped a GTM engineering team go from manual workflows to a fully deployed Claude Code automation stack in 8 weeks...
and we used ONLY Claude Code
the most UNDERUTILISED GTM build tool of 2026
so i just sat down, pressed record, and dropped a full breakdown going through the ENTIRE Claude Code GTM setup behind how we did this
here's what's included:
1. the exact terminal setup across Mac, Linux, and Windows that gets Claude Code running correctly from day one
2. the project brain file structure: what to include, what to leave out, and the auto-update mechanic that fixes repeated mistakes permanently
3. plan mode from scratch: the cost comparison, the exact steps, and when to skip it for simple tasks
4. 5 GTM skill files ready to install: lead scraping, email labeling, proposal generation, outbound sequence writing, and client onboarding
5. the MCP install process, token cost checks after every install, and the best MCPs for GTM work
6. how to cut token usage by 50 to 100x by converting MCPs into skills
7. sub-agents and agent teams: the 3 cases where they earn their cost and the reliability math for parallel runs
8. context management: what is eating your context before you type anything and how to fix it
9. /compact and /clear: when to use each and what happens when you get it wrong
10. model selection for parent vs sub-agents: Opus, Sonnet, and Haiku mapped to the right task
11. Modal deployment: any skill as a live URL in under 2 minutes with a form interface
12. connection to n8n, Make, or Zapier so your Claude Code automations trigger from anywhere in your stack
plus a bonus walkthrough of a real GTM engineer who replaced his entire Clay enrichment workflow with Claude Code and now processes 272 leads per second at 40-60% lower cost
like + comment "CODE" and i'll DM you EVERYTHING
(must be following + RT for priority access)
i should never be sharing this but f*ck it
seedance 2.0 + claude code + tiktok is THE best combo for AI videos
i cracked the formula for generating videos that look hyper realistic & make your audience feel like "holy shit this person gets me"
i'm finally sharing my FULL system with you..
here's what you're getting:
- my prompting method for realistic voices (works every time)
- my realistic human movements & breathing claude skill (this is key for realistic videos)
- my exact method on how to make infinite length videos that maintain consistency
- how to get AI tools for dirt cheap (90% off)
RT + reply 'UGC' and i'll send you the step-by-step system (must follow so i can dm)
Jeff Bezos is raising $100 billion to buy factories. Why?
Because sensor data from real production floors is the only way to train Physical AI that actually works.
This guy already has one. And he’s inviting people to come build in it.
The factory exists. The missing piece is a measurement layer that proves what happened on the floor, tied to the humans who built it, cryptographically signed.
That infrastructure exists too.
We built Guardian Bikes into a vertically integrated factory doing $100M+ in revenue - tube lasers, robotic welding, CNC, powder coating, assembly - all under one roof in Indiana with 500,000+ sq ft of production space.
Here’s what I’ve realized: we’re sitting on one of the rarest assets in robotics and physical AI, a real, high-volume American factory with full operational control and the willingness to let you break things.
Most robotics companies are building incredible technology but struggling to find real deployment environments. Demo cells and lab setups only get you so far. You need messy, high-mix, real-world production to actually train and validate.
We have that. And we’re building an AI-native MES from the ground up with full sensor instrumentation and computer vision baked in.
So here’s an open invitation: if you’re building robotics or physical AI for manufacturing - humanoids, manipulation, autonomous mobile robots, vision systems, whatever - and you need a real factory to develop and prove your technology, let’s talk.
We’ll give you the environment. You bring the technology. We’ll build the future of American manufacturing together.
DMs open.
Joanna Stern spent 12 years at the WSJ covering tech. When she was deciding whether to leave, she asked everyone around her. Friends hedged. Colleagues hedged.
ChatGPT said quit.
She did.
Her explanation: "People do not tell you what to do because if it went wrong, you'd feel bad." AI doesn't carry that weight.
We're so focused on whether AI will take our jobs. We're not paying attention to the fact that it's already advising us on whether to leave them.
In 10 days, three levels of government gave three different answers to the same question: who controls where AI gets built?
The White House wants to preempt state laws and fast-track permitting. Sanders and AOC want to freeze construction until Congress passes worker and environmental protections. 12+ states filed their own moratorium bills and aren't waiting for either side.
One data center consumes as much electricity as 100,000 homes. $98 billion in projects blocked or delayed last year. Communities are treating these facilities the way they once treated industrial polluters. And the federal government's response is to try to take the decision out of their hands.
1/ Dorsey and Botha published "From Hierarchy to Intelligence." The core argument: hierarchy is an ancient information-routing protocol. Roman legions, Prussian generals, railroad org charts. All solving one bottleneck. AI makes that bottleneck disappear.
The historical framing is good.
2/ Block is building a "company as intelligence." Two world models (internal operations + customer transaction data), a centralized AI coordination layer, and three roles: ICs, temporary DRIs, player-coaches. No permanent middle management.
The bet: speed compounds as competitive advantage.
3/ The essay diagnoses centralized human information routing as the constraint. Then proposes centralized AI information routing as the solution.
I watch this same pattern in data infrastructure. Replacing the humans inside a centralized system just makes centralization faster. The control structure stays where it was.
4/ The strongest pushback from the replies: "replacing a layer is not the same as replacing the function."
Managers route information, yes. They also get two people who disagree to commit to the same direction. An AI can surface every data point in the company and still can't do that.
5/ Block laid off roughly 4,000 people, nearly half the company, about a month before publishing this. The essay frames it as permanent restructuring.
The question the comes up yet again for me: who owns the intelligence layer in this new architecture? Whoever controls the context controls the coordination. That was true of VP chains. It's true of algorithms.
The White House AI framework preempts state-level privacy and oversight laws.
The federal government stepped in to prevent states from building their own data governance. That's the function of this policy.
U.S. hyperscalers control 85% of Europe's cloud market. 61% of European CIOs say they're increasing reliance on sovereign solutions in 2026.
Europe saw centralization clearly enough to start organizing against it. America made it official.
$200B in net new value pools over five years.
That's BCG's sizing for agentic AI services. And they chose banking loan origination as the engineering proof point.
The headline number isn't what matters most. The talent data is. 🧵
3/ The talent signal everyone should be reading twice: AI product engineer roles growing 40-50% CAGR. Manual QA and Java dev roles slowing. BCG says providers must "actively rebalance their workforces through targeted hiring and rapid reskilling."
I've been tracking an architecture that closes exactly this gap. Physics-based verification. Every data point locked to a specific place and time, sealed cryptographically so the output is provably real.
The Bezos approach puts all the created value with the fund. Amazon did it with retail. Google did it with search. The platform captures the value. The participants get a service.
This architecture is built on a different principle: the proof of the work belongs to whoever did the work.
Jeff Bezos is raising $100 billion through Project Prometheus, co-founded with former Google exec Vik Bajaj, to acquire aerospace, chipmaking, and defense manufacturing companies and run AI through them.
The factories are the entrance ticket. Decades of sensor data sitting in servers is the actual acquisition. And that data has a provenance problem that determines whether any of this actually works for defense and aerospace. 🧵
Train an AI on unverifiable data, you get an unverifiable AI. For consumer apps, tolerable. For aerospace components and defense contracts, where the Iran war is simultaneously elevating demand and disrupting supply chains, that's a catastrophic liability.
The timing matters: these acquisitions are getting more expensive and more strategically urgent at the same time.
Those factories have been collecting sensor data for decades. Temperature, pressure, vibration, error rates. Most owners don't know what to do with it.
Bezos's AI needs real-world industrial physics to train on. You only get that by owning the machines that produced it.
But a sensor reading right now is just a number in a database. No proof the timestamp is accurate. No verified calibration record. No chain of custody connecting it to a specific machine at a specific moment.