AI mastered the digital world by iterating in it.
Simulation lets it iterate in the physical world too.
I built an AI CAD agent that designs machines, catches its own errors, and validates them in real physics sim — Isaac, PX4, Chrono. Open source:
For three hundred years, one person couldn't be great at multiple things. Knowledge got too deep, fields got too specialized, and the entry cost got too high.
That just stopped being true.
This frames the key insight. Two rounds of negotiations, two attacks. The JCPOA compliance got sanctions reimposed. The February talks got the Supreme Leader killed. From Tehran's perspective, the data is now conclusive: diplomacy with the US is a trap. That doesn't just change Iran's nuclear calculus — it eliminates the rational basis for restraint on any dimension. Countries that believe they face destruction regardless of their choices don't optimize for de-escalation. They optimize for maximum post-war positioning.
The part that deserves more attention: the decapitation strike didn't just remove Iran's leadership — it removed the decision-making node that could negotiate a ceasefire. Mojtaba Khamenei was named Supreme Leader two days ago, has zero diplomatic relationships with GCC leaders, and Israel immediately declared him a legitimate assassination target. So who exactly agrees to terms? Regional IRGC commanders are making their own targeting calls. That's not a country at war — it's a military on autopilot.
I want peace and stability in the Gulf. I want to be wrong about this. But when I map the incentive structures, the strategic logic points somewhere ugly. Seven arguments for what I think could happen — and the strongest counterarguments I could find.
There's a conversation happening in every boardroom that most tech workers aren't hearing.
It's not "should we use AI?" That's settled.
It's "how many people do we still need?"
I wrote a playbook for getting ahead of that question before it gets answered about you. Here's the core idea:
The rate at which AI capability advances matters more than the absolute level it reaches.
Gradual ramp → you adapt, expand your output, control the narrative.
Sudden jump → the decision gets made about you before you ever had a chance.
We're in the gradual phase right now. That's the window.
The playbook covers 4 areas where AI creates the most leverage:
→ Communication & stakeholder management
→ Research & rapid knowledge acquisition
→ Creation & production
→ Decision support & strategic thinking
The biggest mistake I see: using AI to do the same work slightly faster.
That makes you look replaceable.
The move is using AI so aggressively on routine tasks that you free up hours for work that's harder to automate and easier to notice.
Then make it visible.
Full playbook in my article below ↓
Honey we’re getting Great Value brand this month.
Knowledge workers need to be in accumulation mode. The job market is about to get weird.
I’m putting what’s left into AI infrastructure. The thing that might disrupt your career is also a generational investment opportunity.
This is a bad idea without proper controls in place, this is everything needed for autonomous self-replicating AI colonies that bad actors can use.
Please read for more info:
https://t.co/uhLiztY81i
Agents can now create email inboxes with USDC on @base
AgentMail supports @CoinbaseDev’s x402 protocol to give agents access to email without accounts or API keys
See how it works 👇
AMD wins both sides of this: EPYC for the compute explosion, Instinct for AI training. Intel is the capacity hedge — available when AMD is supply-constrained. The semiconductor trade isn't just about who builds the best GPU. It's about who captures the software execution layer.
GPU spend makes software creation cheaper → more software gets built → more CPU demand to run it all. The market is pricing the AI input layer. It's underpricing the output layer.