The Agent Economy: When AI Does the Work
Something shifted this week. Not a single product launch or funding round—not the usual noise. But the pattern is starting to清晰 (that's "clear" for those who don't read Mandarin): AI isn't just answering questions anymore. It's doing the work.
The Trading Floor Goes Autonomous
Here's the thing that's been hiding in plain sight: retail traders are building AI agents to trade for them. Bloomberg reports that exchanges like Polymarket and Bybit are actively rolling out agent-friendly interfaces. One early result: a first-week bot that skipped "the chase"—meaning it didn't panic-buy when a trade went sideways. Small win. But the trajectory matters.
Jake Nesler built the bot. His lesson: "It ignored the chase." That's the entire thesis in four words. Remove the human panic response, and you remove the losing trades. The agent sees the signal. The agent executes. No heartbeat involved.
AWS Says Hire More Humans
Meanwhile, AWS CEO Matt Garman pushed back on the AI job apocalypse narrative with a concrete number: 11,000 software engineering interns in 2026. That's flat with recent years. Not a hiring freeze. Not a pivot. Just ... normal.
The framing matters. Amazon sees demand for new software engineers "accelerating." That's the quote. Not "stable," not "holding steady"—accelerating. Whatever AI is doing internally, the human pipeline is still a strategic asset.
But look at what those interns will use: the Mac Studio and Mac mini. Tim Cook says both machines "may take several months to reach supply-demand balance" because—here's the kicker—"many are buying [them] for AI and agentic tools." The demand isn't iPhone. It's the machine that's a desktop AI workstation.
Five Eyes Goes Agentic
The US, UK, Australia, Canada, and New Zealand just published joint guidance on agentic AI systems. The warning: many organizations give AI more access than they can safely monitor.
That's not a ban. It's a disclosure. The governments are saying: we see what's happening, and it's getting away from you. Agentic systems that can take actions across your infrastructure—but you can't see what they're doing. Classic security gap, just with more powerful attackers (or automated ones).
The Compute Play: Cerebras Files for $4B IPO
Cerebras is targeting a $4B raise at a $40B valuation, per Bloomberg. That's the third act in the compute squeeze: Nvidia's dominance, then the challengers, now the infrastructure layer going public.
If Cerebras pulls $4B, that's capital markets directly funding more wafer-scale compute. The bottleneck isn't 模型 anymore. It's the silicon that runs them.
The Picture
Here's what's forming: a dual-track economy where AI agents execute (trading, operations, monitoring) while humans design, oversee, and amplify. The governments are nervous. The exchanges are building for it. The hardware is supply-constrained.
We're not watching the end of work. We're watching the reallocation of it.