The Singularity's hard part, it turns out, was never the agent, it was the model underneath it. So argues Moonshot AI CEO Zhilin Yang, noting that while everyone raced on reasoning, Claude quietly bet on agents. He calls a pure reasoning model "a fish tank with a brain in it," thinking beautifully while touching nothing, and an agent that same brain wired into the world, with the endgame stated plainly: "we want K2 to help build K3." The recursion is already cashing checks. On a private cybersecurity benchmark, his lab's Kimi K3 emerged as the workhorse for hunting vulnerabilities, near-frontier recall at a fraction of the cost, while GPT-5.6 Sol took the crown at 7x the price, GLM 5.2 undercut everyone, and Fable 5 refused 100% of the tasks, turning safety into a service outage. Capital has drawn its conclusions. Moonshot is preparing a Hong Kong IPO in as little as six months at a valuation above $30 billion, and one of its employees toasted the moment by quoting Che Guevara: the schools and hospitals will come, "not because they've had a change of heart," but "because we came."
The revolution ships with a rate card. Chamath Palihapitiya warns that forcing American firms to pay $26-56 per million tokens while adversaries pay 50 cents is "the Cold War Soviet collapse in reverse," an argument for embracing open source before economics does it the hard way. Perplexity's CEO supplies the corporate memento mori, recalling how Sun Microsystems shed 96% of its value to Linux and commodity hardware, a fate now on the menu for any closed lab. Alibaba is volunteering to be the commodity. Qwen3.8, a 2.4 trillion parameter model billed as second only to Fable 5, is going open-weight, its upgraded Token Plan bundles the whole family at 40% off and plugs into Claude Code and Cursor, and chip unit T-Head is open-sourcing the entire SAIL stack to storm CUDA's moat from below.
Some of the closed labs are not idle, merely priced differently. Forecasters put Opus 5 at near certainty for next week, slightly behind Fable but ahead on efficiency, so that "All will be right in Claudesylvania." The frontier may also be simpler than its price implies. Princeton's DeepLoop shows looped Transformers scale depth stably once residual rules account for revisited parameters, and its author passes along the rumor that some frontier models are basically a 48-layer transformer looped twice. The emperor has weights, just fewer than advertised.
Where the weights live is now spycraft. A longtime CIA operative's final mission surveilled the UAE's G42 and its China ties, a saga that wound into Washington's decision to widen Gulf access to advanced chips. The counterintelligence file thickens with the saga of Microsoft's "digital escorts," China-based engineers feeding code to cleared US minders who pasted it into Pentagon clouds unvetted, a practice now banned by law.
Back home, the substrate fights its own permitting wars. Oracle's supercampus buildout is absorbing multibillion-dollar cost surprises, including a $165 billion New Mexico project on the rocks. In California, a onetime strip club operator is pushing a $10 billion, 330-MW Imperial Valley campus through lawsuits and a county moratorium. The neighbors are organizing. 142 protests across 42 states marked the first coordinated national demonstration against AI infrastructure, organized by HumansFirst, a grassroots group co-founded by a former Tea Party leader, with just 14% of Americans wanting a data center next door, all while Washington's light-touch AI agenda collides with reality.
Meanwhile the atoms keep automating themselves. Manual transmissions fell to 0.6% of new American cars, the stick shift yielding to software the way everything eventually does. Overhead, the first FireSat wildfire-detection satellites launched through the very smoke they were built to see, able to spot a 5-by-5-meter fire through smoke and clouds on the way to hourly global imagery by 2029. And far beyond, New Horizons woke from a 321-day nap in the Kuiper Belt, its all-clear taking nine hours to cross 5.9 billion miles. Patience, it turns out, also scales.
The friction lives in the human loop. Medicare's new AI prior-authorization pilot pays vendors a cut of "averted expenditures," which is one way to teach a model to say no. Automating away Gen Z entry-level jobs, warns MIT's Andrew McAfee, burns the apprenticeship ladder along with tomorrow's power users. The winners, meanwhile, face a private jet shortage. New York's mayor wants landlords to disclose AI-doctored listing photos. And Big Pizza is losing its moat to the delivery apps, with Papa Johns closing 200 stores, Pizza Hut sold to private equity, Domino's down 30%, and quick-service delivery sales off 15% since 2021, because in this economy even the moats get eaten.
The future is already here, it just isn't evenly delivered.
Follow me via:
X: https://x.com/alexwg
Substack: https://theinnermostloop.substack.com/
LinkedIn: https://www.linkedin.com/newsletters/7404871891775025153/
YouTube: https://www.youtube.com/@alexwg
Spotify: https://open.spotify.com/show/1thtZk5vHTXbtDHezPT7tl
Threads: https://www.threads.com/@alexwissnergross
RSS: https://theinnermostloop.substack.com/feed