Musk Launches Grok 4, Amazon Agent Store, YouTube AI Video Crackdown

xAI released its Grok 4 flagship model and introduced a $300-per-month SuperGrok Heavy plan. The launch also includes Grok 4 Heavy, a multi-agent version promised to deliver higher performance.

xAI says Grok 4 beats Google’s Gemini 2.5 Pro and OpenAI’s latest on multiple benchmarks, while Grok 4 Heavy scores 44.4% on Humanity’s Last Exam with tools. The new tier becomes the priciest among major AI providers and grants early access to upcoming coding, multimodal, and video models.

The debut follows Grok’s antisemitic X posts and the sudden resignation of X CEO Linda Yaccarino earlier the same day. Controversies now shadow xAI’s push to win enterprise adoption and cloud partnerships.

Amazon Web Services will launch an AI Agent Marketplace at the AWS Summit in New York on July 15, giving enterprises a centralized hub for autonomous software. Anthropic is an inaugural partner, offering its in-house agents and developer API through the store.

Startups can list agents, set standalone pricing, and share revenue with AWS. Enterprise users gain a single place to browse, install, and compare agents without juggling multiple vendor silos.

Sources say the marketplace exposes Anthropic’s agents to customers already using competing tools, broadening its reach beyond direct channels. The launch puts AWS in the same competitive lane as Google Cloud and Microsoft, which opened similar agent marketplaces earlier this year.

YouTube will update its Partner Program monetization policies on July 15 to bar revenue from “inauthentic” content like mass-produced and repetitive videos. The company says the move targets material now easily churned out by AI tools.

Help pages already require “original” uploads, and the new language will spell out examples of what is disqualified. Creator liaison Rene Ritchie calls it a minor clarification and says formats such as reaction videos remain eligible.

The article states that unchecked AI-generated “slop” threatens YouTube’s reputation and value. It adds that the clearer rules give YouTube grounds to expel such channels from its Partner Program.

MIT and University of Wisconsin researchers used a machine-learning pipeline to design and 3D-print two autonomous underwater gliders inspired by marine animals. The AI-generated prototypes moved more efficiently and expended less energy than a handmade torpedo-shaped glider in pool tests.

The system trained on dozens of deformed 3D models, optimized each shape’s lift-to-drag ratio, and delivered predictions within 5% of wind-tunnel results. It produced a jet-like two-wing craft tuned for 9-degree dives and a four-fin flat-fish model tuned for 30 degrees, both printed as lightweight hollow shells around standard pumps and sensors.

AI now opens a vast design space beyond traditional tubes, yielding glider shapes engineers had never trialed. The semi-automated process replaces slow physical prototyping and surfaces energy-saving configurations ready for ocean research.

Every’s Thesis column features an essay from investor-designer Willem Van Lancker declaring that intentional friction, not effortless prompts, builds lasting mastery in the AI era. He argues that removing all struggle through automation strips newcomers of the formative grind that shapes judgment and taste.

Drawing on all-nighters at RISD and stints at Apple and Google, Van Lancker shows how repeated critique, failure, and iteration create the “proof of work” employers now seek. He warns that entry-level jobs, the traditional training ground, are disappearing as AI generates polished outputs in seconds.

The piece contends that knowledge risks becoming transactional and shallow when AI handles the hard parts. In this environment, visible evidence of sweat equity—projects built, iterations survived—quickly outranks degrees or titles.