Company@X — 2026-06-10#

Signal of the Day#

Anthropic made a massive strategic and policy move, publishing an Economic Policy Framework with a $200 million evaluation fund and a $150 million national AI fellowship, while publicly urging governments to establish authority to block or revoke unsafe frontier AI models. This signals a coordinated shift toward proactive, highly funded initiatives to shape the inevitable regulation and economic impacts of advanced AI models.

Key Announcements#

Google · Source Google released “DiffusionGemma,” an experimental open-weight model (Apache 2.0) that generates text up to 4x faster by replacing sequential token prediction with a diffusion process. By drafting and refining entire blocks of text simultaneously, it unlocks real-time self-correction and excels at formatting complex code and math. This represents a significant architectural exploration beyond standard autoregressive LLMs.

Anthropic · Source CEO Dario Amodei released “Policy on the AI Exponential,” accompanied by an Advanced AI Framework that explicitly calls for government authority to block unsafe model releases. To back these frameworks, Anthropic is launching a $200 million fund to evaluate economic disruption and a $150 million national AI fellowship program.

Google · Source Google is introducing “Antigravity,” a feature that allows Search to build custom, ongoing experiences—effectively “mini apps”—for complex workflows like event planning. The feature will roll out in the coming months, initially limited to Google AI Pro and Ultra subscribers in the U.S..

UseKernel & Anthropic · Source A new partnership equips Claude managed agents with browser capabilities for the first time. Managed agents can now navigate the web and execute long-horizon, scheduled tasks autonomously without requiring developers to manage the underlying infrastructure.

xAI · Source xAI launched the Grok Voice API, claiming state-of-the-art performance in timing and tone at a fraction of competitors’ costs. According to ServiceNow AI Research, the underlying model (Grok Voice Think Fast 1.0) sets a new Pareto frontier on the EVA-Bench benchmark, outperforming existing systems on accuracy without sacrificing user experience.

AWS · Source AWS made its Amazon EC2 M9g and M9gd instances generally available to developers. Powered by the new AWS Graviton5 processors, the instances deliver up to 25% better compute performance for cloud workloads.

Cursor · Source Cursor updated its code review agent, making it over 3x faster and 22% cheaper while finding 10% more bugs. Developers can also use the /review command to run Bugbot locally to catch issues before pushing code.

Also Noted#

• Amplitude (Source): The company launched Wave, a proactive product agent that analyzes user data to draft complete product specs and track experiment outcomes.
• EliseAI (Source): The vertical AI startup crossed $200 million in ARR with 100% year-over-year growth, proving the viability of highly regulated enterprise AI in housing and healthcare.
• Waymo (Source): Partnering with TU Delft, Waymo published new research in Nature Communications that establishes a behavioral benchmark for how competent human drivers respond to conflict and uncertainty.
• Tesla (Source): The company is officially rolling out its Summon feature for the Cybertruck.
• Extend UI (Source): The team open-sourced an MIT-licensed library of 14 React components specifically designed for building document agents.