Company@X — 2026-05-12#

Signal of the Day#

Google fundamentally shifted its hardware strategy by announcing “Googlebook,” a new high-performance laptop designed specifically for its newly unveiled “Gemini Intelligence” platform. Coming this fall, it signals Google’s move toward tight, vertical hardware-software integration centered entirely around AI agents, matching similar OS-level AI automation capabilities rolling out to upcoming Android devices.

Key Announcements#

[Google DeepMind] · Source Google DeepMind unveiled experimental demos for an “AI-enabled mouse pointer” that allows users to prompt Gemini directly on their screens using motion, speech, and natural shorthand. The system understands what the cursor is hovering over, removing the need for precise text instructions and offering a glimpse into future agentic UI patterns. These experiments are now live in Google AI Studio.

[AWS Cloud] · Source AWS launched Amazon Redshift RG, powered by their custom Graviton silicon . Built to meet the demands of modern analytics and agentic AI workloads, the new instances promise up to 2.4x the performance of previous generation RA3 instances at a 30% lower cost per vCPU, while supporting open formats like Apache Iceberg and Parquet .

[Hugging Face] · Source The Hugging Face Hub officially crossed 1 million public datasets. CEO Clément Delangue noted a sharp acceleration in dataset creation driven by AI agent development, pointing out that the number of hosted datasets doubled in just the past eight months.

[Cursor] · Source Cursor introduced a new “Fast mode” for Anthropic’s Claude Opus 4.7. Aimed at latency-sensitive developer workflows, the new feature delivers 2.5x the standard generation speed, though it comes at a 6x cost premium.

[Vapi] · Source Voice agent platform Vapi announced a $50M Series B led by Peak XV Partners, bringing their total funding to $72M. The platform has crossed 1 billion processed calls, pointing to massive enterprise adoption of platform-layer infrastructure for predictable latency, guardrails, and human escalation in voice interfaces.

[Meta] · Source Meta silently released Sapiens2, a new family of high-resolution models trained on 1 billion human images. Available in six sizes up to 5 billion parameters, the models achieve state-of-the-art results for pose estimation, body-part segmentation, and surface normals at up to 4K resolution.

Also Noted#

• [Interfaze] (Source): YC-backed Interfaze launched a new model architecture that outperforms Claude Sonnet 4.6 and Gemini 3 Flash on nine benchmarks including OCR, vision, and speech-to-text tasks.
• [Hugging Face] (Source): Hugging Face researchers teased the release of two new open-source SLMs dropping this week, with one allegedly matching state-of-the-art accuracy while being up to 93x smaller.
• [Waymo] (Source): Waymo released footage of its autonomous driving system successfully detecting and defensively evading a 125+ mph street racing crash on LA’s I-10 freeway .
• [OpenMed] (Source): OpenMed Agent shipped in preview, utilizing Hugging Face endpoints and MCP to orchestrate over 1,000 medical models for clinical extraction.