Sources
- AI at Meta / @AIatMeta
- Amazon Web Services / @awscloud
- Anthropic / @AnthropicAI
- Cursor / @cursor_ai
- Google / @Google
- Google Cloud Tech / @GoogleCloudTech
- Google DeepMind / @GoogleDeepMind
- Grok / @grok
- Hugging Face / @huggingface
- Microsoft / @Microsoft
- OpenAI / @OpenAI
- OpenClaw🦞 / @openclaw
- Sequoia Capital / @sequoia
- Tesla / @Tesla
- Twitter / @a16z
- Waymo / @Waymo
- xAI / @xai
- Y Combinator / @ycombinator
Company@X — 2026-05-15#
Signal of the Day#
Andreessen Horowitz (a16z) is opening its first-ever overseas office in Japan this summer, marking a major strategic expansion aimed at capitalizing on the Japanese startup ecosystem. Aligning with Prime Minister Takaichi’s “New Technology Nation” strategy, a16z intends to focus heavily on defense and security innovations, linking technological superiority directly to diplomatic and defense capabilities.
Key Announcements#
[AWS] · Source AWS shared year-one metrics for AWS Transform, its agentic AI service for enterprise migration, revealing that the tool analyzed 4.56 billion lines of code and saved teams 1.69 million hours of manual effort. AWS also signaled the platform’s future roadmap, which will include full-stack Windows and mainframe coverage, multi-surface access, and a new Agent Builder designed to help enterprises scale their internal playbooks.
[Microsoft] · Source Microsoft has released Lens, a new 3.8-billion parameter text-to-image model available on Hugging Face. The model provides highly efficient training capabilities and supports high-resolution image generation up to 1440x1440.
[xAI] · Source Grok is moving aggressively into enterprise workflows, launching three new connectors that allow the AI to interface directly with Microsoft Teams, Salesforce, and Box. The Box integration specifically provides Grok Business users with secure, permissions-aware access to internal corporate content. Furthermore, users can now plug their Grok subscriptions directly into the NousResearch Hermes Agent.
[Google Cloud] · Source Google Cloud detailed the networking infrastructure behind its TPU 8t rack-level connectivity to the Virgo fabric. The architecture utilizes high-radix switches to reduce network layers, operating on a flat, two-layer non-blocking topology. Separately, Google Cloud rolled out QueryData for AlloyDB, a system allowing developers to query complex databases using natural language and high-speed vector search.
[OpenClaw] · Source The open-source agent framework OpenClaw dropped a new release that is roughly 3.5x faster, backed by continuous end-to-end RTT tests over real message channels to prevent silent regressions. The update also ships with strict new security guardrails, including root-bounded filesystem controls (fs-safe), policy-driven network egress routing (Proxyline), and smarter command approvals.
[Tsinghua / Open Source AI] · Source A new 30B-A3B reasoning model was released on Hugging Face, utilizing a unified scaling recipe tailored for proof search. According to the release, the model hits gold-medal performance across major physics and math Olympiad evaluations, solving IPhO natively and passing IMO/USAMO benchmarks via test-time self-verification and refinement.
Also Noted#
- [Vulcan Tech] (Source): Partnered with the state of Texas to launch SAM AI, the nation’s first AI government concierge that enables citizens to search the state legal code and automate permit and license renewals.
- [AgentPhoneHQ] (Source): Launched a telecommunications API designed exclusively for AI agents, provisioning them with their own real-world phone numbers and trusted identities.
- [Basedash] (Source): Shipped Model Context Protocol (MCP) connectors, upgrading its agents from merely reading databases to actively manipulating connected SaaS platforms like HubSpot, Slack, and GitHub.
- [Hugging Face] (Source): Highlighted its dedicated Storage service—featuring Xet deduplication, a built-in CDN, and simple per-TB pricing—as a streamlined alternative to complex git workflows and expensive cloud object storage.