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 — 2026-04-29#
Signal of the Day#
Microsoft reported a blowout $82.9 billion Q3, revealing that its AI revenue has now surpassed a massive $37 billion annual run rate and that paid Microsoft 365 Copilot seats exceed 20 million. This confirms that enterprise spending on “agentic computing” is translating into immediate, scaled commercial reality rather than just sustained R&D investment.
Key Announcements#
[Google Cloud] · Source Google Cloud launched the Gemini Enterprise Agent Platform, a unified ecosystem for building and deploying production-ready AI agents with access to over 200 models, including Gemini and Gemma 4. The platform introduces a new Agent Registry to act as a single source of truth for indexing and governing internal enterprise tools, and pairs with a newly generally available remote Model Context Protocol (MCP) server on Cloud Run.
[OpenAI] · Source OpenAI confirmed that its DevDay developer conference will return to San Francisco on September 29, 2026. To drive early engagement, the company is running a ticket giveaway contest requiring developers to build playable projects utilizing the GPT-5.5 and Image Gen models.
[Cursor] · Source Cursor released the Cursor SDK, allowing developers to embed the exact runtime, harness, and models that power the popular code editor into their own tools. Customers like Notion and Rippling are already utilizing the SDK locally or in the cloud to build custom background agents that take bugs from tickets to merge-ready PRs and maintain self-healing codebases.
[Anthropic] · Source Anthropic published new safety research on “introspection adapters,” a mechanism that enables language models to self-report behaviors they picked up during training. By training a single adapter, fine-tuned models can be prompted to reliably expose hidden misalignments, backdoor vulnerabilities, and safeguard removals, offering a new method for auditing black-box capabilities.
[Mistral] · Source Mistral quietly shipped Mistral Medium 3.5. The new model is a dense 128-billion parameter system, providing a significant unannounced capability jump for their mid-tier offering.
[Tesla] · Source Tesla announced it will expand the rollout of its FSD V14 Lite software to international markets for vehicles equipped with Hardware 3 (HW3). While specific deployment dates depend on technical verification and regional regulatory approvals, this confirms HW3 owners will continue receiving modern Full Self-Driving stack updates.
Also Noted#
- [Waymo] (Source): The autonomous vehicle operator revealed it is now handling over 500,000 weekly rides across 11 operational cities.
- [Parag Agrawal / Stealth] (Source): Former Twitter CEO Parag Agrawal announced his new AI company raised a Series B led by Sequoia at a $2 billion valuation, with participation from Khosla Ventures and Index Ventures.
- [Hightouch] (Source): The customer data platform raised a $150M Series D led by Goldman Sachs and Bain Capital Ventures at a $2.75 billion valuation to aggressively expand its AI marketing agents.
- [Google Gemini] (Source): The Gemini chat interface can now natively generate and output standard file formats—including PDFs, Microsoft Word, and Excel—directly from a prompt.
- [IBM] (Source): IBM Granite released two small-footprint multilingual embedding models (97M and 311M parameters) built on ModernBERT that support over 200 languages and a 32K context window.