Claude docs 📃, Microsoft & Anthropic 🤝, MCP Registry 🌐

Claude can now generate and edit documents, spreadsheets, slides, and PDFs directly in the app. Turn prompts and data into downloadable files. ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With TLDR AI 2025-09-10 How Snowflake is adapting usage-based pricing to the AI era (Sponsor) Snowflake's usage-based pricing has been one key driver of its market leadership. In this webinar, you'll get an insider view of how the team managed its pricing and billing, and how they're preparing for the Al era of monetization. Join Ryan Campbell, Director of Product Finance at Snowflake, and Scott Woody, Metronome CEO, to learn how Snowflake aligns finance, product, GTM, and engineering around pricing decisions. On the agenda: Reducing friction and accelerating pricing decisions Applying rigor and experimentation to your monetization model Lessons from Snowflake on designing pricing for predictability, visibility, and control Join live: How Snowflake built its Monetization Operating Model 🚀 Headlines & Launches Claude Can Now Create and Edit Files (2 minute read) Claude can now generate and edit documents, spreadsheets, slides, and PDFs directly in the app, allowing users to turn prompts and data into downloadable files. Microsoft to lessen reliance on OpenAI by buying AI from rival Anthropic (1 minute read) Claude will be integrated into Office 365 apps, ending Microsoft's exclusive reliance on OpenAI. OpenAI has continued to put strain on its relationship with Microsoft by announcing competitive products, including an AI-powered LinkedIn rival and custom chips built by Broadcom. Nvidia unveils new GPU designed for long-context inference (1 minute read) Nvidia has announced a new GPU called the Rubin CPX designed for context windows larger than 1 million tokens. The GPU, meant to be used as part of a broader 'disaggregated inference' infrastructure approach, is optimized for the processing of large sequences of context. It performs better on long-context tasks like video generation and software development. The Rubin CPX will be available at the end of 2026. 🧠 Deep Dives & Analysis The whole point of OpenAI's Responses API is to help them hide reasoning traces (5 minute read) OpenAI's Responses API replaces the previous /chat/completions API for inference. The new API has a lot more features, but the main difference is that it is stateful. Users no longer have to pass the entire conversation history with each request, they just have to pass an ID representing the state of the conversation and the provider keeps models up to date. This allows OpenAI to keep its reasoning traces secret. The Training Imperative (6 minute read) Every serious AI company will eventually train its own models. The barrier to doing so is collapsing. Distillation, fine-tuning, and post-containing get easier every month. Soon, the only way to stay relevant will be to own your own models. Thoughts on Evals (14 minute read) Production monitoring reveals real issues that pre-deployment evals will inevitably miss, especially as AI products become more unpredictable and personalized. Evals are collections of already-known failure cases, but agents can and often do fail in ways that produce no error codes. 🧑‍💻 Engineering & Research Discover how Google, AWS, Databricks, and ServiceNow approach data strategy & AI at CData Foundations (Sponsor) Get your data ready for the future of analytics and AI at CData Foundations 2025! One free registration gets you two virtual summit days - Analytics on September 17 and AI on September 24. Join for expert insights, real-world success stories, and practical strategies to modernize your data architecture. Save your spot (free) ByteDance's Backward Reasoning (6 minute read) ByteDance's REER (REverse-Engineered Reasoning) is a paradigm that derives step-by-step reasoning from known good answers rather than building it forward via RL or imitation. Introducing the MCP Registry (4 minute read) The Model Context Protocol Registry provides a standardized way to distribute and discover MCP servers. A community-driven project, it allows organizations to create private enterprise registries while maintaining compatibility through shared API schemas and moderation guidelines. 🎁 Miscellaneous Gemini rolls out 'Tools' redesign on web as Nano Banana growth continues (3 minute read) Gemini Web has rolled out its Tools redesign. The prompt bar is now centered with a new 'Tools' menu that lets users access Deep Research, Veo, image creation, Canvas, Guided Learning, and Deep Think. Gemini now groups suggested prompts together for a cleaner look. Screenshots of the new interface are available in the article. The Gross Margin Debate in AI (8 minute read) AI companies face varied gross margins across sectors. Chips maintain around 70% margins, while cloud services see margins pressured by AI investments, estimated between 50-55%. Application-level margins range widely, with AI "Supernovas" starting at 25%, potentially negative, while others hit 60%, highlighting pricing strategies and diversified revenue models to improve margins over time. ⚡ Quick Links Warp Code: Go from prompt to production with Warp's benchmark-beating coding agent (Sponsor) Warp Code adds a built-in editor and code review to proven CLI and MCP tools, bringing everything together in one powerful app. Top-rated on Terminal-Bench and SWE-bench Verified. Download today OpenAI Launches $50M People-First AI Fund (6 minute read) OpenAI announced a $50 million fund to support nonprofits working on AI literacy, community-driven innovation, and equitable economic opportunity. Four scenarios for open vs frontier AI model competition (1 minute read) The main question is whether smaller models will continue to lag closed source models by 6-12 months, or if the gap will expand as capabilities advance. Nvidia CEO Says AI Will Actually Make Us Busier in the Future (5 minute read) Nvidia CEO Jensen Huang predicts AI and robotics will increase human workloads, countering Elon Musk's vision of more leisure time. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/0b6a6dc1/2 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of AI professionals and decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! If you have any comments or feedback, just respond to this email! Thanks for reading, Andrew Tan, Ali Aminian, Jacob Turner & Sahil Khoja Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR AI isn't for you, please unsubscribe.