Sep 11 · AI Daily

SpaceX CFO Bret Johnsen stated that no competitor will reach the level of Starship in the near term.

Reusable rockets give SpaceX an advantage, allowing them to drive AI compute costs well below current levels. Having achieved the top spot in rockets and Starlink, SpaceX is now bringing that same drive to building the best AI products.

Cursor now has the most gorgeous sidebar of any AI coding tool. I don't have any normal chats anymore, everything is a Project. Organizing my work has never been easier.

I predict all other coding apps will follow suit soon.

Google officially released a standalone Gemini client for Windows 10/11. Once installed, it can be summoned from any interface using the Alt + Space shortcut, which can be customized. The app interface is largely the same as the web version, supporting image and video generation, calling connected apps, and providing a standalone settings menu to manage microphone, camera, and location permissions, view usage stats and crash reports, and set launch at startup.

Slack launched Slackforce Surfaces. After users describe what they need to Slackbot, AI gathers information from relevant conversations and connected apps like Google Drive and Salesforce to generate interactive reports, polls, dashboards, presentations, and mini-websites right inside the chat.

The author argues that AI makes copying existing products fast and easy, meaning the true moat is no longer features or price, but the ability to continuously originate. The article looks back at the creative explosion of the Flash era and its objective flaws, noting that roughly 35% of new websites today are AI-generated and generally mediocre because users skipped key creative decisions. It suggests using AI to accelerate experimentation rather than generating copies, questioning product premises, tolerating numerous failed attempts, and making the invention of new solutions a habit.

Cursor released Projects (beta), letting users handle large jobs like feature development, migrations, and ongoing maintenance through a coordinator agent. The coordinator itself does not write code, but instead dispatches thousands of sub-agents to run in parallel.

Krea AI launched Flux Video Edit, which lets users replace anything in a video with high precision and super-fast speed. The feature is now open for testing.

Perplexity is developing a new Automations feature for Perplexity Computer.

Users will be able to schedule workflows on a schedule or invoke them based on specific triggers.

"Automate everything with Computer -- run workflows on a custom schedule, or when specific conditions are met, and automatically get results."

OpenAI released ChatGPT for Financial Services, available to eligible financial institutions, allowing various financial data to be connected to ChatGPT conversations. Powered by GPT-6 Astra under the ChatGPT Work mode, it scored 69.9 on the OfficeQA Pro benchmark, beating GPT-5.6 Sol by 9% and Claude Fable 5.1 by 7%.

Clean up your sidebar! One suggestion I have is that you can think of Projects in Cursor as a special folder for chats! A coordinator agent oversees all agents within the project, and you can also drag existing chats from your sidebar into the project. If you have a lot of related chats scattered all over the place, consider moving them into a project! Even completed agents can be dragged in—they make great context sources for the rest of the project.

Powered by gpt-live!

Voice is the most magical way to interact with AI.

There is something beautiful about a future where bringing an idea to life starts with saying it out loud.

So excited for Devin voice!

💚

Nemotron 3 Embed 8B ranked first in overall nDCG@10 on the Q2D-Web benchmark, covering 190 million web documents and nearly 70,000 agent-rewritten queries across 10 languages.

Former Google CEO Eric Schmidt stated that the true limit for AI is funding rather than energy. At roughly $50 billion per gigawatt, 10 gigawatts would require about $500 billion, and very few companies and countries can afford trillion-dollar capital investments. He said China has the capability to do so, the US can borrow the money thanks to its capital markets, and Europe will struggle to do so. The full video is from the Special Competitive Studies Project YouTube channel.

SpaceX plans to launch its first batch of orbital AI compute satellites in 2027, followed by deploying "massive compute" in space starting in 2028.

CFO Bret Johnsen said almost all AI leaders believe orbital compute is the future, with its advantages lying in clean energy and fewer restrictions than Earth.

Google proposed ToolGrad, a data generation framework that first has LLMs generate tool-calling chains and then back-annotates user queries, published at ACL 2026.

Redis released LangCache, a fully managed semantic caching service now available in public preview on Redis Cloud, accessible via REST API and Python and JavaScript SDKs.

The author shared that they used DeepSeek as their primary model for daily knowledge work today, noting its exceptional cost-performance ratio, especially excelling at browser navigation, available to try on Energy. The content cites DeepSeek-V4.1-Flash released by @deepseek_ai: the smallest model in the new architecture family, featuring native visual understanding, designed for stronger capabilities, faster inference, higher throughput, and scalability to larger models. Screenshots show it scoring 74.2 on DeepSWE v1.1, 88.1 on CyberGym, 54.8 on Automation-Bench, and 30.0 on Terminal-Bench 3.0.

Pocket FM used AI-driven content production to double its revenue run rate to $500M ARR within a year, adding $250M net last year while remaining EBITDA positive. The company reportedly cut the time to produce 100 hours of audio from a year down to one day, lowering costs by about 80x; it develops its own writing and TTS models.

Elvis Saravia recommended paying attention to the "business world models" proposed by Lightfld, suggesting that as the world becomes increasingly agentic, different forms of world models will emerge. Starting from CRM, Lightfld models business complexity using world models to make it easier for AI agents to extract value, with the potential to expand into adjacent domains on appropriate agent infrastructure.

OpenAI announced the resolution of a long-standing open problem regarding the Navier-Stokes equations, releasing a Lean 4 formal proof alongside the human-readable proof. Citing estimates, author ibobev noted that formalizing its 166-page paper using old standards would have taken about 132,800 person-hours, whereas OpenAI completed the Lean verification in 17 hours, cutting costs by about four orders of magnitude. The author believes formal verification can also be used for security policies, smart contracts, and critical algorithm checks.