TLDR AI 2025-12-11

As services multiply, retries, cron jobs, and glue code can make orchestration brittle and hard to operate. Netflix Conductor was designed to solve this, but it's tricky to run at enterprise scale and security.

This whitepaper explains how the creators of Netflix Conductor built Orkes Conductor -- the control tower for microservices that adds enterprise capabilities to the proven workflow engine. Read it for a deep dive into Orkes, including:

• Delivering durable event driven workflows with clear SLAs, and automatic recovery.
• Scaling to millions of tasks per second with up to 99.99% uptime
• Enabling agentic orchestration that pulls humans in at the right moment

Read the whitepaper

Google DeepMind will open a research lab for discovering new materials in the UK as part of Alphabet's newly announced partnership with the British government. The agreement involves Alphabet tailoring several of its AI models for scientists, teachers, and public employees in the UK. The new lab will research materials that could improve medical imaging, solar panels, and chips. Using AI to unearth novel materials could drastically reduce costs and time.

Shopify merchants' products can now be discovered on AI platforms like ChatGPT, Perplexity, and Microsoft Copilot. Agentic storefronts allow merchants to sell their products across popular API chatbots. Agents can answer customers' questions while giving merchants insight into search trends and topics customers are asking about. Shopify plans to make every Shopify store agent-ready by default.

Meta is developing a closed model codenamed Avocado, expected next spring, after years of loudly supporting open-source. The new model is trained on the outputs (or “distilled”) from rival models, including Google's Gemma, OpenAI's gpt-oss, and Alibaba's Qwen. New Chief AI Officer Alexandr Wang is reportedly chafing under Zuckerberg's “micromanagement”. Long-time AI head Yann LeCun left last month after being sidelined under the new org structure.

No vector databases, no RAG over conversation history. The system uses four layers: ephemeral session metadata (device, location, and usage patterns), long-term facts stored via a dedicated tool, lightweight summaries of recent messages, and a sliding window of the current conversation. OpenAI prioritizes speed and token efficiency over detailed historical context.

No one can definitively say whether we are currently in a bubble or not. No one should go all-in without acknowledging that they face the risk of ruin if things go badly. At the same time, no one should stay all-out and risk missing out on one of the great technological steps forward. A moderate position seems like the best approach. Intelligent investment requires sober, insightful judgment and skillful implementation.

Tangle is a visual experimentation platform that lets you build ML pipelines using drag-and-drop, without writing orchestration code. Intelligent content-based caching means unchanged steps never recompute, and cached results are shared across your whole team. Works with any language and framework, including Python, R, Go, Java, or Shell. Try Tangle (open source)

Designing fast search engines is a subtle optimization problem. It takes effort to select the right algorithm and then tune it to get the best out of the CPU. Agents write longer queries than humans, so it is becoming increasingly important for text search to scale well with the number of terms. For text search, this means the cursor has shifted more and more from WAND to MAXSCORE.

Agents are becoming the dominant paradigm for real-world AI applications. However, practitioners still rely on heuristics rather than principled design choices. This study attempts to derive quantitative scaling principles for agent systems.

Claude Code CLI now has async subagents, instant compact, customer session names, and usage stats. There are also new keyboard shortcuts on the /resume screen. Run 'claude update' to get all of the new features.

Cursor has introduced Debug Mode, a new agent loop that incorporates runtime logs and human interaction to fix stubborn bugs. Inspired by expert debugging workflows, the mode generates multiple failure hypotheses, instruments code with logs, and loops in the user to reproduce errors and verify fixes.

Gemini is aimed at large organizations, not individual developers. It is designed for maximum compliance with government regulations, not for efficiency or productivity. Google doesn't want your business unless you're an organization that employs tens of thousands of people. OpenAI and Anthropic are much smaller and more nimble. They are able to make the process of setting up a developer account quick and easy.

Advances in AI world models like Marble and Genie 3 generate explorable 3D environments from text, enabling creators to build interactive worlds and game mechanics with natural language. These systems open the door to generative multiverses where players act as co-creators and participate in emerging digital economies built on interoperable assets. The shift promises new storytelling formats and valuable training grounds for agents and robotics.

Nvidia-backed Starcloud successfully trained the first AI model in space, using an Nvidia H100 GPU to run Google's Gemma LLM on their Starcloud-1 satellite. These orbital data centers promise significant energy savings compared to Earth-based facilities, addressing the growing power demands and environmental impact of terrestrial data centers. Starcloud plans a 5-gigawatt orbital data center to capture constant solar energy, with implications for both commercial AI workloads and real-time data analysis for military and emergency response operations.

Sourcegraph launched Amp 9 months ago as an experimental coding agent. It quickly grew to become one of the leading coding agents. Now, Amp is its own company operating at the frontier of software development.Read more on the new Amp website >

Adobe has launched Photoshop, Adobe Express, and Acrobat within ChatGPT, allowing 800 million users to create and edit by simply describing their goals.

Google is expanding features like Preferred Sources in Search, allowing users to prioritize their favorite outlets in Top Stories results.

The updated Flash and Pro TTS previews improve expressivity, context-aware pacing, and multi-speaker consistency across 24 languages.

Google has released full-managed, remote MCP servers, providing a unified layer across all Google and Google Cloud services.