TLDR AI 2025-07-11

Search has long been a the most important piece of the online shopping experience — and in recent years, AI has turned that experience on its head.

Generative AI is changing how developers approach the architecture and capabilities of user-facing systems like search, and how data are indexed and retrieved in the back-end.

Read How generative AI affects search: a full stack story to learn how you can align your stack with the latest developments and sharpen your competitive edge. Get the ebook

Mistral AI and All Hands AI released Devstral Medium and upgraded Devstral Small 1.1, optimized for agentic coding tasks. Devstral Small 1.1 is open-source under Apache 2.0 and outperforms all other open models for code agents.

Amazon is considering another multibillion-dollar investment in AI firm Anthropic to strengthen their strategic partnership. It already committed $8 billion in November last year. Amazon is one of Anthropic's largest shareholders. Google, another shareholder, has invested over $3 billion.

AWS will launch an AI agent marketplace next week at the AWS Summit in New York City, with Anthropic as a partner. The platform will enable startups to offer AI agents directly to AWS customers, potentially boosting Anthropic's reach and revenue. This move follows similar initiatives by Google Cloud and Microsoft, aiming to streamline AI agent distribution.

The success of a coding app depends on context engineering and software architecture, not raw model capabilities. Both platforms share four core components: typed prompts with test-driven development, MCP servers for sandboxed execution, agent loops for state management, and real-time frontend coordination.

There is widespread enthusiasm among Google software engineers about how AI is helping write code. This blog post looks at the company's newest AI-powered improvements and presents a method for building AI products that deliver value. Google monitors developer productivity and satisfaction carefully - the improvements covered in the post directly impact these. Google prioritizes by technical feasibility and impact, iterates quickly and extracts lessons learned, and extensively monitors productivity and satisfaction metrics to measure effectiveness.

A randomized controlled trial of 16 experienced open-source developers found AI tools decreased task completion time by 19%, despite developers self-reporting a 24% speedup.

Want to stay ahead of the AI curve? At GitHub Universe — from October 28-29 in San Francisco — you'll explore the newest AI agents, automations, and tools that can make your work more efficient and fun. You'll also get a chance to connect with fellow devs and leaders you admire. For a limited time, save $400 on your ticket with Early Bird pricing. Grab yours today.

Google's T5Gemma is a suite of encoder-decoder LLMs adapted from decoder-only Gemma 2 models. Designed for tasks like summarization and translation, T5Gemma includes pretrained and instruction-tuned variants in sizes ranging from 2B to XL.

asyncmcp is a regular MCP that works over queues. A lot of context is not always readily available, and it takes time for applications to process it. asyncmcp uses an async transport layer, so it doesn't have to immediately respond to any requests - it chooses to direct them to internal queues for processing and the client doesn't have to stick around for the response.

The Gemini API now has an asynchronous endpoint designed for high-throughput, non-latency-critical workloads. The Gemini API Batch Mode allows users to submit large jobs, offload the scheduling and processing, and retrieve results within 24 hours, with a 50% discount compared to the synchronous APIs. It is the perfect tool for when users have data ready upfront and don't need an immediate response.

The first of its kind, this architecture allows data contributors to activate or deactivate their contributions in real-time without retraining the entire model, enabling instant modification. The system trains separate expert modules on private datasets using a frozen public model as an anchor point, then merges them via domain-informed routing without requiring joint training.

Google has added speech generation to Flow, its AI video tool, allowing users to transform images into talking clips using Veo 3. Alongside this, Flow and Google AI Ultra are now available in over 140 countries, expanding access to advanced video and AI features worldwide.

Before companies can get to a future where real work happens through AI agents that interact with both data and people, companies have to solve the action interface for human operators. The action interface matters because most operators aren't prompt engineers - their job is to keep the world moving. Operators need a clean interface that lets them react to and confirm actions and a way to proactively ask the system to take actions. This article discusses a new paradigm for building interfaces that creates a seamless experience for operators.

You'll probably like our flagship newsletter. It's all about tech, science, and programming.

Same quick format. Still free.

Subscribe now.

gemini-beta-3.0-pro was just referenced in the latest Gemini-CLI commit.

Elon Musk confirmed that xAI's chatbot Grok will be integrated into Tesla vehicles "next week at the latest."