Brief

Google AI has introduced Vibe Coding XR, a new workflow designed to simplify the creation of interactive XR experiences. This system leverages Gemini's capabilities with the open-source XR Blocks framework to translate natural language prompts into functional, physics-aware WebXR applications for Android XR devices. The goal is to accelerate prototyping by allowing creators to quickly test intelligent spatial experiences without extensive coding knowledge, with applications deployable in under 60 seconds. Google plans to demonstrate Vibe Coding XR at ACM CHI 2026. AI

OpenAI has launched "Instant Checkout" within ChatGPT, enabling users to purchase products directly from merchants without leaving the chat interface. This feature is powered by the newly released Agentic Commerce Protocol, an open standard co-developed with Stripe. Initially available for U.S. users to buy from Etsy and soon Shopify merchants, the protocol aims to facilitate seamless AI-driven commerce by allowing AI agents, people, and businesses to collaborate on purchases. OpenAI is open-sourcing the protocol to encourage broader adoption and integration by developers and merchants. AI

Chris Aquino, a software engineer at Thunderbird, discussed the development of a privacy-focused AI assistant for the email client. The project involved careful consideration of design choices such as remote but confidential inference and local encryption. Aquino detailed the evolution of the assistant's architecture, moving from a single large language model approach to a more distributed system utilizing multiple specialized models for different tasks. AI

Hugging Face has released a guide detailing techniques to optimize the performance of large language models using the Transformers library. The blog post, inspired by OpenAI's open-source contributions, focuses on practical methods for accelerating inference and training. It covers strategies such as quantization, efficient attention mechanisms, and optimized kernels to help developers achieve faster results with their models. AI

OpenAI has launched SafetyKit, a platform that utilizes its most advanced models, including GPT-5 and GPT-4.1, to build multimodal AI agents for detecting fraud and prohibited activities. These agents can process text, images, and financial transactions with over 95% accuracy, handling billions of tokens daily. The system aims to enhance risk assessment, compliance, and safety operations for various platforms by automating content review and reducing human moderator exposure to harmful material. AI

Bitrig, a new iOS app developed by Kyle, Jacob, and Tim, allows users to create native Swift applications directly on their iPhones through AI-powered chat. The app utilizes Claude Sonnet 4.0 and a custom Swift interpreter to enable on-device app development, a feat previously requiring Xcode on a Mac. Users can preview their creations instantly, share them via URL, and even connect a paid developer account to compile and distribute apps through App Store Connect. AI

OpenAI has updated its Codex tool, enhancing its integration with VSCode. This update allows for better synchronization of tasks with the Codex Cloud platform. The changes aim to improve the developer experience and workflow when using AI-assisted coding tools. AI

Blue J, a tax research company founded by law professors, has successfully scaled its AI-powered system to three countries and over 3,000 firms. Their solution, built on OpenAI's GPT-4.1 model and a Retrieval-Augmented Generation (RAG) system, provides expert-grade tax answers with citations in seconds. The system incorporates user feedback through a 'disagree' button to continuously improve accuracy and address edge cases, demonstrating a rapid and effective application of advanced AI in a complex, regulated field. AI

Anthropic's Claude 3.5 Sonnet model can now generate images through a partnership with Hugging Face. This integration allows users to create visuals directly within the Claude interface by describing them in text. The feature leverages Hugging Face's diffusion models to render the images, expanding Claude's capabilities beyond text-based tasks. AI

Basis, a startup founded in 2023, has developed AI agents that leverage OpenAI's latest models, including GPT-5, to automate tasks for accounting firms. These agents help reduce time spent on routine work by up to 30%, allowing accountants to focus on advisory services and business growth. The system uses a multi-agent architecture, with GPT-5 acting as a supervising agent for complex workflows and GPT-4.1 for faster interactions, ensuring transparency and control for users. AI

Figma is integrating AI across its design platform to enhance creativity and streamline workflows for both designers and developers. The company's Head of AI Products, David Kossnick, highlighted tools like Figma Make, which generates code from prompts, and AI agents that handle repetitive tasks. These AI features aim to empower users by reducing technical barriers and accelerating the ideation and prototyping process, while still emphasizing human judgment and craft as essential to the design process. AI

Phind.design has launched a new AI-powered image editor and design tool. The platform leverages OpenAI's GPT-4o model, alongside custom models, to assist users in their creative processes. This integration aims to provide advanced capabilities for image manipulation and design tasks. AI

Invideo AI has launched a new platform that leverages multiple OpenAI models, including GPT-4.1, image generation, and text-to-speech, to significantly accelerate video production. This tool allows users to create professional-quality videos from simple text prompts, with AI agents handling scripting, visuals, narration, and optimization for different platforms. The integration aims to reduce video creation time by up to tenfold and has already helped users double their revenue, with the platform supporting over 50 million users and 7 million videos monthly. AI

Hugging Face has introduced ScreenEnv, a new framework designed to enable the deployment of full-stack desktop agents. This tool allows developers to create agents capable of interacting with graphical user interfaces (GUIs) across various operating systems. ScreenEnv aims to bridge the gap between AI models and real-world desktop applications, facilitating more complex automation tasks. AI

Retell AI has launched a no-code platform for creating customizable voice agents, leveraging OpenAI's GPT-4o model. These AI-powered agents can handle customer service tasks like scheduling appointments and resolving issues, reportedly reducing call handling costs by up to 80%. The integration of GPT-4o allows for more natural, dynamic conversations and simplifies the development process, enabling businesses to deploy agents in days rather than weeks. AI

Hugging Face has integrated its Transformers library with SGLang, an open-source language model serving system. This integration allows developers to leverage Hugging Face's extensive model hub directly within SGLang for more efficient model deployment and inference. The collaboration aims to simplify the process of serving large language models, making advanced AI capabilities more accessible to a wider range of users and applications. AI

OpenAI has partnered with Mattel, the toy and entertainment company, to integrate AI into its iconic brands. This collaboration aims to create new ways for fans to interact with Mattel's products and experiences. Mattel will also utilize ChatGPT Enterprise to boost its internal operations, including product development and creative processes. AI

Together AI has introduced YAQA, a novel post-training quantization technique for large language models. This method aims to preserve the original model's outputs more effectively than existing algorithms by directly minimizing KL divergence. YAQA achieves this by approximating the Hessian of the KL divergence, leading to over 30% reduction in KL divergence compared to current rounding methods and improved performance on downstream tasks. AI

The AI Engineer World's Fair (AIE) featured discussions and workshops on Google's Gemini models and real-time voice AI. Speakers like Logan Kilpatrick, Shrestha Basu Mallick, and Kwindla Hultman Kramer presented on topics such as the progress of Gemini, building real-time AI workflows with the Gemini Live API and Pipecat, and developing voice agents. The event highlighted advancements and future directions for Gemini, particularly in real-time applications. AI

Wix has launched a new AI website builder that allows users to create complete websites simply by conversing with the AI. This tool, powered by OpenAI's GPT-4o and other models, handles everything from layout and image selection to content generation and business app integration. Users describe their needs, and the AI constructs a functional website, significantly reducing the time and technical expertise required for site creation. AI

Mistral AI has launched an Agents API, signaling a move towards an "LLM Operating System" for 2025. This new API allows developers to build and deploy autonomous AI agents that can perform complex tasks. The company aims to facilitate the creation of more sophisticated AI applications by enabling agents to interact with each other and external tools. AI

CodeRabbit has enhanced its AI-powered code review tool by integrating OpenAI's o3, o4-mini, and GPT-4.1 models. This integration allows the tool to analyze code with greater accuracy and speed, reportedly enabling developers to ship code four times faster and reduce production bugs by half. The system enriches code diffs with contextual information from repositories and conversations before performing recursive reviews using multiple OpenAI models tailored to specific team standards. AI

Lowe's has integrated OpenAI's GPT-4o technology into two new applications to enhance both customer and associate experiences. The Mylow virtual advisor on their website now offers conversational guidance for home improvement projects, while the Mylow Companion mobile app provides in-store associates with quick, expert answers about products and installations. These tools aim to equip associates with extensive product knowledge, enabling them to better assist customers with complex projects and purchasing decisions. AI

Lowe's has been integrating AI into its operations for nearly a decade, deploying over 50 machine learning models for tasks like pricing and supply chain management. The company is now expanding its use of generative AI, collaborating with OpenAI on tools like Mylow and Mylow Companion. These AI-powered applications aim to assist customers with home improvement projects and provide store associates with instant access to product and project information, enhancing both customer and employee experiences. AI

Andrej Karpathy has developed MenuGen, a web application that generates images for menu items based on a photo of the menu. This tool aims to help users understand unfamiliar dishes by providing visual context. Karpathy utilized AI tools, specifically Cursor and Claude 3.7, for the entire development process, noting that 100% of the code was generated by AI. He encountered challenges with API integrations, including hallucinations from Claude and rate limiting issues with OpenAI and Replicate APIs. AI

Google DeepMind has enhanced its Music AI Sandbox with new features and expanded access for users. The platform now includes Lyria 2, a new music generation model, and has incorporated feedback from musicians and industry partners. This update aims to provide more sophisticated tools for music creation and collaboration. AI

OpenAI has made its GPT-Image-1 model accessible via API. This model is distinct from the recently announced GPT-4o. The availability through the API allows developers to integrate image generation capabilities into their applications. AI

Speak, a language learning application, is leveraging OpenAI's advanced AI capabilities to create a personalized and highly interactive tutoring experience. The company, which began in 2016, has evolved significantly with the advancements in speech recognition and large language models, enabling features like real-time feedback and conversational role-playing. Speak's strategy involved an early focus on the South Korean market to prove its AI-native model before expanding globally, and it is now investing in AI-generated curriculum to personalize learning paths across various domains. AI

Together AI has launched a new fine-tuning platform that allows users to continuously improve open-weight language models. The platform now supports preference optimization and continued training, enabling models to adapt based on user feedback and new data. A new web UI simplifies the process, allowing developers to manage datasets, specify parameters, and monitor experiments directly from their browser. AI

Canva is integrating more advanced AI capabilities into its design platform, moving beyond simple tools like background removal to assist with entire creative workflows. The company is leveraging partnerships with AI leaders like OpenAI, alongside its own research, to enable users to generate and refine designs, images, and text. This expansion aims to make the creative process more accessible and efficient, allowing for human-AI collaboration from initial idea to final product. AI

OpenAI has reportedly adopted the "Model Coordination Protocol" (MCP), a system designed to manage and coordinate the development and deployment of AI models. This move suggests a structured approach by OpenAI to handle the complexities of its rapidly advancing AI research and product releases. The adoption of MCP could indicate a new phase in how large AI organizations manage their internal model pipelines and external releases. AI

Hebbia has launched Matrix, a multi-agent AI platform designed to automate 90% of finance and legal work. The platform orchestrates multiple AI agents, including OpenAI's o3-mini, o1, and GPT-4o, to process vast amounts of offline data and perform complex tasks. Matrix achieves 92% accuracy on deep research tasks, significantly outperforming traditional RAG methods, and offers substantial time and cost savings for professionals in investment banking, private credit, private equity, and law firms. AI

Booking.com has partnered with OpenAI to enhance its travel planning services by integrating large language models into its platform. This collaboration aims to create a more personalized and intuitive travel discovery experience for users, moving beyond traditional search filters. The company developed an AI Trip Planner that can understand natural language prompts to suggest destinations and build itineraries, leveraging both structured and unstructured data for tailored recommendations. AI

EliseAI, a company focused on the housing and healthcare sectors, has leveraged advancements in AI, particularly conversational models like GPT-4 and Whisper, to enhance operational efficiency. Initially using models such as BERT, EliseAI has progressively integrated newer technologies to automate and streamline existing workflows, making them more accessible to industries less familiar with advanced technology. The company measures its success by its ability to positively impact key business metrics for its clients, such as occupancy rates and maintenance resolution times. AI

Nubank, a major digital financial services platform serving over 114 million customers, has partnered with OpenAI to enhance customer experience and internal operations. The company is utilizing OpenAI's GPT-4o and GPT-4o mini models to power an AI-powered assistant that resolves 55% of Tier 1 customer inquiries, reducing chat response times by 70%. Additionally, Nubank has implemented an internal enterprise search tool for employees and a call center copilot to assist human agents, improving efficiency and customer satisfaction. AI

Anthropic engineer David Hershey has developed a system allowing their Claude 3.7 model to play the video game Pokémon Red. The model uses a custom harness to perceive the game's visuals, process information, and make decisions. Claude 3.7 has demonstrated progress in the game, notably escaping Mt. Moon. AI

Mercari has integrated OpenAI's GPT-4o mini into its platform to enhance the online marketplace experience for sellers. The AI Assistant optimizes product listings by analyzing them against best practices and suggesting improvements, leading to a significant increase in average sales per user. Additionally, a new AI Listing Support feature allows users to upload photos and have the AI automatically generate titles, descriptions, and category suggestions, drastically reducing listing time and effort. AI

Endex, an AI platform for financial firms, is developing an AI Analyst that leverages OpenAI's latest reasoning models, including GPT-4o, o1-mini, o1-preview, and o3-mini. This AI Analyst is designed to retrieve, synthesize, and reason through complex financial data, aiming to augment the workflows of investment professionals. By integrating internal data, public disclosures, and financial sources, Endex's agents can perform tasks like precedent transaction overviews, earnings summaries, and due diligence, with a focus on precision and structured reasoning. AI

Hugging Face has introduced Remote VAEs, a new method for efficient decoding in their Inference Endpoints. This approach allows for the offloading of the VAE decoding process to a separate, specialized service. This separation aims to improve inference speed and reduce the computational load on the main model. AI

Fireworks.ai has joined the Hugging Face Hub, making its suite of AI models accessible to a wider community. This integration allows developers and researchers to easily discover, use, and fine-tune Fireworks.ai's offerings directly through the Hugging Face platform. The partnership aims to foster collaboration and accelerate the development and deployment of advanced AI technologies. AI

OpenAI has introduced Operator, a new AI agent designed to perform tasks on behalf of users. This marks OpenAI's first foray into agent technology, aiming to automate complex actions. Operator is expected to enable more sophisticated interactions and task completion within various applications. AI

Hugging Face has integrated support for Visual Language Models (VLMs) into its smolagents framework. This enhancement allows agents to process and understand visual information alongside text. The update aims to enable more sophisticated agent capabilities by combining multimodal understanding with agentic reasoning. AI

OpenAI has launched Operator, a new AI agent designed to perform web-based tasks by interacting with websites through its own browser. This agent, powered by a new model called Computer-Using Agent (CUA), can fill forms, order groceries, and more by typing, clicking, and scrolling. Initially available to Pro users in the U.S. as a research preview, Operator is being integrated into ChatGPT and will expand to other user tiers. The agent has demonstrated state-of-the-art performance on web interaction benchmarks. AI

Bertelsmann, a global media and education company, is integrating OpenAI's technology across its various brands worldwide. This partnership will involve a significant deployment of ChatGPT Enterprise, enabling employees globally to leverage AI for product development and service enhancement. The collaboration aims to boost productivity in creative processes and daily workflows, with specific applications planned for journalism, marketing, and personalized content recommendations on streaming platforms. AI

NVIDIA has released a new tool called LogitsProcessorZoo, integrated into Hugging Face's Transformers library. This tool allows developers to control and modify the output of large language models during the generation process. By manipulating logits, developers can enforce specific constraints or styles on the model's responses, enhancing control over AI-generated text. AI

OpenAI has updated its Voice Mode for ChatGPT to include visual input capabilities, allowing users to show the AI objects and receive information about them. This feature, which was previously demonstrated by Google's Gemini, enables ChatGPT to analyze images and provide contextually relevant responses. The update aims to enhance the interactivity and utility of the voice assistant by integrating visual understanding. AI

Smol AI has released its o1 API, enabling developers to integrate its models into applications. The update also brings 4o and 4o-mini models to the Realtime API, supporting WebRTC for enhanced communication. Additionally, Smol AI has introduced Direct Preference Optimization (DPO) finetuning, a method for improving model performance based on user preferences. AI

Sidekick has launched as an AI-powered commerce assistant specifically designed for Shopify merchants. This tool integrates with a merchant's business data, including products, orders, and customer information. It leverages generative AI and large language models to provide expert assistance within the Shopify ecosystem. AI

OpenAI has officially launched ChatGPT Canvas, a new feature that allows users to generate images directly within the ChatGPT interface. This integration aims to streamline the creative process by enabling users to describe their desired visuals and have them produced without leaving the chat environment. The feature is now generally available, marking a significant step in making AI-powered image generation more accessible and user-friendly. AI

Replit has launched its AI coding Assistant and Agent tools, designed to enable users to create software through natural language conversations. The Agent assists in building applications from scratch, even from visual designs, while the Assistant helps refine existing projects. Replit is implementing a new 'checkpoint' billing model for these tools, offering monthly credits for a set number of Agent checkpoints and Assistant edits, with pay-as-you-go options available beyond those limits. AI