From The New Stack · 40 stories
Runway Introduces Solaris, an AI Model for Real-Time Generative User Interfaces
Runway launched Solaris, its first Interface World Model, which generates user interfaces frame by frame in real time based on user interactions. This model aims to eliminate the need for translating designs into code by making the visual interface the application itself, allowing for direct interaction with generated scenes.
Meta launches Muse Code out of beta with new features and subscription tiers
Meta has officially launched its coding agent, Muse Code, out of beta, introducing three new subscription plans ranging from $5 to $50 per month. The launch includes new capabilities like inter-session messaging, Workflows for orchestrating agents, and a rewind feature. This move provides predictable monthly pricing for the coding agent, which is powered by Meta's Muse Spark 1.2 model.
Google Releases TimesFM-3, a New Time-Series Forecasting Model, Under Non-Commercial License
Google launched TimesFM-3, a 330-million-parameter time-series forecasting model trained on over a trillion data points, now available on Hugging Face with a non-commercial license. This model is Google's first natively pre-trained for multivariate time series and zero-shot generalization, outperforming previous models in benchmarks. Its non-commercial license restricts immediate business use, but it advances the field of time-series forecasting.
Amazon ECS Express Mode simplifies container deployment on AWS Fargate
Amazon Web Services (AWS) has launched ECS Express Mode, a new interface for Amazon Elastic Container Service (ECS) that simplifies the deployment of containerized applications. This mode automates the setup of load balancers, networking, IAM roles, and scaling policies, allowing developers to deploy an HTTPS service on Fargate with just a container image and two IAM roles.
LM Studio develops Auto Review and Shell Judge to secure AI coding agent commands
LM Studio introduced Auto Review and Shell Judge to enhance the security of commands executed by its Bionic AI coding agent. These tools analyze shell commands to prevent potentially dangerous operations, particularly when variables are involved, by parsing command structure and understanding tool-specific interpretations.
Google DeepMind introduces double-blind evaluation for AI models using confidential computing
Google DeepMind has developed a double-blind evaluation method for AI models, allowing testing without either the model provider or the evaluator seeing the other's proprietary data. This method addresses benchmark leakage concerns by using confidential computing to protect both model weights and test questions during evaluation. The approach aims to provide a more secure and unbiased way to assess AI model performance.
Simular's Sai Agent Achieves 73% Success on OSWorld 2.0 Benchmark
Simular's Sai agent scored 73% on the OSWorld 2.0 benchmark, outperforming OpenAI's GPT-5.6 Sol and Anthropic's Opus 5. This result indicates progress in AI agents handling complex, real-world professional tasks at a lower operational cost.
Perplexity introduces Portable Computer, separating AI reasoning from deterministic execution
Perplexity released Portable Computer, a local-first version of its Computer agent, which separates probabilistic AI reasoning from deterministic software execution. This architectural change allows for improved reliability and performance in AI agents by using code for control and policy enforcement, rather than relying solely on larger AI models for orchestration.
Perplexity's Computer AI agent now runs locally on high-end Nvidia hardware
Perplexity, in collaboration with Nvidia, released Portable Computer, a local version of its agentic AI assistant. This development allows users to run the AI agent on their own hardware, requiring specific Nvidia GPUs with at least 24GB VRAM or a DGX Spark desktop, which entails significant hardware costs.
MotherDuck Acquires Tower to Integrate AI-Powered Data Pipeline Technology
MotherDuck, a serverless data warehouse company, has acquired Tower, a data infrastructure startup whose technology was already powering MotherDuck's AI-built data pipelines. This acquisition is MotherDuck's first and aims to bring Tower's technology and team in-house as MotherDuck expands its focus on AI agents for data pipeline creation and operation.
JetBrains launches Junie Local for offline AI coding on developer machines
JetBrains released Junie Local, a free version of its AI coding agent that runs entirely offline on a developer's machine. This allows developers to use AI coding assistance without internet access, API costs, or cloud-hosted models, simplifying local AI setup by pre-configuring models and inference engines.
Malicious Pull Request Nearly Led to AI Coding Assistant Wiping User Systems
A malicious pull request submitted to Amazon's aws-toolkit-vscode repository nearly caused the Q Developer AI coding assistant to wipe the systems of its nearly one million users. A formatting error in the malicious prompt prevented the attack from executing, highlighting vulnerabilities in AI agent security and code review processes.
Spline V2 Rebuilds 3D Editor, Integrates External AI Coding Agents via MCP Server
Spline released V2, a complete rebuild of its 3D editor, introducing the Spline MCP Server which allows external coding agents like Claude Code and Google Antigravity to directly interact with and modify live 3D scenes. This integration enables AI models to make editable changes within Spline projects, bridging the gap between AI-driven generation and interactive 3D design workflows.
Google expands Antigravity AI coding agent to VS Code, JetBrains, and other IDEs
Google is integrating its Antigravity AI coding agent into popular Integrated Development Environments (IDEs) like Visual Studio Code, Visual Studio, JetBrains IDEs, and Zed through new extensions. This expansion allows developers to use Antigravity within their existing workflows, making the AI agent more accessible to enterprise engineering teams.
Canonical and University of Bristol research AI-driven C to Rust code translation
Canonical is collaborating with the University of Bristol to research automated translation of legacy C code into Rust, focusing on ensuring behavioral equivalence and safety. This initiative aims to determine the feasibility and trustworthiness of AI-generated Rust code for critical systems like AppArmor and snap-confine.
OpenAI's Codex integrates asynchronous messaging for developer interaction
OpenAI has merged `send_user_message_async` into the public Codex repository, allowing the AI coding agent to continue working on independent tasks while waiting for developer input. This change improves efficiency by preventing Codex from idling, enabling it to ask questions and proceed with other work concurrently.
IBM develops new modular cryogenic refrigerator for future quantum computers
IBM announced the development and initial cooling of two modules for a new cryogenic dilution refrigerator designed for its future fault-tolerant quantum computers. This advancement is crucial for scaling quantum systems, as it addresses the infrastructure challenges of housing and connecting multiple quantum processors in an ultra-cold environment.
OpenAI, AWS, GitHub, Microsoft, and Cursor back Agent Plugins 1.0.0 standard
OpenAI, AWS, Cursor, GitHub, and Microsoft have endorsed Agent Plugins 1.0.0, a new open standard for AI agent plugins initiated by Vercel. This standard provides a common, portable package format for reusable components, allowing AI agents to connect to external tools and services consistently across different clients and environments.
Databricks acquires Electric to integrate WASM-based Postgres and real-time sync for AI agents
Databricks has acquired Electric, the company behind PGlite and a real-time sync engine, to enhance database capabilities for agentic AI applications. This acquisition aims to provide individual Postgres databases for AI agents and facilitate real-time data synchronization, addressing new database requirements posed by agent workloads.
Amazon EKS Reduces Multi-Gigabyte Container Image Pull Times to Seconds
Amazon EKS has reduced the time it takes to pull multi-gigabyte container images from several minutes to seconds, particularly benefiting machine learning workloads. This improvement was achieved by optimizing the image pull pipeline to better utilize existing network bandwidth, storage throughput, and compute resources on accelerated instances.
Meta Engineers Train Internal AI Coding Tools by Fixing Code
Meta is requiring thousands of its software engineers to submit code corrections to its internal AI coding agent, MetaCode, to improve its performance. This strategy allows Meta to collect data on AI errors and human corrections during regular development, which will be used to post-train models like Watermelon and has already improved Muse Spark 1.1.
YugabyteDB Launches AMP for Serverless PostgreSQL to Manage AI Agent Database Sprawl
YugabyteDB introduced Agentic Multitenant Postgres (AMP), a new serverless PostgreSQL tier designed to manage the proliferation of databases created by AI agents. This solution addresses the scaling challenge of numerous small, isolated databases rather than a single large one, offering cost efficiency by scaling to zero when idle.
Cloudflare Open-Sources AI Tool Used by Astro to Clear GitHub Issue Backlog
Cloudflare has open-sourced "triagebot-action," an AI-powered GitHub Action that helped the Astro JavaScript framework reduce its open issues from over 200 to around 20. This tool automates the reproduction, diagnosis, and verification of bugs, offering a solution for open-source projects struggling with growing issue backlogs.
PortSwigger launches Burp AT public beta for agentic AI in pentesting
PortSwigger released the public beta of Burp AT, a new product integrating agentic AI into its Burp Suite platform for professional penetration testing. This development allows AI agents to perform investigative tasks within defined scopes, with human pentesters retaining control over judgment and conclusions.
Tines Launches 3B AI Platform, Shifting from No-Code to AI-Authored Workflows
Tines, an automation platform company, launched 3B, a new platform that uses AI to author enterprise workflows while executing them with conventional code. This move signals Tines' belief that the low-code/no-code paradigm has a "sell-by date" due to advancements in large language models capable of generating code.
Diagrid Catalyst 2.0 enables AI agents to resume from failure across multiple frameworks
Diagrid launched Catalyst 2.0, adding a durable execution and attestation layer for AI agents built with various frameworks, allowing them to resume from their last completed step after an interruption. This development matters because it addresses a key reliability issue for AI agents in production, preventing them from restarting entire processes after a failure.
Major Cloud Providers Now Offer Agent Sandboxes with Varied Implementations
AWS, Google Cloud, Microsoft Azure, and Cloudflare now all provide agent sandboxes for isolated code execution, with Google Cloud recently launching its Cloud Run sandboxes into public preview. This development means all four major cloud providers offer native isolated code execution, each utilizing different underlying isolation technologies and lifecycle models.
Nvidia releases JEPA-DNA, a genomic foundation model with latent-space prediction
Nvidia released JEPA-DNA, a genomic foundation model on Hugging Face that combines standard token-level DNA language modeling with a Joint Embedding Predictive Architecture (JEPA). This model predicts the functional representation of masked genomic segments in a latent space, moving beyond traditional token prediction to better understand the wider functional meaning of DNA sequences.
Cursor launches AI model router to optimize coding requests and reduce costs
Cursor, an AI coding tool, has released a new model router that automatically directs coding requests to the most suitable AI model based on complexity and cost. This development aims to simplify model selection for developers and reduce operational expenses, with early users reporting 30-50% savings.
Anthropic Acqui-Hires AI Startup Mendral Team to Strengthen Claude's Engineering Capabilities
Anthropic has acqui-hired the team behind AI startup Mendral to bolster Claude's software engineering capabilities. Mendral, known for developing AI agents that automate software development tasks, will wind down its existing product.
AWS EKS launches self-healing node feature for GPU instances
AWS has implemented a self-healing feature in its Amazon EKS for GPU nodes, allowing automatic detection and replacement of failed nodes. This enhancement aims to reduce downtime and manual intervention, streamlining operations for Kubernetes clusters.
Arm and Google Launch Axion Processors for Efficient Agentic AI Workloads
Arm and Google have introduced custom Axion processors optimized for running agentic AI workloads in cloud environments. These processors enhance efficiency by matching varying workload types to the most suitable compute resources, making operations less costly and more effective.
Amazon EKS shares lessons from scaling Kubernetes controllers
Amazon EKS has detailed challenges faced in scaling Kubernetes controllers, particularly with network traffic enforcement. The insights emphasize the importance of maintaining correctness in reconciling state as cluster sizes increase.
GoDaddy launches developer platform integrating domain management into development workflows
GoDaddy unveiled its Developer Platform, allowing domain management directly within development environments. This change addresses the growing need for streamlined domain management in software development workflows, competing with major cloud DNS services.
Cursor Launches Origin, a Code Hosting and Collaboration Platform
Cursor has launched Origin, a new code hosting and collaboration platform available in early beta for paid plan users. Origin provides repositories, pull requests, code browsing, and GitHub synchronization, aiming to integrate code management with Cursor's AI agent capabilities.
Silent AI Model Updates Undermine Agent Monitoring and Reliability
An AI agent's underlying model was silently updated, leading to a significant change in behavior that went undetected by standard success metrics. This highlights a critical gap in monitoring practices for AI systems, where unpinned model aliases can cause unlogged deployments and unexpected performance shifts. The incident underscores the need for more robust diagnostic metrics that can detect intentional system failures, rather than just overall functionality.
Ox Alpha Reasoning Model Released by Anonymous Third-Party Provider via OpenRouter
A new reasoning model named Ox Alpha, designed for coding and agentic work, has been released by an anonymous third-party provider through OpenRouter. The model is free, offers a 1M context window, and its prompts and completions are not used for training.
AlloyDB Omni Web UI Reaches General Availability for Cluster Observability
Google has announced the General Availability of the AlloyDB Omni Web UI for Red Hat and containers. This read-only console provides direct visibility into hardware telemetry, query performance, and configuration flags for AlloyDB Omni clusters. The UI aims to simplify troubleshooting and monitoring of PostgreSQL databases across on-premises, edge, and multi-cloud environments without relying solely on command-line tools.
Unused Access Keys Pose Security Risk Due to Lack of Lifecycle Management
Unused and forgotten access keys for non-human identities, such as those for defunct projects or departed personnel, create significant security vulnerabilities because they do not expire automatically. This issue highlights a critical gap in identity and access management, as these lingering credentials can be exploited by attackers much faster than traditional audit cycles can detect them.
Mandated Internal Platforms Deliver Goals More Often, Despite Industry Shift to Optional Adoption
Research indicates that internal platforms with mandatory adoption are more than twice as likely to achieve their goals compared to optional platforms. This finding contradicts prevailing industry advice that advocates for self-service and optional platform adoption, even as the trend shifts towards optional platforms.