From The New Stack · 40 stories
Perplexity's Personal Computer AI Agent Now Available on Windows
Perplexity has expanded its Personal Computer feature, an agentic AI designed for multi-step tasks, to Windows 10 and 11. This feature, previously Mac-exclusive, allows the AI to access local files, native applications, and cloud services like OneDrive and Google Drive, requiring a paid Perplexity subscription.
OpenAI develops Pando, a new Git backend, to handle increased monorepo commits
OpenAI developed Pando, a new Git backend for its monorepo, to address performance bottlenecks caused by a significant increase in commits from developers and AI agents. Pando reduced push latency by nearly five times and processes close to a petabyte of Git data daily, while maintaining compatibility with the existing Git ecosystem.
Git Worktrees Lack Isolation for AI Agents, Creating Security and Infrastructure Challenges
Git worktrees, often used for parallel execution of AI coding agents, do not provide true isolation, allowing agents to manipulate repository state or install malicious hooks outside their designated worktree. This lack of isolation poses security risks and creates bottlenecks in existing runtime infrastructure, as multiple agents generate changes in parallel, overwhelming shared resources like CI queues and staging environments.
OpenSearch Project Shows Decreased AWS Contribution and Increased Self-Managed Deployments
Five years after its creation by AWS, the OpenSearch project is demonstrating reduced reliance on its original backer. AWS's share of contributions has decreased, and self-managed deployments of OpenSearch are growing faster than Amazon's managed service, indicating a shift towards broader community involvement and adoption.
Amazon CEO Predicts 6-8 AI Models Will Dominate Most Workloads
Amazon CEO Andy Jassy stated that the AI model landscape will likely feature 6-8 models that are equally capable for over 95% of workloads, rather than a single dominant model. This prediction suggests that companies will use multiple models based on specific task requirements, influencing how services like Amazon Bedrock are designed to offer access to various foundation models.
AWS Hydro Team Explores Compiler-Enforced Correctness Checks for AI-Generated Code
The AWS Hydro team is developing methods to use compilers for catching errors in code written by AI agents, particularly in distributed systems. This initiative addresses the challenges of reviewing large code contributions from agents and the limitations of agent-generated tests, aiming to establish domain-specific guardrails.
Google opens Android development tools to third-party coding agents
Google is providing deeper access to Android development tools, including devices, profilers, and crash reports, for third-party coding agents like Anthropic's Claude Code and OpenAI's Codex. This initiative allows agents to investigate bugs, test changes, and verify fixes directly within Android Studio and the Android CLI, aiming to accelerate high-quality Android app development.
Vercel pushes vendors to adopt OIDC, but its own deployment token remains static
Vercel is requiring Marketplace vendors to adopt OpenID Connect (OIDC) and short-lived tokens by January 2027, citing security concerns with static credentials. However, developers noted that Vercel's own GitHub Actions deployment workflow still uses a persistent token, creating a double standard.
Microsoft Introduces New Voice AI Models for Real-time Speech Processing
Microsoft Foundry has released three new AI models, including MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash. These models enable real-time speech-to-text transcription and expressive text-to-speech, allowing applications to process and respond to spoken requests more quickly.
OpenAI's GPT-6.1 Sol achieves 97.9% accuracy in COBOL migration tests
OpenAI announced its GPT-6.1 Sol model achieved 97.9% accuracy in internal COBOL migration tests, a significant increase from its predecessor. This development indicates progress in using AI for modernizing legacy COBOL systems, which are critical for financial infrastructure.
Cloudflare open-sources Void to manage employee-built AI applications
Cloudflare released Void, a collection of three open-source tools, including a self-hosted platform for deploying and managing employee-created AI applications. This initiative addresses the challenge of 'shadow IT' by providing IT departments with oversight and control over internal applications built by non-engineers.
Vercel Introduces Daemons for Always-On Background Workers and General Availability of Python Support
Vercel launched Daemons, a new service for continuously running background workers, and made its Python support generally available. These additions expand Vercel's backend capabilities, allowing developers to run more complex, continuously processing workloads directly on the platform.
Linus Torvalds Used Claude Code for Linux Page Cache Fix, Reviewers Find Race Condition
Linus Torvalds used Anthropic's Claude Code to generate patches for a long-standing Linux kernel page-cache performance bottleneck, resulting in a 23-fold throughput improvement in a synthetic benchmark. However, reviewers subsequently identified a race condition and memory reclamation concerns in the submitted 10-patch series. Torvalds has transferred further development to Kiryl Shutsemau.
Workday Opens HR and Finance Data to External AI Agents with Headless Platform
Workday has made its Trusted Agent Platform "headless," allowing developers to build AI agents that interact directly with HR and finance data using tools like Claude Code and Cursor. This change provides direct SQL access to live Workday data and ensures all agents operate under existing Workday permissions and audit trails, addressing data access control concerns.
Amazon ECS introduces automatic repair for failing GPUs and instances
Amazon ECS now automatically repairs failing GPUs and instances to improve application resilience. This update shifts some infrastructure failure recovery responsibility from users to ECS, simplifying operational management for applications running at scale.
Percona releases Valkey-proxy to simplify migration to Valkey Clusters
Percona introduced Valkey-proxy, an open-source proxy designed to allow existing applications to connect to a Valkey Cluster without requiring code rewrites. This development addresses a significant barrier to Valkey adoption for applications originally built to interact with Redis as a single instance.
OpenSearch Veterans Launch Infino, an Agent Retrieval Platform for AI Agents
Infino, founded by OpenSearch veterans, launched its agent retrieval platform, which provides a unified database system for AI agents to query all data from a single location. This platform addresses the fragmented data stack currently used by AI agents, which were not optimized for their unique query patterns.
Google's Docsy documentation project joins Linux Foundation, adds AI-focused features
Google's Docsy, an open-source documentation theme for Hugo, has moved to the Linux Foundation. This transition aims to bring the project closer to its users, particularly within the cloud-native community, and includes new features designed to make documentation more accessible to AI agents.
SAP acquires AI company TechWolf to enhance HR agent context in SuccessFactors
SAP acquired TechWolf, a Belgian AI company that maps employee skills and work, to integrate its technology into SAP SuccessFactors HR agents. This acquisition aims to provide more context for HR tasks like skills-based hiring and talent management, making AI agents more efficient and reducing operational costs.
Dynatrace Completes Acquisition of Arize for AI Observability
Dynatrace has finalized its acquisition of Arize for $915 million, integrating Arize's AI model and agent observability capabilities with Dynatrace's existing application and infrastructure monitoring. This acquisition addresses the new challenges in observability posed by large language models (LLMs) and AI agents, which require specialized monitoring beyond traditional infrastructure performance.
Palo Alto Networks launches Cortex XCOR for AI-driven observability and automated incident response
Palo Alto Networks introduced Cortex XCOR, an AI-driven observability platform designed to automatically identify root causes and recommend fixes for system outages. This platform aims to shift incident response from manual dashboard monitoring to autonomous AI agent investigations, building on the acquisition of Chronosphere.
Cohere Releases Embed 5 with Pro and Fast Models for Optimized RAG Workloads
Cohere launched Embed 5, introducing Pro and Fast models that share an embedding space, allowing users to index data with the Pro model and query with the faster, cheaper Fast model. This setup aims to reduce latency and cost in RAG and agent workloads, with Cohere's tests showing minimal impact on retrieval quality.
Microsoft Fabric Integrates AI Agents for Enterprise Data Understanding
Microsoft is positioning its Fabric data platform as a central hub for AI agents to understand enterprise operations. This integration allows agents to access business context, with Fabric IQ now feeding Microsoft 365 Copilot and Power BI gaining capabilities to convert definitions into applications.
pgEdge Launches Starfleet, a Postgres Cloud Platform for Database Branching and AI Agents
pgEdge introduced Starfleet, a Postgres cloud platform that provides isolated database branches for AI coding agents. This platform allows each agent to work on a separate copy-on-write database branch, addressing challenges in moving AI agent prototypes to production environments.
Azul Launches AI Assistant for Real-time Java Security and Licensing Risk Management
Azul introduced its Intelligence Cloud AI Assistant, a natural-language query interface designed to provide real-time insights into security vulnerabilities and licensing infringements within live Java production environments. This tool addresses the challenge of static reports quickly becoming outdated, especially as AI accelerates the weaponization of newly disclosed vulnerabilities.
Cursor launches Rollouts bot for production code change tracking after Firetiger acquisition
Cursor launched Rollouts, a new bot that tracks code changes through to production and monitors their behavior. This release follows Cursor's acquisition of Firetiger, a startup specializing in AI agents for monitoring software changes from pull request to deployment.
Xiaomi releases MiMo-V2.6, a trillion-parameter open model, and details training costs
Xiaomi launched MiMo-V2.6, a new open-weight AI model with a trillion parameters and multimodal capabilities, ranking highly among open models. The company publicly livestreamed its reinforcement learning (RL) training process, disclosing costs of approximately $3.5 million for this phase.
Vercel tightens free-tier deployment retention rules due to increased usage
Vercel has updated its Hobby plan policy, immediately deleting older, unprotected deployments for teams exceeding 10GB of storage. This change aims to maintain the viability of the free tier amidst a significant increase in deployment volumes.
Intel Reduces LLM Weight Storage to 1.485 Bits Using New BITCOS Format
Intel researchers developed BITCOS, a new storage format that compresses ternary LLM weights to 1.485 bits per weight without altering the model itself. This method improves decoding throughput by up to 27% on GPUs and 18% on CPUs by optimizing how zero weights are stored. The innovation addresses the inefficiency of traditional ternary weight storage, which assumes an equal distribution of values, by accounting for the higher frequency of zero weights in real models.
Meta enables AI agents Claude and Codex to configure WhatsApp Business via MCP
Meta now allows AI coding agents like Claude and Codex to configure WhatsApp Business accounts through its MCP server. This change simplifies the setup process for businesses by automating tasks such as phone number verification, webhook configuration, and message template management, which previously required manual developer intervention.
Kubernetes 1.36 Restores Consistency Group Guarantee for Database Backups
Kubernetes 1.36 addresses a long-standing issue where multi-volume database backups could be inconsistent due to the lack of a consistency group feature, which was lost in the transition to cloud-native storage. This update allows for coherent restores of stateful applications by ensuring all volumes are snapshotted at the same instant, preventing data corruption during recovery.
Cohere restricts commercial use of new North Small Translate 1.0 model weights
Cohere released its North Small Translate 1.0 model with open weights for non-commercial use, but requires a commercial license and deployment through its Model Vault for enterprise production use. This move contrasts with Cohere's stated emphasis on AI sovereignty, as it limits how enterprises can deploy the model independently.
Salesforce introduces Enterprise AI Harness to unify six core capabilities
Salesforce launched its Enterprise AI Harness, integrating six existing platform tools with a new AI control plane to create an open and composable AI ecosystem. This initiative addresses the challenge of siloed AI automation within enterprises by providing a unified approach to leverage customer data and operational intelligence. The harness aims to differentiate enterprises by their ability to securely turn proprietary context into action, rather than just model choice.
OpenAI Launches GPT Images 2.5 with Flare and Sunburst Models for Image Editing
OpenAI released GPT Images 2.5, introducing two new models, Flare and Sunburst, designed to improve localized image editing without affecting other parts of an image. This update provides developers with options for faster or more precise image generation, impacting workflows that require specific image modifications.
Andela Analysis Identifies 23 New AI-Driven Engineering Roles from 47,000 Job Listings
Andela analyzed 47,000 engineering job postings from Fortune 500 companies, identifying 23 new job titles and over 2,000 skills emerging due to AI. This research indicates that AI is creating specialized roles by merging existing skill sets, rather than promoting generalist positions.
IFM Releases K2 Horizon: Six AI Models with Open Training Data and Code
The Institute of Foundation Models (IFM) released K2 Horizon, a suite of six AI foundation models ranging from 0.9 billion to 375 billion parameters. IFM defines these as "fully open-source" by committing to publish training and evaluation code, data, configurations, logs, and intermediate checkpoints, though some components are not yet available for all models. This initiative aims to enable developers to inspect, reproduce, and adapt the models.
DeepSeek to hire 150 engineers for AI agent infrastructure, not model development
DeepSeek is hiring approximately 150 engineers to focus on server-side engineering and Agent Elastic Compute, rather than AI research, to scale its infrastructure for AI agent workloads. This hiring initiative addresses the growing demands of training, evaluation, and backend systems as the company expands its concurrent AI agent sandboxes.
PhiloLabs AI Agents Reconstruct Virtual Union Square for $33, Identify Visual Errors
PhiloLabs used Claude Fable 5.1 AI agents to reconstruct a 3D virtual model of San Francisco's Union Square for $33 in API calls. The experiment highlighted the agents' ability to not only build functional code but also identify visual inaccuracies using Playwright for comparison against real-world images.
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.