2026-03-24 16:07
🚀 Exciting news in AI! The Allen Institute of AI (Ai2) has launched MolmoWeb, an open-source visual web agent designed to perform tasks in web browsers. MolmoWeb offers two sizes (4B and 8B parameters) and is built to navigate websites, fill forms, and search for products. It stands out by using synthetic data for training, rather than proprietary models. While it shows strong performance against some benchmarks, proprietary models still lead overall. Ai2 aims to provide an alternative for...
Source: The New Stack
Frederic Lardinois
2026-03-24 16:01
AI-generated code presents security risks, including potentially harmful commands. Dan Phillips discussed this at Wasm I/O in Barcelona, highlighting the need for reliable isolation for AI agents. WebAssembly offers a solution through its unique memory model, avoiding shared kernel pitfalls of traditional sandboxing methods. 💻🔒 For effective execution environments, WebAssembly could be the key. #WebAssembly #AISecurity #TechInnovation #CyberSafety #LLM
Source: The New Stack
B. Cameron Gain
2026-03-24 15:38
Broadcom VMware has announced the donation of Velero to the CNCF sandbox project. This initiative aims to enhance Kubernetes backup and disaster recovery for users across private and public clouds. 🌐 The move responds to user feedback and expands integration with open-source options, providing more flexibility for Kubernetes infrastructure. Broadcom also plans to enhance its vSphere Kubernetes Service with API compatibility for various third-party suppliers. This donation allows the open-...
Source: The New Stack
B. Cameron Gain
2026-03-24 15:20
IBM, Red Hat, and Google have introduced llm‑d, a new Kubernetes blueprint for deploying large language model (LLM) inference, now donated to the CNCF. 🤖 This open-source framework aims to streamline the serving of foundation models, enhancing scalability and cloud-native capabilities. It disaggregates inference processes and incorporates intelligent routing and scheduling to optimize performance. Founding collaborators include NVIDIA and CoreWeave, supporting a community-governed approach. 🌐...
Source: The New Stack
Steven J. Vaughan-Nichols
2026-03-24 14:20
🚀 Brendan Burns, co-founder of Kubernetes, discusses the impact of AI on infrastructure in a recent interview. He emphasizes how AI-generated code will soon be as overlooked as assembly language. 💻 Kubernetes is adapting to AI workloads, requiring a rethink of scheduling and monitoring. GPU scheduling has become crucial, as it involves complex resource requirements beyond just CPU and memory. 🔄 The shift from stateless to batch workloads highlights the need for new strategies in handling...
Source: The New Stack
Frederic Lardinois
2026-03-24 13:00
PwC is reshaping its consulting model by introducing AI agents directly to clients through the new PwC One platform. Clients can now describe their issues, allowing autonomous agents to handle tasks, while consultants review the results. This approach moves away from traditional engagement methods, integrating AI into core workflows. The goal is to create a seamless, self-service experience, with human oversight only where necessary. PwC aims for a more continuous client relationship,...
Source: The New Stack
Darryl K. Taft
2026-03-24 12:43
Discover ExLight, the latest lightweight Linux distribution from Exton Linux! 🌟 Based on Debian 13, it features the Enlightenment 0.27.1 Desktop environment. ExLight is known for its efficiency, stability, and user-friendly interface, making it suitable for both new and experienced users. With tools like Refracta Snapshot and the Calamares Installer, it offers a blend of modern aesthetics and classic functionality. Explore the possibilities with ExLight! 🖥️✨ #Linux #ExLight #OpenSource...
Source: The New Stack
Jack Wallen
2026-03-24 10:00
🚀 Minimus has launched a new initiative aimed at improving security for open-source projects. Eligible maintainers can now access secure container images, SBOM generation, and threat intelligence tools at no cost. This program seeks to enhance software supply chain integrity. 🔒 Minimus emphasizes the need for a ruggedized approach to container security, reducing vulnerabilities and protecting critical digital infrastructure. For developers, this initiative provides essential tools often...
Source: The New Stack
Adrian Bridgwater
2026-03-23 18:00
Unlock the potential of LLMs through effective prompt engineering! 🛠️ The article highlights the importance of crafting clear and specific prompts to achieve reliable outputs. It emphasizes that vague requests often lead to unpredictable results. Understanding how LLMs process prompts is key, as they follow a hierarchy of instructions. To improve your results, research your requirements before prompting and ask the right questions. #PromptEngineering #AI #TechTips #LLM #Developers
Source: The New Stack
Jessica Wachtel
2026-03-23 14:44
🛡️ Want to block ads on every device in your home? Consider using Pi-Hole! Pi-Hole is an open-source solution that operates on a network-wide basis, eliminating the need for individual ad-blockers on each device. To set it up, deploy Pi-Hole and configure your devices to use it as the DNS server. This method also enhances network performance and includes monitoring features. Keep in mind, if you're using certain ISPs like AT&T Fiber, DNS settings might not be changeable on your router. To...
Source: The New Stack
Jack Wallen
2026-03-23 12:00
The rise of Model Context Protocol (MCP) has generated excitement around "Digital coworkers." However, those with established APIs need not be concerned about abandoning their hard work. APIs serve as structured communication bridges between systems. They're like a restaurant menu—specific and predefined, ensuring clarity in data retrieval. While AI agents can interact with APIs, they may also pose risks, such as overcalling endpoints or leaking sensitive data. Nevertheless, APIs are crucial...
Source: The New Stack
Camille Crowell-Lee
2026-03-22 14:00
A recent audit by Mobb.ai assessed 22,511 AI coding skills across public registries, revealing significant security concerns. The study identified over 140,000 security findings, highlighting a critical gap in security measures. While skills are scanned at publication, they execute with full system permissions once installed, posing risks to developer systems. 🔍 CEO Eitan Worcel emphasized the growing reliance on AI coding agents, noting that developers may unknowingly grant extensive access...
Source: The New Stack
Darryl K. Taft
2026-03-22 13:00
What would an AI-first programming language look like? A developer in Spain raised concerns about human-friendly syntax being inefficient for AI, leading to new attempts at creating AI-centric languages. These languages aim for clarity and efficiency, targeting autonomous AI agents. Andrea Griffiths from GitHub notes that while there are experiments, none have gained significant traction yet. Existing programming ecosystems create a barrier to widespread adoption of new languages. The...
Source: The New Stack
David Cassel
2026-03-21 17:00
OpenClaw's popularity is evident, with nearly 1,000 people lining up outside Tencent's Shenzhen headquarters for software installation. This open-source AI tool, developed by an Austrian company, is part of a trend among Chinese startups to innovate with government support. Anthropic has now introduced Claude Dispatch for Cowork, enhancing AI interactions with local drives. This tool aims to streamline user experience without needing complex setups. For those on a Pro plan, a research preview...
Source: The New Stack
David Eastman
2026-03-21 15:45
WebAssembly (Wasm) and Kubernetes serve different purposes in cloud deployment. An article highlights that while WebAssembly simplifies application deployment, it does not replace Kubernetes. It emphasizes the integration of WebAssembly with Helm, particularly with Helm 4 supporting direct Wasm modules, enhancing security and isolation. This development allows for better policy implementation and multi-language support in applications. #WebAssembly #Kubernetes #DevOps #CloudComputing #Helm
Source: The New Stack
B. Cameron Gain
2026-03-21 12:52
🚀 This week in AI: Cursor's new Composer 2 coding model outperforms Opus at a fraction of the cost, marking a significant shift in AI development. Composer 2 scored 61.7% on the Terminal-Bench 2.0, while Opus lagged at 58%. In other news, Nvidia is forming a coalition to build open base models, and Congress introduced the 300-page Trump America AI Act to unify AI regulations. Stay tuned for more updates! 🤖💡 #AI #TechNews #Cursor #OpenAI #Innovation
Source: The New Stack
Matthew Burns
2026-03-20 20:18
AI is changing how teams approach infrastructure as code (IaC), but challenges remain. Marcin Wyszynski, co-founder of Spacelift, highlights that while AI can automate coding for tools like Terraform, understanding the underlying changes is crucial. A lack of comprehension can lead to significant risks in production environments. As teams seek faster access to infrastructure, balancing ease and safety becomes vital. Spacelift's new product, Intent, aims to address this gap....
Source: The New Stack
Frederic Lardinois
2026-03-20 17:46
China is advancing in the open-source AI sector, recently surpassing the US in model downloads. 🇨🇳📈 Platforms like Hugging Face highlight this shift, with Chinese models now constituting 41% of downloads. While China leads in model production, it still relies on US infrastructure, primarily Nvidia. Key reports indicate a significant change in the AI landscape. #OpenSourceAI #AIEcosystem #China #HuggingFace #Nvidia
Source: The New Stack
Paul Sawers
2026-03-20 16:00
Kubernetes is evolving into a key infrastructure for AI, as discussed in a recent post by Hyperframe Research. This shift indicates we are moving beyond using Kubernetes just for its own sake. Now, its value is defined by the AI workloads it supports. To optimize AI inference, developers need Kubernetes to be reliable and easy to use, avoiding complexities that can hinder performance. The focus is on creating a seamless experience for deploying AI models close to users, ensuring that...
Source: The New Stack
Danielle Cook
2026-03-20 14:33
📢 OpenAI has acquired Astral to integrate its open source Python developer tools into the Codex ecosystem. Both companies assure that popular tools like uv, Ruff, and ty will continue to be supported as open source. However, details on governance and contributions remain unclear. OpenAI aims to enhance Codex, which has seen significant growth, by incorporating Astral's expertise to streamline the entire Python development workflow. #OpenAI #Astral #Python #Coding #DeveloperTools
Source: The New Stack
Meredith Shubel
2026-03-20 14:00
Kubernetes networking offers flexibility but can become complex as systems scale. 🌐 Flat network security models often lead to challenges like debugging issues and enforcing compliance. This is due to a lack of manageable priorities in policies, making it hard to predict changes. 🔍 Introducing a security hierarchy can help by providing clear order and responsibility for network policies, reducing risk and improving operations. 🔒 #Kubernetes #CloudNative #NetworkSecurity #DevOps #TechTrends
Source: The New Stack
Reza Ramezanpour
2026-03-20 13:00
The observability industry is shifting from proprietary systems to open frameworks, moving away from the traditional pillars of metrics, logs, and traces. With OpenTelemetry, data is now consolidated into unified streams, reducing information overload for operators. This integration allows for AI-powered platforms that streamline analysis and enhance workflows. The focus is on transforming how operators detect and resolve issues, moving towards user-friendly solutions that present data...
Source: The New Stack
Ted Young
2026-03-20 11:30
The Linux kernel faces unique challenges in maintaining backward compatibility while evolving as the most popular operating system. 🖥️ A year ago, the decision to become a CVE numbering authority (CNA) aimed to enhance transparency in security disclosures. However, this has led to an overwhelming increase in reported vulnerabilities. 📈 In 2025, the Linux kernel was identified as the most vulnerable technology, with many issues being less critical in nature. This high volume of alerts is...
Source: The New Stack
Jed Salazar
2026-03-19 18:37
Backend development, or server-side development, is essential for building applications that run on developer-controlled computers. It focuses on data management and business logic, contrasting with frontend development, which centers on user experience. In 2026, backend developers have the freedom to choose their programming languages, while frontend developers are limited to those supported by browsers. A key focus area now includes the integration of AI agents with backend services,...
Source: The New Stack
TNS Staff
2026-03-19 18:01
Developers are increasingly using AI coding tools, influenced by their trust in automation and project flexibility. Jellyfish's AI Engineering Trends study analyzed data from over 700 companies and 200,000 engineers. Findings show that more than half of companies consistently use AI tools, leading to improved code generation and increased productivity. 📈 However, fully autonomous code agent activity remains low, yet is rapidly growing. This trend may have significant implications for the...
Source: The New Stack
Adrian Bridgwater
2026-03-19 15:39
🚀 Cursor has launched Composer 2, its latest in-house coding model that outperforms Anthropic’s Opus 4.6 on key benchmarks while being significantly cheaper. 💰 The cost for Composer 2 is as low as $0.5 per million input tokens, compared to Opus 4.6's $5. 📊 On Terminal-Bench 2.0, Composer 2 scored 61.7%, closing the gap with competitors. ⚙️ This new model uses innovative 'self-summarization' training for improved performance. #Cursor #AI #Coding #TechNews #Innovation
Source: The New Stack
Frederic Lardinois
2026-03-19 15:39
🚀 Cursor has launched Composer 2, its latest coding model, now outperforming Anthropic's Opus 4.6 on key benchmarks at a much lower cost. The pricing starts at $0.5 per million input tokens, while Opus 4.6 is priced at $5. Composer 2 scored 61.7% on Terminal-Bench 2.0, marking significant progress since previous versions. This model utilizes a new training technique called 'self-summarization' for enhanced performance. #Cursor #AI #Coding #TechNews #Innovation
Source: The New Stack
Frederic Lardinois
2026-03-19 15:00
In modern observability, distributed tracing is vital for capturing execution context in systems. OpenTelemetry plays a key role in enabling span collection across various technologies. However, with high volumes of data generated, effective querying can be challenging. Sampling helps by selectively retaining portions of tracing data, reducing complexity. Two main sampling approaches exist: 1. **Head sampling** decides upfront whether to collect spans. 2. **Tail sampling** records all...
Source: The New Stack
Michele Mancioppi
2026-03-19 12:00
🚀 Adopting large AI models on Kubernetes can seem straightforward initially, but complexities arise in operations as usage scales. Day 0 may feel manageable, but Day 1 and 2 present challenges like latency sensitivity and unpredictable traffic. A real-world example highlights how AI platforms must handle incident triage efficiently. Common issues include fragmented GPU capacity, inconsistent inference interfaces, and the need for reliable multi-stage workflows. #AI #Kubernetes #Infrastructure...
Source: The New Stack
Sachi Desai
2026-03-18 17:00
At CES 2026, NVIDIA CEO Jensen Huang emphasized the importance of open innovation for the growth of AI across all industries. He suggested that AI infrastructure must also be open-sourced to support this shift. 🤖 Kubernetes has adapted to run AI workloads, but challenges remain in resource allocation and scheduling. Dynamic Resource Allocation (DRA) is a key development, allowing better management of GPU resources. The KAI Scheduler aims to enhance gang scheduling and placement, optimizing...
Source: The New Stack
Nathan Taber
2026-03-18 16:35
🚀 Oracle has released Java 26 at the JavaOne 2026 conference, introducing 10 JDK Enhancement Proposals (JEPs) focused on performance, security, and language improvements. While this release is not an LTS version, it includes updates like HTTP/3 support, enhanced garbage collection, and cryptographic tooling. Experts highlight that Java 26 is important for developers seeking cutting-edge features, especially in AI applications. #Java26 #SoftwareDevelopment #TechNews #AI #JavaOne
Source: The New Stack
Darryl K. Taft
2026-03-18 16:24
🚀 Chainguard is tackling a growing security issue in software development caused by AI coding assistants. These tools often rely on outdated training data, leading to the use of older, insecure library versions. 🔍 As AI accelerates code generation, attackers are also leveraging AI to exploit vulnerabilities faster than ever. A recent breach of the Trivy project highlights this risk, showcasing how AI can automate the discovery of misconfigurations. 🛡️ To combat these challenges, Chainguard...
Source: The New Stack
Darryl K. Taft
2026-03-18 16:19
🚀 Sauce Labs has launched AI for Test Authoring to address challenges in software testing. With AI speeding up code generation, testing has become a bottleneck. Their new tool allows users to describe desired application behavior in natural language, generating executable test suites automatically. CEO Prince Kohli highlights that testing has long been deprioritized, leading to a gap in quality. Intent-driven testing aims to streamline the process and enhance efficiency. #SoftwareTesting...
Source: The New Stack
Frederic Lardinois
2026-03-18 15:02
Capital One has deprecated an AI tool, emphasizing the importance of enabling their 14,000 engineers. Catherine McGarvey, SVP of Developer Experience, highlights the need for proper tools and knowledge to ensure success. The focus is on platform engineering and standardization to streamline processes. This approach reflects a commitment to continuous improvement and efficiency in a regulated tech environment. #CapitalOne #DeveloperExperience #AI #TechInnovation #Engineering
Source: The New Stack
Jennifer Riggins
2026-03-18 12:00
🚀 Chronosphere achieved a remarkable 74% reduction in storage costs by transitioning from ext4 to Btrfs for managing petabytes of time-series data. Btrfs, a copy-on-write filesystem for Linux, offers features like transparent checksumming and compression. This change significantly decreased their data storage footprint, allowing for more efficient resource use. For those interested, Btrfs can be tested on Google Cloud Platform. #Btrfs #CloudStorage #DataManagement #TechInnovation #CostSavings
Source: The New Stack
Motiejus Jakštys
2026-03-18 11:00
Engineering teams are facing rising observability costs, often attributing this to vendor pricing. However, the real issues lie within their telemetry practices. 📉 Many organizations generate unnecessary data, lacking ownership and useful attributes, leading to inflated bills. One analysis revealed 82% of data points lacked basic identifiers. 🏷️ Additionally, sensitive information is frequently leaked into observability pipelines, raising security concerns. Effective governance and...
Source: The New Stack
Juraci Paixão Kröhling
2026-03-17 20:04
🚀 Chainguard is addressing container security challenges faced by DevOps teams. Their new Chainguard OS Packages allow teams to build secure Linux images tailored to their needs, without the hassle of tracking CVEs. This package uses zero-known-CVE components, ensuring safety while offering customization. Chainguard CEO Dan Lorenc compared it to a meal kit from a Michelin-starred supplier, where organizations can craft custom images from trusted packages. For more control and security in...
Source: The New Stack
Steven J. Vaughan-Nichols
2026-03-17 18:57
🚀 OpenAI has introduced GPT-5.4 mini and nano, designed for specific tasks like codebase searches and file reviews. These smaller models offer faster performance at lower costs, with mini available in various platforms and nano being the most affordable option. Early benchmarks show mini closely matches the flagship model in coding tasks, while nano excels in basic operations but is less efficient for complex tasks. Learn more about these advancements in AI models. #OpenAI #GPT5 #AIModels...
Source: The New Stack
Frederic Lardinois
2026-03-17 18:50
✨ WebMCP enables any Chrome web page to serve as an MCP server for AI agents. This development simplifies how agents interact with web content, making navigation more efficient. 🌐 With a Chrome extension, agents can access the DOM and perform tasks using standard HTML components. This method enhances collaboration between AI and users, allowing for real-time queries about displayed content. 🔍 The approach includes two APIs: the Declarative API for standard actions and the Imperative API for...
Source: The New Stack
David Eastman
2026-03-17 18:42
GitHub Copilot is reshaping how developers allocate their time. A recent study surveyed 187,000 open-source developers and found a notable shift in work patterns. After using Copilot, coding time increased by 12.4%, while project management tasks decreased by 24.9%. However, concerns remain about the quality of AI-generated code, with many developers expressing distrust in its reliability. As AI tools continue to evolve, their impact on developer roles and productivity is becoming more...
Source: The New Stack
Steven J. Vaughan-Nichols