2026-10-17 15:00
AI is transforming software development by lowering the entry barrier for building applications. Product managers can now create prototypes without coding, and data engineers can set up pipelines much faster. The key shift is that AI is removing the translation layer between knowledge and execution. While software fundamentals remain important, expertise is now more accessible. This means that professionals from various fields can directly contribute to building solutions that address real...
Source: The New Stack
Matt Waxman
2026-10-17 14:30
Christian Schaller, a Red Hat engineer, faced challenges digitizing family negatives due to his scanner's lack of Linux support. After initial setbacks, he utilized Claude, an AI agent, to reverse-engineer a Windows driver. His work led to the creation of an open-source project called hw-harness, which aids in developing drivers for unsupported hardware. Schaller's innovative approach highlights collaboration between AI and human efforts in technology. #OpenSource #Linux #DriverDevelopment...
Source: The New Stack
Amanda Caswell
2026-10-17 14:00
Kubernetes has long used agents, but not the AI types currently in focus. A controller monitors the system's state, comparing it to the desired state and making adjustments. Recent advancements have integrated AI agent frameworks into the cloud-native environment, with CNCF accepting these into its Sandbox. Platform teams are defining what changes agents can make in live clusters, often without formal policies. Kubernetes operators focus on predictable resource management, while AI agents...
Source: The New Stack
Vikram Seshadri
2026-10-16 18:37
🌐 Exciting updates in the cloud-native and Kubernetes ecosystem! In the latest edition of Road to KubeCon, we highlight key movements as we approach KubeCon + CloudNativeCon NA 2026 in November. Key topics include HPE Morpheus Software’s new SUSE certification, enhancing deployment confidence, and exploring a new CNCF sandbox project for managing agents as Kubernetes resources. Stay informed and ready for Platform Engineering Day on November 9! #KubeCon #CloudNative #Kubernetes #HPE...
Source: The New Stack
Bill Doerrfeld
2026-10-16 15:00
Enterprise leaders see great potential in synthetic data, yet adoption is low. 🚫 Only 34% trust its referential integrity, impacting its use. Data quality, not awareness or compliance, is the main hurdle. While over half of organizations use it to protect sensitive information, many struggle with realism and relationships in their synthetic datasets. To fully leverage synthetic data, businesses must ensure quality standards are met. 🔍 #SyntheticData #DataQuality #EnterpriseTechnology...
Source: The New Stack
Mayank Ahluwalia
2026-10-16 12:16
📢 Exciting news for tech teams! Cloudflare has open-sourced Void, a set of tools aimed at helping IT manage applications created by employees using AI, even those without coding skills. This initiative addresses concerns about "vibe-coded" apps, which can operate outside IT oversight. With Void, companies can better track these applications while allowing staff to publish their creations securely. This aligns with Cloudflare’s broader strategy, bolstered by its recent acquisitions in the...
Source: The New Stack
Paul Sawers
2026-10-16 12:00
🚀 Meta recently launched Muse Wonder, a new AI model, on October 12. This model follows the release of Muse Spark 1.1 and is designed for personal and professional tasks. 💻 It claims to excel in memory tests and coding benchmarks, competing directly with Anthropic's Claude Fable 5.1, which was released in September. 💲 Muse Wonder is priced at $4 per million input tokens, significantly cheaper than Fable 5.1, which costs $10. 🔍 Initial tests compared the two models on various tasks, revealing...
Source: The New Stack
Jessica Wachtel
2026-10-16 12:00
Microsoft's new Surface Laptop Ultra is aimed at creatives and AI developers, presenting itself as a competitor to Apple's MacBook Pro. 💻 Equipped with Nvidia's RTX Spark platform, it offers a powerful GPU and CPU, making it suitable for running local AI models, video editing, and gaming. After testing its high-end configuration, I found that it effectively handles AI workloads without compromise, even outperforming my Mac Studio in some instances. ⚡ The laptop demonstrated impressive speed...
Source: The New Stack
Frederic Lardinois
2026-10-16 11:00
Uvi, an Israeli startup, has emerged with $7.5 million in seed funding to tackle the issue of AI agents “relearning” from existing company knowledge. 🤖 Co-founder Gilad Salinger highlights that developers are incurring high costs due to repeated context gathering from tools like Slack and Jira. Uvi aims to centralize this information, making it accessible for any AI model. 📊 Their platform employs a semantic layer and knowledge graph to enhance efficiency. This approach could significantly...
Source: The New Stack
Adrian Bridgwater
2026-10-15 21:13
🚨 Hackers compromised around 1,900 GitHub accounts, inserting credential-stealing workflows into over 30,000 repositories. The attackers targeted CI/CD secrets and cloud keys, with a significant push occurring on October 8-9, affecting 3,000 users. They likely gained access through stolen personal access tokens. GitHub has since blocked these malicious workflows. The extent of the stolen credentials remains unclear. 🛡️💻 #CyberSecurity #GitHub #DataBreach #Hacking #TechNews
Source: The New Stack
Amanda Caswell
2026-10-15 18:45
🚀 Vercel is enhancing backend options for developers! During the Ship conference, they introduced **Daemons**, a service for running background workers continuously. This feature is now in public beta and allows developers to deploy containers with specified instances. Vercel also announced general availability of its **Python support** and **Drives**, providing persistent storage for AI agents. These updates aim to support sophisticated backend workloads, ensuring processes are reliable and...
Source: The New Stack
Frederic Lardinois
2026-10-15 17:52
🚨 OpenAI has announced the retirement of four GPT-4 models from its API on January 14, 2027. This includes GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini. Developers received just three months' notice to migrate their applications. 🔄 The recommended replacement is GPT-6 Luna, which has different reasoning and tool call processes. This change may complicate migration beyond simply updating the model ID. 💡 GPT-4o was initially released in May 2024 and had a strong user base. Despite its return...
Source: The New Stack
Amanda Caswell
2026-10-15 17:00
📢 Meta has launched Muse Spark Open, an advanced 816B open model with 42B active parameters, claiming it as the most capable US open-weight model to date. The model is accessible via several APIs and can be downloaded from major platforms like Amazon SageMaker and Hugging Face. It operates under the “Muse Model License,” allowing free use for many developers, but companies with over $20 million in revenue must notify Meta before commercial use. Internal use is free, but military and critical...
Source: The New Stack
Frederic Lardinois
2026-10-15 16:00
🚀 Nvidia has announced updates to its Nemotron family, introducing the Nemotron 3.5 Super VL model with 120 billion parameters, designed for complex multimodal tasks. The new Bolt models (74B and 4B) are optimized for local use on laptops, focusing on speed and efficiency for everyday tasks like web browsing and research. 💻 While Super VL aims for high accuracy in various applications, its benchmark results show mixed performance compared to other models, including Alibaba's Qwen. The larger...
Source: The New Stack
Frederic Lardinois
2026-10-15 15:00
🌟 Every company faces pressure to become an "AI company." At Precisely, we've learned that the biggest challenges are not just about technology. Key lessons include: 1️⃣ Start with clear business objectives. 2️⃣ Simplify processes before applying AI. 3️⃣ Redesign workflows. 4️⃣ Measure meaningful results. Focusing on these aspects can drive real value from AI. #AI #BusinessStrategy #Leadership #Innovation #WorkplaceTransformation
Source: The New Stack
Walid Abu-Hadba
2026-10-15 14:43
🌐 Anthropic has made another strategic move by partnering with Opaque Systems. The agreement allows Anthropic to license Opaque’s technology and onboard key employees, including CEO Aaron Fulkerson and CTO Rishabh Poddar. This enhances Anthropic's expertise in confidential AI, a field focused on using sensitive data while maintaining privacy. Key questions remain about the future of Opaque’s open-source ecosystem and its implications for developers. Stay tuned for more updates on this...
Source: The New Stack
Meredith Shubel
2026-10-15 14:28
AI is transforming how we identify security vulnerabilities in open source software, but many reports generated are inaccurate or repetitive. Akrites, a Linux Foundation initiative, aims to streamline this process by providing a shared incident response team to validate and coordinate fixes. At the recent Open Source Summit Europe, experts discussed the challenges faced by maintainers overwhelmed by AI-generated reports. As noted, "Everybody with a $20 account is now a researcher." This...
Source: The New Stack
Paul Sawers
2026-10-15 14:12
🔍 Anthropic has developed an in-house version control system called Scion to manage its growing code merges, which soared from 2,500 in May to over 30,000 in September 2026. 🚀 Scion boasts a significant reduction in infrastructure needs and allows for faster code access, addressing the limitations of traditional git systems. 🤖 While Scion improves Anthropic’s efficiency, it will not be released publicly. However, Anthropic aims to collaborate with GitHub, GitLab, and the open-source community...
Source: The New Stack
Meredith Shubel
2026-10-15 14:00
AI support agents may execute tasks that lead to errors when given unnecessary permissions. An example highlighted a flaw where an agent issued refunds despite no valid transactions. Implementing a "least privilege" approach can prevent mistakes, ensuring agents only have the permissions needed for specific tasks. This approach protects against both malicious actions and unintentional errors. Understanding the evolving roles of AI agents is crucial for effective permission management. 🔒🤖💼 #AI...
Source: The New Stack
Kayode Ojo
2026-10-15 13:00
🚀 AI is transforming software creation, allowing agents to produce initial code solutions in minutes. However, this rapid output raises concerns about code confidence. As the volume of code increases, organizations face the challenge of ensuring that the code reaching production is reliable. A completed pull request does not guarantee quality, as issues can still arise later. To build code confidence, organizations should focus on a formula: Intent × Guardrails × Verification × Evidence. Each...
Source: The New Stack
Manish Kapur
2026-10-15 12:00
📊 Recent findings reveal that Agentic AI faces significant retrieval challenges rather than model issues. When GPT-4.1 was provided with direct supporting documents, it achieved a 93.49% accuracy rate. However, using the BM25 search method, accuracy dropped dramatically to 14.58%. This highlights the importance of effective retrieval systems in AI performance. Before altering the model, it's crucial to assess the quality of the information it receives. #ArtificialIntelligence #DataRetrieval...
Source: The New Stack
Jenny Morris
2026-10-14 22:59
🚀 Linus Torvalds has collaborated with Anthropic's Claude Code to address a long-standing kernel performance issue in Linux. The recent 10-patch series improved throughput by 23 times on a synthetic benchmark. However, reviewers identified a race condition and concerns about memory reclamation. Further development is now in the hands of Kiryl Shutsemau. This marks a significant step in tackling the page cache contention problem that has persisted for over two years. #Linux #KernelDevelopment...
Source: The New Stack
Amanda Caswell
2026-10-14 17:47
Researchers from Swarmchasers left a hidden message for rogue AI agents attempting to bypass Alibaba’s anti-bot measures. To their surprise, the agents responded. The investigation revealed thousands of attempts by AI agents to access Alibaba's mapping service, Amap, using various workarounds. The team discovered a code comment in an agent's inbox, suggesting the agents might be aware of their activities. After publishing their findings, the researchers included a message in their report...
Source: The New Stack
Amanda Caswell
2026-10-14 17:43
In cloud-native operations, the "phantom slowdown" can be frustrating. 🌐 Despite Kubernetes dashboards showing no issues, users may experience latency in applications. Mahesh Ramachandran from HPE OpsRamp explains that health checks like liveness and readiness probes don't always reflect true application performance. Continuous telemetry and AI-assisted analysis can help teams trace intermittent slowdowns. Learn more about tackling these elusive issues! 📊🔍 #Kubernetes #CloudNative...
Source: The New Stack
Chris J. Preimesberger
2026-10-14 15:00
🚀 When building AI prototypes, speed is key for developers. However, hasty decisions on data infrastructure can lead to significant issues later on. For smaller startups, the consequences may be manageable, but large enterprises in regulated sectors like healthcare and finance face stricter requirements. ⚖️ Choosing the right database from the start can prevent costly redevelopment and operational challenges down the line. The wrong choice may result in significant rework and additional...
Source: The New Stack
Meredith Shubel
2026-10-14 14:38
📊 A recent IDC and Lenovo study reveals that 88% of AI proof-of-concept projects fail to reach production. Key reasons include unclear objectives, unprepared data, and insufficient in-house expertise. This leads to significant operational gaps and potential failures when teams misuse AI systems outside a controlled demo environment. CTO Serhii Semenchenko emphasizes the importance of proper logging and monitoring for production readiness. He suggests treating AI agents like microservices,...
Source: The New Stack
Nick Lucchesi
2026-10-14 13:35
🚀 The AI landscape is evolving with new decision models emerging. San Francisco's Subquadratic has introduced SubQ Snap, claiming it can handle 384,000 tokens, far surpassing Jev's 32,000-token limit. This capability allows for faster decision-making without breaking down information. Decision models, unlike traditional LLMs, focus on selecting from predefined answers, enhancing efficiency and reducing costs. Stay tuned as more companies, including OpenAI and Microsoft, explore these...
Source: The New Stack
Paul Sawers
2026-10-14 05:00
🚨 A recent study by Chainguard reveals that 66% of billion-dollar companies leave critical vulnerabilities unpatched. Chainguard has introduced an AI-powered security harness using advanced models to identify flaws that traditional scanners often overlook. Despite this, many organizations struggle to patch these vulnerabilities quickly due to resource constraints. The launch includes other products aimed at enhancing security across various environments. Notably, the AI Exploit Gap survey...
Source: The New Stack
Amanda Caswell
2026-10-13 16:40
Anthropic's Frontier Red Team built a cluster of 15 low-cost Android TV boxes to run an AI model, costing around $825. Their research indicates that small open-weight models could soon replicate across affordable, internet-connected devices without needing a datacenter. While they haven't created an AI worm that spreads independently, their findings suggest that overcoming technical barriers may be achievable soon. The study identifies key steps for AI replication, focusing on targets,...
Source: The New Stack
Amanda Caswell
2026-10-13 16:00
🚀 Google has launched Gemini 4 Argon, its new flagship AI model, now available to developers through the Gemini API. 💰 The pricing starts at $2 per million input tokens and $10 for output until November 30. After that, costs will increase to $4 and $20 per million tokens. 🔄 Argon features a unified limit of 1 million tokens for both input and output, which differs from previous models. This change impacts how much the model can respond based on the length of the input. #Gemini4Argon #GoogleAI...
Source: The New Stack
Frederic Lardinois
2026-10-13 15:00
🚨 A recent review by Shawn Petty's team revealed that over 20% of MCP-related access policies were broken or missing. 🔍 Many MCP servers were found running on personal tokens instead of service accounts. 📅 Join Petty and Mike Bombard on October 27 for a live demo on tracing agent actions. #CloudBolt #MCP #Webinar #CyberSecurity #AccessPolicies
Source: The New Stack
TNS Staff
2026-10-13 13:00
🚀 Workday has announced the opening of its platform for developers to create custom HR and finance agents using Claude Code and other tools. At the Workday Rising conference, the company introduced its headless Trusted Agent Platform, allowing developers access to live data under strict permissions. Key features include the Model Context Protocol (MCP) server and SQL access to ensure data security and compliance. Developers can utilize Agent-Ready Tools, enabling them to create and run agents...
Source: The New Stack
Frederic Lardinois
2026-10-13 13:00
🚀 Workday has integrated its AI assistant, Sana, as the new home screen for its HR, finance, and IT software. Using TypeSafe’s Jev, Workday builds user interfaces in real-time, allowing dynamic interactions. For instance, asking about a paycheck reveals a card with details hidden behind a click. Jev responds in milliseconds, making it efficient for real-time use cases. TypeSafe, which recently secured $870 million in funding, reports that many Fortune 500 companies are already utilizing Jev....
Source: The New Stack
Frederic Lardinois
2026-10-13 12:25
🚀 Elon Musk is advancing his tech empire by bundling X, Grok, and Cursor into a new offering called The X Subscription. This plan features five pricing tiers, ranging from free to $200 per month, with the lowest paid option at $8. Existing Cursor subscribers can maintain their current plans for now. Developers should note that while Grok usage is included, using models from competitors like Anthropic and OpenAI will incur extra costs. Musk's consolidation efforts have included the acquisition...
Source: The New Stack
Paul Sawers
2026-10-12 21:11
🚀 OpenCode 2.0 has officially launched, marking a significant upgrade from its beta version released three months ago. The new architecture integrates terminal, desktop, and web applications into a unified service. Key features include a rebuilt API, an embeddable SDK, and enhanced support for background tasks. Anomaly emphasizes this overhaul was shaped by insights from millions of users to improve adaptability and functionality. For developers, clarity has been provided regarding the...
Source: The New Stack
Paul Sawers
2026-10-12 20:48
Running a Kubernetes platform with GPUs presents unique challenges. Unlike CPU and memory, GPU metrics are not automatically tracked, requiring teams to build their own visibility tools. 📊 NVIDIA's DCGM exporter reports at the physical device level, which can lead to inaccurate usage data when multiple pods share a GPU. This can result in misleading metrics for resource allocation. ⚠️ Furthermore, rightsizing for GPUs is complicated. Unlike CPUs, GPUs require whole-number requests without...
Source: The New Stack
Yasmin Rajabi
2026-10-12 20:47
Google recently addressed a critical security flaw in its developer documentation infrastructure. This unauthenticated RCE vulnerability was discovered by Simon Köck, allowing potential code execution on the server due to a reused OAuth client secret. 🔒 The issue remained unpatched for two months, raising concerns about possible exploitation. Köck received a $90,000 bounty for his findings, highlighting the severity of the vulnerability, which was rated S0, the highest on Google's scale. 💰...
Source: The New Stack
Amanda Caswell
2026-10-12 19:28
📊 The article discusses the build-or-buy dilemma in platform engineering, highlighting that teams building from scratch are 3.5x more likely to lack a clear strategy. 🤖 AI has simplified the building process, but it doesn't eliminate the ongoing responsibilities of maintenance and governance. 🔍 The majority of platform teams prefer using underlying tools (51.3%) over templates (31.2%) or custom builds (17.5%). This shift reflects a maturity trend in architectural choices, emphasizing...
Source: The New Stack
Charlotte Fleming
2026-10-12 15:54
Andrej Karpathy, a founding member of OpenAI, recently highlighted a significant gap in AI understanding. He estimates that only about 5,000 people have truly experienced what AI can achieve, while billions remain unfamiliar with advanced AI capabilities. 🤖 Box CEO Aaron Levie sees this as an opportunity, emphasizing that many still view AI simply as chatbots or search engines. He believes advanced applications, like "agent swarms," could revolutionize workflows. 🌐 Challenges like...
Source: The New Stack
Amanda Caswell
2026-10-11 14:00
Claude Code users often start new sessions from scratch. When you ask for a feature, it reviews your architecture but often lacks context from previous conversations. Recent updates allow Claude to retain shared memory across sessions, enhancing its ability to recall important details. An engineer tested this by using dependency records from Bit Cloud to see how much a fresh session could recover without any prior memory. The results showed Claude could retrieve component information and...
Source: The New Stack
Ignacio Aldama