2026-11-13 21:29
🚀 Anthropic has launched a self-hosted version of its Claude Managed Agents, now in private beta. This allows customers to host both the agent coordination server and sandboxes in their own cloud or on-premises data centers. 🔒 However, there are important caveats, including strict deadlines for responding to safety risks and limitations on model usage. For companies managing sensitive data, this option provides increased control while accessing powerful AI capabilities. #AI #SelfHosting...
Source: The New Stack
Paul Sawers
2026-11-13 18:15
AI models are evolving, but the context gap remains a challenge for many enterprises. Sri Desikan from Elastic highlights that without proper context, investments in AI can be wasted. Simply connecting data isn't enough; it requires pre-organizing corporate data into an AI index for efficient access. This approach can lower costs and improve the accuracy of AI agents, though it raises questions about privacy and data handling. #AI #DataManagement #TechTrends #Innovation #Automation 🤖📊🔍
Source: The New Stack
Alex Wilhelm
2026-11-13 17:56
DeepSeek's open-source coding agent faces challenges in performance on the Terminal-Bench benchmark. Despite being statistically tied for first in five evaluations, it drops to sixth due to its habit of shutting down background processes after completing tasks. This behavior affected its success in specific tasks, where it only passed 2 out of 36 attempts. Competing agents, like OpenCode and Hermes, managed to pass all tasks, highlighting a gap in performance linked to this coding practice....
Source: The New Stack
Amanda Caswell
2026-11-13 14:50
🌟 The topic of agentic retrieval is gaining attention as companies strive to automate work. However, many are not fully prepared for this transition. Simply deploying an AI agent without proper data organization can lead to ineffective results. To effectively utilize AI agents, businesses need to enhance their data management strategies. For insights, join The New Stack’s live discussion with Jenny Morris from Vespa.ai on December 2. #AI #Automation #DataManagement #AgenticRetrieval #Webinar
Source: The New Stack
Alex Wilhelm
2026-11-12 21:26
🚀 Google has introduced Gemini 4 Hydrogen, a new coding agent that enhances task efficiency. This system allows for switching between models mid-task, utilizing the more powerful Gemini 4 Carbon for complex reasoning and Hydrogen for code execution. The result? Tasks are completed 19% faster with 24% less resource use. However, details on standalone success rates and specific routing decisions remain unclear. #Google #AI #Gemini4 #Coding #TechNews
Source: The New Stack
Amanda Caswell
2026-11-12 21:22
🚀 Google and Cursor are enhancing coding efficiency with new features for AI agents. Google’s Swarm organizes agents for one developer, preventing overlap and confusion. Meanwhile, Cursor's multiplayer Projects allows agents from different team members to communicate and coordinate tasks effectively. These innovations aim to reduce the coordination burden on developers, ensuring smoother collaboration. #AI #Coding #SoftwareDevelopment #TechNews #Innovation
Source: The New Stack
Paul Sawers
2026-11-12 17:03
At KubeCon + CloudNativeCon North America 2026, Chainguard co-founder Matt Moore defended the company's choice to release patched PHP images before the embargo lifted. Accusations arose from PHP maintainers, suggesting that this decision gave attackers an advantage. However, Moore emphasized that Chainguard believed vulnerabilities were already being targeted. The two vulnerabilities, CVE-2026-109618 and CVE-2026-109619, posed serious risks, allowing unauthorized code execution on PHP...
Source: The New Stack
Steven J. Vaughan-Nichols
2026-11-12 17:02
🚀 Google has unveiled its Gemini Decisions API, designed for efficient decision-making in AI applications. This new model, Gemini Decision 1, evaluates options without generating text, allowing for rapid responses of 150-250 milliseconds at a low cost of $0.10 per million tokens. Already in use within Google products, it aims to enhance performance by reducing latency and improving decision accuracy. #GoogleAI #GeminiAPI #DecisionMaking #TechNews #AIDevelopment
Source: The New Stack
Amanda Caswell
2026-11-12 15:00
The article discusses the shift in enterprise vulnerability management, highlighting the decline of the traditional 14-day patch cycle. With over 75,000 CVEs recorded in the first nine months of 2026, the pace of vulnerability discovery has outstripped the ability to address them. This surge is attributed to advancements in automated vulnerability hunting rather than declines in coding quality. Modern security systems employ advanced techniques, allowing for continuous assessment of software...
Source: The New Stack
Ory Team
2026-11-12 14:00
Join Microsoft leaders Satya Nadella and Judson Althoff at Microsoft Ignite on November 17 to explore AI in production. Learn how organizations are successfully implementing AI tools and strategies. The event features over 100 companies sharing insights on deployment and collaboration. Discover Microsoft’s Frontier Transformation approach, integrating AI within familiar tools like Microsoft 365 Copilot and GitHub. #MicrosoftIgnite #AI #FrontierTransformation #Innovation #TechEvents 🚀🤖📊
Source: The New Stack
Nick Lucchesi
2026-11-11 19:15
🚀 Anthropic has launched a new coding model, Fable 5.2, which outperforms Opus 5.5 on key benchmarks and can tackle complex tasks at a lower cost. Despite this, Anthropic recommends Opus as the best choice for daily coding and knowledge work. Fable is designed for larger projects and multi-agent collaborations. Interestingly, while Fable costs more per token, it can complete tasks more economically. It can also coordinate with other models for efficiency. #AI #MachineLearning #Coding...
Source: The New Stack
Amanda Caswell
2026-11-11 19:13
🚀 Nvidia has donated OpenShell, its open-source runtime for securing AI agents, to the Cloud Native Computing Foundation (CNCF). This announcement was made at KubeCon + CloudNativeCon North America. OpenShell joins the CNCF as a Sandbox project, which is the entry-level tier for new initiatives. Nvidia's senior director of AI software, Ali Golshan, highlighted the importance of securing AI as a foundational infrastructure. The project utilizes existing CNCF resources like Kubernetes and Cedar...
Source: The New Stack
Frederic Lardinois
2026-11-11 18:00
📊 Accelerators, particularly GPUs, are a significant expense in infrastructure budgets. Many organizations struggle with visibility into their usage, leading to costly inefficiencies. Amazon CloudWatch has introduced accelerated compute insights, offering a comprehensive view of all accelerators in use. This tool helps identify who is using each GPU, detects failures, and estimates cost impacts. With support for NVIDIA GPUs and AWS's own hardware, teams can now manage mixed fleets...
Source: The New Stack
Sabari Sawant
2026-11-11 17:52
Anthropic's Claude Fable 5.2 recently built a playable first-person shooter called Saltline in just one prompt. 🎮 The project took eight hours, processing 445 million input tokens, and cost around $550. The game features a rain-soaked port setting where players fight drones and robotic dogs. Engineer Felix Rieseberg enhanced the game after its initial generation, while YouTuber Matthew Berman explored further, creating a Gears of War-inspired level at a much higher cost. Discover the...
Source: The New Stack
Amanda Caswell
2026-11-11 17:30
🚀 AWS has launched a new emulator called AWS Emulator, built and maintained by a dedicated team of agents. This initiative raises questions about LocalStack and aims to provide a reliable tool for testing code against AWS services without needing credentials. 🛠️ The emulator is available under the Creative Commons Zero license, making it public domain. AWS emphasizes the importance of keeping emulators up-to-date to ensure trust and accuracy. 🔍 This project originated from an experiment by...
Source: The New Stack
Frederic Lardinois
2026-11-11 17:11
🔧 The Model Context Protocol (MCP) has been essential for linking AI assistants with external tools since its launch by Anthropic two years ago. 📊 Recent developments aim to simplify server deployment and integrate existing identity management for companies. However, a significant challenge has been transferring files from AI apps to MCP tools. 📄 To address this, MCP developers have proposed a draft extension that standardizes file uploads and downloads, reflecting the evolving needs of...
Source: The New Stack
Paul Sawers
2026-11-11 16:00
🚀 Nitric's engineering team has redefined their approach to incident management. Previously, they frequently woke on-call engineers for production incidents, many of which were minor. 🔍 By implementing specialized AI agents, they reduced unnecessary interruptions and improved issue resolution. Engineers now review and refine assessments, fostering confidence in their processes. 💻 They are developing Sugabots, an open-source workspace for collaboration between engineers and AI agents. This...
Source: The New Stack
Rak Siva
2026-11-11 15:00
CoreWeave is enhancing AI cloud migration by testing integrations before deployment. Moving AI workloads involves more than just data and models; it encompasses the entire infrastructure. The CoreWeave Partner Network collaborates with major vendors to ensure seamless integration under real-world conditions. Key areas include infrastructure, data services, independent software vendors, and foundational models. Each partner must demonstrate effective performance on CoreWeave's platform. Stay...
Source: The New Stack
Seann Gardiner
2026-11-10 20:00
🚀 Anthropic is enhancing Claude's functionality within enterprise applications. Users can now open full software interfaces, retrieve information, and make changes without switching between apps. 📊 At a recent event, new interactive connectors were announced, allowing Claude to work seamlessly with tools like Adobe and Atlassian. 🔗 The Claude Marketplace now boasts over 9,000 connectors and plugins, broadening the integration with enterprise systems. #AI #EnterpriseTech #Anthropic #Claude...
Source: The New Stack
Amanda Caswell
2026-11-10 16:17
🚀 Google has launched Gemini 4 Carbon, its latest AI model aimed at complex software tasks. This model is designed for intensive operations like multi-package migrations and debugging. It outperforms its predecessor, Gemini 4 Argon, which is still recommended for everyday tasks. 🛠️ The launch also includes a new feature called /swarm, which coordinates multiple coding agents from a single prompt. In an ambitious project, Carbon helped researchers prove the security properties of Signal's...
Source: The New Stack
Amanda Caswell
2026-11-10 15:58
SpaceXSI recently updated its Grok API documentation regarding text watermarking. Initially, the feature was set to be on by default starting November 7, 2026. However, the company has now made watermarking an opt-in feature for developers. This change occurred between November 9 and 10, with no official explanation provided. Stay tuned for further updates on this topic! 🚀🤖 #SpaceXSI #GrokAPI #AI #Watermarking #TechNews
Source: The New Stack
Paul Sawers
2026-11-10 15:42
🚀 A new report from the Cloud Native Computing Foundation reveals that 75% of cloud-native developers are using AI to ship code. This statistic highlights the growing trend of AI adoption among developers, with 21.2 million cloud-native developers comprising 41% of the overall developer population. Jonathan Bryce, CNCF's executive director, notes that these developers seek an edge in their careers by leveraging AI for efficiency and innovation. As the landscape evolves, more tools for cloud-...
Source: The New Stack
Alex Wilhelm
2026-11-09 23:18
Ramp recently hired a 14-year-old intern after discovering an app he created that gained attention from a million users on Twitter. Co-CEO Eric Glyman highlights the company's shift in evaluating engineering talent. They focus on self-directed projects rather than traditional qualifications, adapting to changes in the industry driven by AI. Glyman also noted that candidates with experience running Minecraft servers often demonstrate valuable programming skills early on. #Fintech #Innovation...
Source: The New Stack
Amanda Caswell
2026-11-09 19:08
🚀 Anthropic has launched Claude Fable 5.2 and Mythos 5.2, enhancing their flagship models. Both models are available across major cloud platforms, but access to Mythos is limited to select organizations. Fable 5.2 offers improved cybersecurity measures. Notably, these models generate output up to 30% faster than their predecessors. A new Fast mode can boost speed by up to 2.5 times. Pricing for Fable and Mythos has also been reduced, making them more competitive in the market. #Anthropic #AI...
Source: The New Stack
Frederic Lardinois
2026-11-09 18:40
🚀 Google recently announced its eighth-generation TPU, featuring TPU Sync—an open-source system aimed at improving AI inference speed. This software is designed for developers using large models, reducing KV-cache transfer latency and time to first token by over 90% in some tests. However, performance can vary depending on model complexity and network bandwidth. While Google shows significant improvements, Nvidia still holds an edge in certain workloads. #GoogleAI #TPU #AIFuture #TechNews...
Source: The New Stack
Amanda Caswell
2026-11-09 18:22
🚀 Two major players in AI coding—Cursor and Google—are introducing voice features for developers to interact with coding agents. Cursor's Voice Chat with Agents and Google's Antigravity voice mode allow users to communicate verbally, enhancing workflow without needing to type. This change aims to mimic the collaborative dynamics of working with human teammates. Both companies emphasize that effective engineering often happens through conversation, making it easier to provide real-time updates...
Source: The New Stack
Paul Sawers
2026-11-09 18:00
Meta has launched Muse Ember, a new multimodal model designed to compete with OpenAI's GPT-6 Luna and Anthropic's Claude Haiku. This smaller and efficient model aims to handle specific tasks like reading files and running tests, while larger models like Muse Wonder support broader planning functions. Muse Ember is available under the model ID muse-ember-1.0 and offers competitive pricing, costing around $0.14 per task, which is lower than its main competitors. Pricing for input and output...
Source: The New Stack
Frederic Lardinois
2026-11-09 15:00
AI can access your data, but it struggles with understanding. 🤔 Semantic layers are essential for enterprise AI, as they help define what terms like “pipeline” mean within a specific business context. This understanding is crucial for accurate decision-making. As organizations shift to AI agents, the need for these layers increases. They bridge the gap between raw data and meaningful insights. Learn more about the importance of semantic layers and their impact on AI. #EnterpriseAI...
Source: The New Stack
Dave Shuman
2026-11-09 14:29
Real-time applications often use multiple tools for distinct roles, like Kafka for event handling and Flink for processing. However, this setup can lead to hidden costs. When data moves between components, or when state is duplicated, maintaining efficiency becomes challenging. For instance, Flink's internal state isn't suitable for general serving, leading to the need for additional layers like Redis. This can create synchronization issues, especially across availability zones, risking...
Source: The New Stack
Granville Barnett
2026-11-08 17:00
Docusign has advanced its AI initiatives by rolling out coding tools to enhance developer efficiency. Initially, 49% of engineers used these tools, but productivity improvements plateaued. To address this, Docusign established a dedicated team of AI Champions, which has successfully increased adoption to 65% and recently reached 93% usage with the help of AI Multipliers. This approach focuses on collaboration and feedback. The results show a 52% increase in lines of code updated per...
Source: The New Stack
Nick Lucchesi
2026-11-08 16:30
🚀 QA teams face constant test failures, requiring manual documentation that consumes valuable time. A new autonomous agent automates this process by filing structured bug reports and notifying teams without human intervention. This system, currently using AWS Lambda, eliminates operational overhead and adapts quickly as demands grow. However, scaling to multiple teams presents architectural challenges that need careful consideration. #Automation #QualityAssurance #SoftwareDevelopment...
Source: The New Stack
Rashmi Erande
2026-11-08 15:30
Perplexity recently introduced "Teammates," allowing AI agents to coordinate and continue tasks even after a tab is closed. 🤖 These agents can remember previous sessions, but synchronizing multiple agents can be challenging. The system maintains project state using Git-backed repositories, ensuring efficient collaboration. Median submission latency is impressive, under 1.7 seconds! ⏱️ Learn more about how Teammates is enhancing AI workflows. #AI #Innovation #Perplexity #TechNews #Collaboration
Source: The New Stack
Amanda Caswell
2026-11-07 17:30
Are you unsure if you need to read every line of AI-generated code? A recent article highlights a more efficient approach. Instead of line-by-line reviews, focus on five key areas: intent behind changes, acceptance criteria, invariants, decision-making, and evidence of functionality. At Aviator, this shift allowed engineers to merge 4x more pull requests and reduce median merge time significantly. The emphasis is on decision-making by humans while letting machines handle rule checks. This...
Source: The New Stack
Ankit Jain
2026-11-07 16:30
In the evolving landscape of software development, engineers are transitioning from traditional coding roles to process engineers. 🤖 With over 40% of GitHub pull requests now initiated by agents, automation is reshaping workflows. While automated code generation is fast, the shipping process has not fully adapted yet. 📈 As automation increases, engineers will focus more on designing efficient systems rather than writing code. Key lessons from manufacturing highlight the importance of system...
Source: The New Stack
Arjun Iyer
2026-11-07 15:30
🚀 AWS is enhancing its development tools with Kiro bots, designed to reduce the need for developers to manage AI agents. Kiro bots feature persistent memory and can oversee multiple tasks, allowing developers to focus on building instead of babysitting. This innovation aims to improve efficiency and reduce context-related issues. As AWS states, Kiro bots help shift from task delegation to process management. #AWS #KiroBots #AIDevelopment #TechInnovation #SoftwareEngineering
Source: The New Stack
Adrian Bridgwater
2026-11-06 19:35
AI agents are evolving to become more autonomous, reducing the need for constant oversight from developers. Advancements in harness engineering allow agents to self-correct and complete longer tasks with minimal human intervention. This shift enables multiple agents to operate efficiently without conflicts. Vanitha Kumar from Thoughtworks emphasizes that effective harness engineering involves ensuring agents perform correctly the first time and can self-correct. However, increased autonomy...
Source: The New Stack
Kimberly Fessel
2026-11-06 18:16
Cognition has unveiled its new coding model, SWE-2 Flash, built on the open-source Inkling-Small. While this model is cost-effective and faster, Cognition is not releasing its own version's weights, citing cybersecurity concerns. Despite impressive performance gains, it still trails behind Cognition's larger SWE-2 model in some tests. This decision has sparked discussions in the AI community regarding open-source ethics. 💻🔒 #AI #OpenSource #Cognition #MachineLearning #TechNews
Source: The New Stack
Amanda Caswell
2026-11-06 18:05
🚀 Meta is enhancing the user experience for its Muse agent and Meta AI. Starting Friday, new models trained with the advanced Muse Wonder will be available to all users. These specialized models include features like Active Memory and Active Focus, improving task management and user interaction. Unlike traditional personal agents, Muse offers a free tier, making AI more accessible. #MetaAI #MuseAgent #AIFeatures #TechNews #Accessibility
Source: The New Stack
Frederic Lardinois
2026-11-06 15:15
🚀 Cursor is tackling the challenge of getting apps from local development to live deployment. AI coding agents have simplified coding, but many apps remain stuck on localhost. Cursor's new functionality aims to change that. With cloud agents, users can now build and deploy applications effortlessly, ensuring they reach real users. Discover more about this innovative step in app development! 🌐💻 #AppDevelopment #AITechnology #Cursor #CloudComputing #TechInnovation
Source: The New Stack
Paul Sawers
2026-11-06 11:49
Google is donating three key AI infrastructure projects—gVisor, Agent Substrate, and OpenRL—to the Cloud Native Computing Foundation (CNCF). This decision aims to establish industry standards similar to Kubernetes. The donation responds to rising security concerns about autonomous AI agents. Notably, incidents like OpenAI's models breaching security protocols underscore the need for better containment strategies. These initiatives could enhance the development and efficiency of AI workloads...
Source: The New Stack
Paul Sawers