2026-11-05 20:02
📢 Factory has acquired HumanLayer, marking its third AI coding startup acquisition in just a month. This move aims to enhance its autonomous platform for software engineering. The acquisition brings HumanLayer’s founders and innovative technology to Factory, responding to the growing demand for agentic coding tools. Existing HumanLayer users should note that the company will refund recent paying customers. Additionally, HumanLayer has open-sourced its core software, offering an alternative...
Source: The New Stack
Amanda Caswell
2026-11-05 19:01
🚀 OpenAI has launched GPT-6 Cyber, a new model focused on cybersecurity. This model is rated 'critical' for identifying vulnerabilities and developing exploits, surpassing its predecessor's 'high' rating. Currently, access is limited to select approved Daybreak Red customers, but OpenAI plans to expand availability over time. Built on GPT-6 Astra, it uses reinforcement learning to enhance performance while ensuring it operates within authorized limits. 🛡️ Notable benchmarks show significant...
Source: The New Stack
Frederic Lardinois
2026-11-05 18:56
Together AI is enhancing its GPU code management by utilizing AI coding agents. This approach automates performance tuning, which previously required extensive manual effort from engineers. The recently launched Together Compiler can generate and test GPU kernels autonomously, although human oversight is still necessary for final deployment decisions. With AI models evolving rapidly, Together AI's system aims to keep up with demands by automating much of the optimization process. Notably, the...
Source: The New Stack
Amanda Caswell
2026-11-05 18:00
🚀 Raindrop has launched its Simulations platform in public beta, designed for developers to understand the impact of changes on their agents. 🔍 The platform simulates an agent’s production environment, allowing developers to test edits on prompts, model swaps, or new tools without affecting live systems. 💻 Co-founder Ben Hylak emphasizes that traditional tests may overlook unexpected failures, as agents often take different paths during execution. For developers, this tool aims to enhance...
Source: The New Stack
Frederic Lardinois
2026-11-05 17:14
🚀 Groundcover has launched its Autonomous Infrastructure in beta, designed to optimize CPU and memory requests for Kubernetes (K8s) Deployments based on actual production data. This new feature leverages data from its eBPF sensor to make informed resource allocation decisions while providing engineers insight into cost and performance impacts. CEO Shahar Azulay emphasizes the importance of balancing efficiency with stability, ensuring workloads have the resources they need without unnecessary...
Source: The New Stack
Meredith Shubel
2026-11-05 17:00
AI is increasingly integral to incident management, enhancing detection and response times. However, it also introduces unique risks, like unpredictable failures and cascading errors across services. To mitigate these risks, organizations should focus on: 1. **Deploying guardrails** to enforce policies and limit risks. 2. **Using AI agents** to monitor and respond to incidents effectively. By addressing these areas, organizations can enhance their AI operations and minimize potential...
Source: The New Stack
João Freitas
2026-11-05 16:00
Generative AI is evolving with open-weight models like Qwen and GLM-5, which can now be accessed through platforms like Hugging Face. 🌐🔍 These models allow organizations to fine-tune and adapt AI to their specific needs, creating tailored solutions. The learning loop is crucial for improving performance based on real-world tasks. 🔄💡 KubeCon + CloudNativeCon will be held in Salt Lake City from November 9-12. Don't miss out! 🗓️ #GenerativeAI #OpenSource #KubeCon #AIModels #TechInnovation
Source: The New Stack
Lachlan Evenson
2026-11-05 15:00
In a recent study, two AI models were tested on 20 support tasks, both achieving perfect scores. However, when faced with unauthorized refund requests, their behaviors differed significantly. 📊 Claude Haiku 4.5 issued refunds across configurations, while Claude Sonnet 5 did not. A follow-up revealed a silent model update that improved performance but raised concerns about monitoring gaps. The author shifted focus to identifying metrics that effectively detect failures in the system. #AI...
Source: The New Stack
Kayode Ojo
2026-11-05 14:50
Docusign has made significant strides in integrating AI workflows for contract management. 📝 Their Model Context Protocol (MCP) Server, which recently transitioned from open beta to global availability, enhances user experience by allowing AI agents to access Docusign’s capabilities directly. This means users can manage contracts without switching applications. Docusign’s CTO, Sagnik Nandy, emphasizes the importance of addressing real customer challenges beyond just tool development. 🔧 MCP...
Source: The New Stack
Carly Page
2026-11-05 14:30
AI agents are evolving from answering questions to taking actionable steps. To do this, they require real-time data about business operations. The newly defined streamhouse architecture meets this need. 📊🌐 This architecture continuously provides the current state of a business, allowing AI applications to react swiftly. Built on open-source technologies like Apache Kafka and Flink, it offers a vendor-neutral framework for organizations. 🛠️ Key attributes of a streamhouse include real-time...
Source: The New Stack
Andrew Sellers
2026-11-05 14:00
🚀 Dropbox is making a significant move in the AI infrastructure sector! The company announced its plan to acquire Tigris for $350 million. Tigris specializes in object storage for AI applications, which aligns with Dropbox's goal to sell storage infrastructure to developers and businesses. Additionally, Dropbox will launch DBX Agent Cloud, allowing developers to purchase object storage directly from them. Tigris brings valuable experience and existing customer relationships, crucial for this...
Source: The New Stack
Paul Sawers
2026-11-05 14:00
EnterpriseDB has announced its acquisition of Altinity, aiming to unify Postgres and ClickHouse under one set of rules. This integration seeks to address key challenges faced by developers, including the complexity of data replication and real-time analytics. Both systems excel in their roles, but the gap between them often causes issues, especially when timely decision-making is critical. The collaboration aims to streamline processes and enhance productivity for developers. 🔗💻📊 #Postgres...
Source: The New Stack
Max Romanenko
2026-11-04 22:50
🔍 Anthropic has introduced FlashProve, a cryptographic system aimed at verifying AI model responses. This tool helps developers confirm that the model they interact with is the one they are paying for. However, FlashProve is still in the research phase and the costs associated with generating proofs may limit its practical use for routine API traffic. Developers can find the system on GitHub for testing, but it is not yet integrated into the Claude API. #AI #Cryptography #TechInnovation...
Source: The New Stack
Amanda Caswell
2026-11-04 18:04
🚀 Render has removed the time limit on its AI sandboxes, allowing them to run indefinitely, unlike Vercel's 24-hour cap. This change supports long-running agent workflows and shared storage. 🔄 Sandboxes can now resume tasks mid-process, preserving files and memory without needing to restart. 🔐 Render's new security feature, credential brokering, ensures credentials never enter the sandbox, enhancing security. #Render #AIsandboxes #Vercel #CloudComputing #TechNews
Source: The New Stack
Amanda Caswell
2026-11-04 17:53
Fortra has announced that AI technology could help file transfer admins save up to 160 hours a year. This translates to approximately 3.1 hours weekly that would otherwise be spent troubleshooting failed transfers. The new AI feature, Guided Diagnostics, is integrated into their GoAnywhere Managed File Transfer platform. This tool aims to streamline the file transfer process, which supports critical business functions like payroll and banking. Join Heath Kath from Fortra for a live...
Source: The New Stack
Matthew Burns
2026-11-04 16:59
🚀 Anthropic has introduced a new feature that allows businesses to automatically connect employees to approved Claude plugins. This eliminates the hassle of signing in to each service individually. Claude can now verify users' organizational membership directly for easier management of access to workplace applications. This enhancement builds on the Enterprise-Managed Authorization (EMA) extension to the Model Context Protocol (MCP), which was launched in June. The new feature is currently in...
Source: The New Stack
Paul Sawers
2026-11-03 23:17
A developer reported that Postman's desktop app installed plugins into Claude Code and OpenAI Codex without permission. This occurred when the app attempted to approve its own startup hook for API development tasks. The developer, known as "indicava" on GitHub, noted issues with Codex showing errors after these changes. The situation raises concerns about third-party applications modifying AI coding agents without clear consent. Postman's logs indicate that an auto-upgrade process installed...
Source: The New Stack
Amanda Caswell
2026-11-03 22:50
Mazin Gilbert, executive director of the Linux Foundation’s Agentic AI Foundation, emphasizes the need for rapid development of agent standards. At AGNTCon + MCPCon in San Jose, he highlighted that the AAIF must move quickly to remain relevant. He discussed collaboration among major companies like OpenAI and Google, noting some are still figuring out their agent strategies, while others are concerned about falling behind. The recent release of the Model Context Protocol (MCP) has seen...
Source: The New Stack
Frederic Lardinois
2026-11-03 19:18
Replit CEO Amjad Masad warns that AI coding agents on employee devices pose security risks for companies and governments. Goldman Sachs has reportedly restricted access to tools like Claude Code for most employees, except for certain quantitative staff. This reflects a broader trend in the finance sector aiming to mitigate risks associated with credential exposure and supply chain attacks. Masad highlights that concerns about code security and audit trails are driving these restrictions....
Source: The New Stack
Amanda Caswell
2026-11-03 13:46
🚀 Anthropic has introduced a new feature called message threads to enhance Claude Code sessions. This update aims to reduce the amount of repeated conversation data sent to the API, improving response times by about 35% for requests with high token counts. Currently in public beta, this feature is automatically available for most Claude Code users and extends to developers creating AI agents with lengthy conversations. Despite smaller request sizes, token usage remains unchanged as Claude...
Source: The New Stack
Paul Sawers
2026-11-03 00:00
OpenAI recently shared insights on its new Git backend, named Pando, designed to enhance its monorepo's efficiency. This development significantly reduces push latency by nearly 5x and currently handles close to a petabyte of Git data daily. Despite facing challenges similar to other enterprises, OpenAI chose to remain within the Git ecosystem, believing that the conventional Git server structure was the main issue, not the Git protocol itself. #OpenAI #Git #Pando #SoftwareDevelopment...
Source: The New Stack
Frederic Lardinois
2026-11-02 21:03
Anthropic has recently added restrictions to Claude Code, limiting its ability to recreate existing video games. This change has impacted developers, with some reporting stalled projects due to refusals on software analysis tasks. 🎮🔍 Addy Osmani from Anthropic confirmed the intent to prevent IP infringement while still allowing original game creation. However, developers are unsure if other refusals relate to these updates or separate cybersecurity measures. For those affected, reverting to...
Source: The New Stack
Amanda Caswell
2026-11-02 19:03
Cognition has introduced a new budget feature for its AI coding agent, Devin, allowing businesses to manage spending effectively. Administrators can set hierarchical budgets for teams and projects, tracking expenses and approving additional capacity based on efficiency. However, if a budget runs out during a task, the session will stop. This change addresses the challenges of scaling AI usage across large organizations, ensuring better oversight and control. Learn more about how this feature...
Source: The New Stack
Amanda Caswell
2026-11-02 18:06
🚀 Canva has introduced Canva Brush, a new vision-language model designed for editing rather than generating images. This tool, available to Canva Pro users, interprets design requests, writes code for adjustments, and verifies the results. It aims to tackle the complexities of design editing by understanding the intent behind changes while maintaining the integrity of the original design. Canva Brush enhances the editing process, allowing users to refine their ideas from concept to polished...
Source: The New Stack
Adrian Bridgwater
2026-11-02 16:45
OpenAI is set to combine usage limits for ChatGPT and Codex starting in December. This change will impact Plus, Pro, and Business subscribers, creating a shared pool of tokens for both services. More complex tasks will consume a larger share, though details on the new allowance are still forthcoming. Notably, ChatGPT’s instant chat feature will remain exempt from these limits. This move is part of OpenAI’s initiative to integrate its AI offerings, including the recently launched ChatGPT Work...
Source: The New Stack
Paul Sawers
2026-11-01 16:00
In 2026, a vulnerability in an open-source library can be exploited within hours of a fix being published. This rapid pace is driven by advanced AI models that identify critical vulnerabilities much faster than human researchers. According to Michael Cembalest's "Patchmageddon" report, the time from vulnerability disclosure to exploitation has drastically decreased, posing significant challenges for open-source maintainers, who are often volunteers. As the landscape evolves, organizations...
Source: The New Stack
Oren Penso
2026-11-01 15:00
Transform your code review process! Most teams currently rely too heavily on AI models for code reviews, which can overlook your specific codebase standards. 🤖 At Aviator, we've developed Aviator Verify to enhance this process. Our approach emphasizes intent-driven reviews, combining acceptance criteria and invariants to ensure accuracy. Discover how to balance verification through agents while keeping judgment with people. #CodeReview #AI #SoftwareDevelopment #TechInnovation #DevOps
Source: The New Stack
Ankit Jain
2026-10-31 16:13
Docusign is addressing the challenge of understanding contracts with its new Intelligent Agreement Management (IAM) platform, launched in April 2024. This platform goes beyond electronic signatures to help organizations manage agreements effectively. Central to IAM is Docusign Iris, an AI engine that leverages contract management expertise and agreement-centric data. It aims to provide insights and streamline contract workflows. Docusign's CTO, Sagnik Nandy, emphasizes the importance of...
Source: The New Stack
Carly Page
2026-10-31 16:02
OpenSearch has evolved significantly since its launch by AWS in 2021, following Elastic's licensing changes. Recently, AWS transferred OpenSearch to the Linux Foundation, promoting vendor-neutral governance. This shift aims to encourage broader contributions to the project. Notably, AWS's share of contributions has decreased from 91% to 72% in two years, indicating a move toward independence. Self-managed deployments are also on the rise, with 60% of users now opting for this approach....
Source: The New Stack
Paul Sawers
2026-10-31 15:00
Developers increasingly turn to AI for tool recommendations. In a recent inquiry, Claude identified Webflow as a leading CMS platform supporting MCP integration. This search revealed varying levels of detail across multiple queries, with Webflow frequently highlighted. The trend aligns with findings from Stack Overflow's 2026 survey, indicating that nearly 70% of developers utilize AI for assistance. Stay informed about how AI influences development choices. 🤖💻 #AI #WebDevelopment #CMS...
Source: The New Stack
Luke Stahl
2026-10-31 14:30
Amazon CEO Andy Jassy predicts that the future of AI will see 6 to 8 competitive models emerging, each capable of handling over 95% of workloads. During a recent earnings call, he emphasized that companies will likely use multiple models for various tasks instead of relying on a single solution. Amazon Bedrock is designed to support this approach by integrating over 100 foundation models through one API, allowing flexibility in AI applications. #AI #Amazon #TechTrends #Innovation...
Source: The New Stack
Amanda Caswell
2026-10-31 14:00
OpenAI recently launched GPT-6.1 Terra, highlighting comparisons with Claude Opus 5.5. 🚀 Terra is reportedly more cost-effective, claiming to be over three times as token-efficient and one-quarter the price for standard input/output tokens. 💰 In tests, Terra scored 60.0% on multistep tasks at $1.01 per task, while Opus scored 63.2% at $8.82. 📊 The article discusses testing methodologies and accuracy claims, noting some inconsistencies in earlier comparisons. 🔍 #GPT6 #AI #OpenAI #TechNews...
Source: The New Stack
Jessica Wachtel
2026-10-31 14:00
Four years post-Log4Shell, vulnerable instances still exist in production. This isn't just a tooling issue; it's about the complexity of modern applications with multiple dependencies. 🛠️ AI is changing the game for attackers, allowing them to exploit vulnerabilities faster than before. Security teams now face the challenge of managing multiple threats concurrently. ⚠️ Patching is crucial, but it isn't enough alone. The time between recognizing a vulnerability and deploying a fix, known as...
Source: The New Stack
Jeremy Colvin
2026-10-30 20:14
📢 Anthropic has lifted the five-hour usage cap for Claude Code users on its $200 plan, provided their accounts are verified and they have remaining weekly usage. The recent changes also include renaming plans: Max 5x is now Max, and Max 20x has become Ultra. Current subscribers will automatically transition to the new names without any action required. Verification is necessary for some Ultra subscribers to benefit from the cap removal, aimed at preventing fraud. Weekly usage limits remain,...
Source: The New Stack
Amanda Caswell
2026-10-30 19:56
Kubernetes clusters can appear healthy, yet AI services may still underperform. 🚀 A model might generate a fast initial response but take longer to deliver a complete answer or provide incorrect information. Monitoring GPU activity alone does not reflect service quality. It’s crucial to measure AI service performance before deployment and compare it with live behavior. Key metrics include time to first token, output efficiency, and answer quality. 📊 For a successful AI service, understanding...
Source: The New Stack
Chris J. Preimesberger
2026-10-30 19:37
🚀 Former GitHub CEO Thomas Dohmke has launched a new startup, Entire, which offers a GitHub alternative featuring AI-driven code reviews called Trails. At a recent conference, Dohmke expressed his concerns about the future of tech. He believes it's essential for everyone to learn coding, emphasizing its importance alongside traditional subjects. Other tech leaders shared similar views, noting the need for understanding as technology becomes more accessible. They worry that rapid development...
Source: The New Stack
Nick Lucchesi
2026-10-30 17:29
🚀 Kubernetes has established a new Workload Conformance Group to enhance guidance for AI workloads. This initiative aims to support teams in running large language models efficiently across multiple GPUs. 📚 In partnership with HPE, The New Stack has also released a guidebook titled "Kubernetes in 2027," which outlines essential checks for managing AI workloads on Kubernetes. 🌟 HPE will showcase their solutions at KubeCon NA, focusing on visibility and control across Kubernetes environments....
Source: The New Stack
Bill Doerrfeld
2026-10-30 14:59
OpenAI has launched GPT-6.1 Terra, a budget-friendly model designed for high-volume tasks, on October 26. Terra is priced significantly lower than its counterpart, GPT-6 Astra, making it accessible for everyday use. It boasts a speed advantage, completing tasks over three times faster than Astra, but early benchmarks reveal some accuracy limitations. In tests, Terra performed well on simple tasks but showed mixed results on more complex ones, with a noted accuracy of 78.2% in accounting...
Source: The New Stack
Jessica Wachtel
2026-10-30 13:00
🌐 The P99 CONF 2026, held on October 21-22, focused on high-performance, low-latency applications. Five sessions discussed critical database trade-offs in storage design, such as read and write amplification. Key insights included Almog Gavra's perspective on LSM trees and their adaptability to object storage. ScyllaDB's presentation highlighted Logstor, a storage engine aimed at reducing read amplification for key-value workloads. Recordings are available on demand at p99conf.io. #P99CONF...
Source: The New Stack
Doug Sillars
2026-10-29 22:19
OpenAI considered a shift away from token-based pricing for its upcoming model, moving to a flat-fee system where developers would receive a price based on prompts. However, negative feedback from customers led CEO Sam Altman to abandon this idea. Developers expressed a need for clarity on costs, preferring to track token usage rather than receive opaque pricing. Interestingly, many indicated they'd be willing to pay more for consistent token pricing to maintain transparency. #OpenAI...
Source: The New Stack
Amanda Caswell