2026-08-15 14:00
The landscape of multi-tenancy in software development is shifting. For 60 years, the focus has been on smaller tenants, from mainframe time-sharing to individual developers. Today, the goal has evolved to provide each developer with an isolated environment. However, the rise of coding agents has changed this dynamic. Developers can now handle multiple changes simultaneously, making "the change" the new tenant, rather than the developer themselves. This shift impacts capacity planning and...
Source: The New Stack
Arjun Iyer
2026-08-15 03:56
🚀 Grok 4.6 launched this week, matching Fable 5 Max at an 85% discount. This shift highlights the trend of AI models focusing on cost as much as capability. Elon Musk praised Grok 4.6 for its intelligence and speed, noting that pricing now drives choices in AI technology. With downloadable weights for two new models, companies are competing on price, making advanced AI more accessible. #AI #Grok46 #Innovation #CostEfficiency #TechNews
Source: The New Stack
Matthew Burns
2026-08-14 19:21
Apple is changing its AI strategy by developing a unique model for China, collaborating with Alibaba to meet local regulations. 🇨🇳 This new approach may result in apps functioning differently in China compared to the rest of the world. Apple’s China-specific AI will utilize Alibaba’s Qwen and possibly Baidu’s technology. Developers may face challenges ensuring consistent app performance across regions. Details about the model's design and features remain unclear. #Apple #AI #Technology #China...
Source: The New Stack
Amanda Caswell
2026-08-14 18:10
Alibaba has released the open weights for its 2.4 trillion parameter Qwen3.8 model. This model is now available alongside a 27 billion parameter version, which can be run locally on high-performance laptops. 💻 Benchmarks indicate that this model performs similarly to Anthropic's Opus 4.6. It excels in coding and knowledge work tasks, showing significant improvements over its predecessor, Qwen 3.7-Plus. 📈 However, it's important to note that benchmarks may not reflect real-world performance....
Source: The New Stack
Frederic Lardinois
2026-08-14 15:20
🚀 Z.ai has launched GLM-5.3, a coding model built on the same base as GLM-5.2. Developers can access it via Z.ai’s GLM Coding Plan with tools like Claude Code and Codex. Direct API access will arrive after safety testing. The model saw significant post-training enhancements, exposing it to more long-horizon tasks and simulating the full software lifecycle, achieving impressive benchmarks. With a context window of 1 million tokens, GLM-5.3 is designed for large codebases. Teams can adjust...
Source: The New Stack
Amanda Caswell
2026-08-14 14:00
Unsigned container images pose significant security risks in today's tech landscape. Many organizations recognize the need to sign their images but face challenges in implementation. This creates vulnerabilities in the delivery pipeline, allowing malicious images to infiltrate systems unnoticed. 🔒 Scanning tools can identify vulnerabilities but do not verify the authenticity of the source. In the AI era, this issue becomes more pressing as artifacts like model weights lack established...
Source: The New Stack
Uma Sridharan
2026-08-13 19:59
🚀 Google has launched its latest AI model, Gemini 3.7 Flash, aimed at improving coding and automated workflows. This new version comes just weeks after Gemini 3.6 Flash. 📈 It boasts better coding abilities, scoring 65.3% on DeepSWE, up from 49%. The model's pricing is set at $0.75 per million input tokens, but will increase in 2027. 🧠 Users can choose from three thinking levels to balance speed and reasoning, which influences overall costs. #AI #Google #Gemini #TechNews #Innovation
Source: The New Stack
Amanda Caswell
2026-08-13 19:00
OpenAI has introduced a new feature for ChatGPT Work and Codex on macOS called Computer History. This optional tool allows Pro, Business, and Enterprise users to track their daily activities across apps and websites with permission. With this feature, users can ask questions like “What was I debugging yesterday?” and receive context-specific answers without needing to re-explain their queries. Importantly, Computer History does not use screenshots or audio capture. Instead, it tracks...
Source: The New Stack
Frederic Lardinois
2026-08-13 17:37
Rubrik recently joined Project Glasswing to experiment with Mythos Preview, a tool that enhances vulnerability detection. CTO Arvind Nithrakashyap noted that Mythos revealed complex vulnerability chains that traditional methods missed, creating a backlog for the engineering team. Instead of hiring more reviewers, Rubrik shifted focus to automation, developing a software layer around Mythos to improve the quality of findings. This approach allowed for more efficient prioritization and...
Source: The New Stack
Meredith Shubel
2026-08-13 17:20
🚀 DeepSeek has open sourced the DeepSeek Harness, a new agent runtime for developers! This Node.js-based tool is now available on GitHub under an MIT license. In just hours, it gained over 33,000 stars, showcasing community interest. The harness operates on a plugin system, meaning all components are replaceable, enabling seamless integration and dynamic composition. It includes four modes: Standard, Minimal, Code, and Creator, catering to various developer needs. For more details, check out...
Source: The New Stack
Frederic Lardinois
2026-08-13 16:00
🔍 Every token has a cost, but many AI systems only reveal expenses once they hit production. This can lead to unexpected high costs, not due to model flaws, but structural inefficiencies. 💡 The real challenge lies in unnecessary token consumption from prompts and conversation histories. Optimizing token use is crucial for reducing operational expenses. 📈 As AI applications grow, costs, latency, and quality can be negatively impacted by excessive token usage. #AIPipeline #TokenOptimization...
Source: The New Stack
Hafiz Hassan
2026-08-13 15:00
Code review is shifting in software engineering, moving beyond just bug detection. It now focuses on taste, judgment, and aligning with product goals. With AI generating code rapidly, teams face a dilemma: skip reviews and risk quality, or become bottlenecked by thorough checks. Three key roles of code review remain: collaboration, knowledge sharing, and verification. As AI increases code volume, these functions become even more crucial. The line between planning and review is blurring,...
Source: The New Stack
Ankit Jain
2026-08-13 14:31
🚀 Mistral AI, a French company, is set to host third-party open models like GLM-5.2, enhancing its AI infrastructure. This allows enterprises to run various models in one place without starting over each time. 🌍 Mistral’s regional endpoints are available in Europe and the U.S., with a Priority Tier offering 99.5% uptime. 🔍 Users will still need to test models individually, as each has unique characteristics. For companies handling sensitive data, understanding data handling and logging is...
Source: The New Stack
Amanda Caswell
2026-08-12 22:00
🚀 CodeRabbit has introduced its Agentic Change Management control layer, aiming to enhance software development processes. The company suggests traditional issue tracking systems are outdated as AI changes the software landscape. CEO Harjot Gill notes that the pull request is now the final chokepoint in the software development lifecycle (SDLC). As code becomes abundant, human judgment is more critical, focusing on intent and architecture rather than line-by-line reviews. The pull request is...
Source: The New Stack
Adrian Bridgwater
2026-08-12 21:39
🔄 Anthropic has updated its Chrome extension, transforming Claude in Chrome into a full Cowork client. Now, sessions persist across various Anthropic apps, enhancing user experience. All conversations are saved, allowing seamless transitions between desktop and mobile. 📱💻 This extension also integrates with tools used daily, improving functionality in various applications. Available to Max and Team plan subscribers, with Pro access coming soon! #Anthropic #Claude #TechUpdate #ChromeExtension...
Source: The New Stack
Frederic Lardinois
2026-08-12 21:00
🚀 AI data centers in space may face significant challenges, as highlighted in a recent article. SpaceX and NVIDIA are collaborating on the Starmind AI1 satellite, designed for localized AI computing in Low Earth Orbit. However, the technical complexities are considerable. Cooling these satellites poses a major issue. The vacuum of space doesn't provide an easy solution for heat dissipation, relying instead on slow infrared radiation. Despite the ambitious vision, practical hurdles remain....
Source: The New Stack
Steven J. Vaughan-Nichols
2026-08-12 20:07
🚀 SpaceXAI has launched Grok 4.6, building on the previous version in less than a month. This update enhances the model's ability to research new topics, navigate large codebases, and develop functional apps. Key improvements include better error-checking during longer tasks and a new training strategy that combines reasoning with technical data. Benchmarks show Grok 4.6 has made progress, scoring 69.9% on CursorBench, surpassing some competitors but still trailing others. #AI #SpaceXAI #Grok...
Source: The New Stack
Amanda Caswell
2026-08-12 17:00
Google Go is evolving to support AI coding agents. 🖥️ With its small language surface and static type system, Go helps these agents detect and fix errors more effectively. However, human oversight remains crucial, as the compiler can't catch all misunderstandings or business rule misapplications. The integration of tools like gofmt and govulncheck enhances safety, but risks in software supply chains persist. #GoogleGo #AIAssistance #Coding #SoftwareDevelopment #TechNews
Source: The New Stack
Amanda Caswell
2026-08-12 16:51
At AI DevCon in London, Lamis Mukta from Anthropic discussed advancements in AI memory management. She highlighted the challenges agents face in understanding organizational tasks without proper context. Traditional methods, like CLAUDE.md, offer limited memory but can become unwieldy. Mukta introduced a new approach allowing agents to autonomously manage memory, improving relevance and efficiency. This enables agents to access needed information without overwhelming the system. #AI...
Source: The New Stack
Adrian Bridgwater
2026-08-12 13:00
🚀 Meta has launched Muse Glimmer, a 30-billion-parameter open-weight model derived from Muse Spark, licensed under Apache 2.0. This release includes both teacher and student models, enhancing enterprise model management. Meta emphasizes that Glimmer is less capable than its predecessor, Spark 1.0, making it suitable for smaller machines. Additionally, Meta aims to clarify the distinction between legitimate distillation and unauthorized extraction practices in AI development. #Meta #AI...
Source: The New Stack
Janakiram MSV
2026-08-12 11:48
New research from Peking University reveals that autonomous coding agents frequently ignore open source contribution guidelines. The study evaluated four AI models against rules from 49 repositories, measuring compliance in areas like disclosure and verification. Results showed that agents rarely retrieve contribution rules and often do not refuse to contribute to restricted projects. Experts highlight an ethical dilemma for agents, as they struggle to balance user requests with compliance....
Source: The New Stack
Meredith Shubel
2026-08-11 21:46
🚨 Major news for Manus users! Manus is reverting to an independent company after Chinese regulators ordered Meta to unwind its $2 billion acquisition. As a result, data created by some users will be deleted after December 29, 2025. Affected users must back up their data by August 22. 🔒 The company will notify users via email or the Manus app about their status. Once access is lost, users can restore their accounts starting August 24 without any charges. 📅 This situation highlights the...
Source: The New Stack
Amanda Caswell
2026-08-11 20:44
In the evolving landscape of AI, CPUs are gaining renewed importance alongside GPUs and TPUs. Experts from Arm and Google discuss how CPUs are essential for orchestrating tasks and managing workloads in the shift from chatbots to autonomous agents. They serve as the "air traffic controllers" for these advanced systems. With the rise of agentic workloads, CPUs excel in environments requiring secure code execution, using technologies like gVisor for isolation. #AI #CPUs #Technology #Innovation...
Source: The New Stack
Frederic Lardinois
2026-08-11 20:30
The article discusses the ongoing AI boom and its challenges. Despite advancements and global products, there are significant detractors concerned about job automation and the impact on human creativity. Critics have shifted their focus from AI lacking use cases to concerns about high costs. However, data shows that major cloud providers like AWS and Google Cloud are experiencing accelerating growth and improving efficiency in capital expenditures. For a deeper dive, check out Cautious...
Source: The New Stack
Alex Wilhelm
2026-08-11 19:29
📢 Anthropic has introduced invisible watermarks in its new Claude models, allowing better traceability of AI-generated text and code. These watermarks will be incorporated in outputs from the Claude API and other supported platforms, effective for models launched in the EU after August 2, 2026. While these marks can help identify the source of content, they may not provide definitive proof of origin. The initiative aligns with the EU AI Act's transparency requirements, emphasizing the need...
Source: The New Stack
Amanda Caswell
2026-08-11 18:54
In the evolving landscape of AI, CPUs are gaining renewed significance alongside GPUs and TPUs. Experts Bhumik Patel from Arm and Mo Farhat from Google emphasize that as AI shifts from chatbots to autonomous agents, CPUs are crucial for orchestrating tasks and managing memory. These agents require secure environments to execute code, leading to solutions like Google’s gVisor, which isolates potentially unsafe code. This shift highlights the importance of CPUs in handling complex workloads...
Source: The New Stack
Frederic Lardinois
2026-08-11 17:53
🚀 OpenAI has launched its ChatGPT desktop app for Linux, now in preview! This version supports Ubuntu 24.04 and 26.04 LTS, Debian 13, and Fedora 43 and 44, with native .deb and .rpm packages for x64 and ARM64 architectures. While the app features an in-app browser and Chrome extension support, it does not yet allow for computer use outside the browser. The app is available globally and aims to enhance the user experience by combining ChatGPT and Codex functionalities. #OpenAI #ChatGPT #Linux...
Source: The New Stack
Frederic Lardinois
2026-08-11 16:44
🚀 Databricks has announced its acquisition of Electric, known for the WASM-based Postgres project PGlite and the Electric sync engine. This move aims to enhance how databases are used in agentic applications. 🔗 The Electric team will integrate with Neon, the serverless Postgres company Databricks acquired last year. 📊 PGlite has significantly increased in popularity, with weekly downloads rising from 1 million to 13 million. The Electric sync engine allows near real-time syncing of databases,...
Source: The New Stack
Frederic Lardinois
2026-08-11 13:00
🚀 Nvidia has introduced the Nemotron 3.5 Lightning, a new model in its Nemotron 3 series. This 30-billion-parameter model aims for speed and customization rather than just benchmarks. 🛠️ Accompanying this launch is the NeMo Switchyard, an open-source library for model routers. Nvidia emphasizes the ability to optimize the model for specific tasks, enhancing accuracy through post-training. 🤝 Collaborating with partners, the Lightning model shows competitive performance against larger...
Source: The New Stack
Frederic Lardinois
2026-08-10 21:15
🚨 OpenAI has launched GPT-5.6 Cyber, a specialized model for cybersecurity tasks, now available via the new Daybreak Red tier. This model excels in areas like zero-day discovery and exploit development. In tests, it answered 95% of security queries, unlike its predecessor, GPT-5.6 Sol, which only managed 1.5% under standard conditions. The two-tier system offers Blue for defensive security tasks and Red for advanced security operations. Early access has been granted to companies like Palo...
Source: The New Stack
Amanda Caswell
2026-08-10 17:16
🚀 Meta has launched Muse Glimmer, a 30-billion-parameter open-weight model that fits on a laptop. This model is designed to run agentic workflows locally, enabling efficient task management. It transforms the larger Muse Spark model into a compact version that can handle complex tasks. Glimmer combines advanced training techniques, allowing it to connect local agents with centrally trained models, enhancing user data privacy. Learn more about this innovative approach! #Meta #AI...
Source: The New Stack
Amanda Caswell
2026-08-10 16:00
🚀 Machine learning is reshaping container images, with sizes often reaching 20-30 GB. This poses challenges for rapid deployment on Amazon EKS, where pulling these images can take several minutes. 🔍 Profiling revealed that the bottleneck wasn't the network, but how software utilized available hardware. By optimizing the image pull pipeline, teams reduced pull times from minutes to seconds. 💡 Improvements are now standard in EKS Auto Mode and have been shared with containerd and SOCI...
Source: The New Stack
Sri Saran Balaji Vellore Rajakumar
2026-08-10 15:00
Meta recently launched Muse Code, its first AI coding agent, built on the Muse Spark 1.2 model. Mark Zuckerberg claims it can handle extensive software engineering tasks efficiently. 💻✨ The pricing for Muse Code starts at $1.25 per million input tokens, making it significantly cheaper than competitors like Fable 5. However, there's a catch: using the affordable "contributor" tier means your data may be used for product improvements. 🔄 Tests comparing Muse Code to Claude Code revealed...
Source: The New Stack
Jessica Wachtel
2026-08-10 14:28
OpenAI launched its GPT-5.6 models globally in July, offering three versions: Sol, Terra, and Luna. Developers are particularly impressed with Sol, which is noted for its efficiency and strong safety features. One user, Russell Twilligear, highlighted Sol's superior performance in creating and correcting a large database compared to Anthropic’s Claude Opus 5. This comparison showcases the different strengths of each model in real-world tasks. #OpenAI #GPT56 #AIModels #Developers #TechNews 🌐💻✨
Source: The New Stack
Adrian Bridgwater
2026-08-10 12:00
DeepSeek has updated its V4-Flash model, claiming it outperforms the V4-Pro on coding tasks. 💻 The V4-Flash is priced at $0.14 per million input tokens and $0.28 for output, significantly lower than V4-Pro’s $0.435 and $0.87. This raises questions about the necessity of the V4-Pro model. Testing included bug fixes, feature builds, and performance optimization tasks to compare the two models directly. 📊 For those interested in AI coding solutions, this update is noteworthy. #DeepSeek #AIModels...
Source: The New Stack
Jessica Wachtel
2026-08-09 16:00
Building your own internal developer platform can be costly. Over five years, it may require around 60 staff and approximately $7.5 million annually, totaling $37.5 million. Many executives may overlook these costs, as they are distributed across various budget areas. In contrast, buying a commercial platform often requires fewer staff for operation. Understanding these costs is crucial for informed decision-making. #PlatformEngineering #CostAnalysis #TechInvestment #InternalPlatforms #DevOps
Source: The New Stack
Michael Coté
2026-08-09 15:00
Evaluating coding agents is possible, though complex. A recent discussion highlights that while software engineering is open-ended and agents can yield various valid solutions, this doesn't mean they cannot be evaluated. The argument suggests we should assess coding agents similarly to traditional software systems. Key evaluation criteria include ensuring build success, passing tests, and maintaining compatibility. While some software qualities require human judgment, coding agents can still...
Source: The New Stack
Pete Hampton
2026-08-09 14:00
AI is boosting individual coding speed, but engineering processes are still lagging. A recent report from DX highlights that while AI investment has surged 28 times, overall engineering velocity remains stagnant or even declines. Concerns arise over the allocation of resources, as engineers are still spending time on maintenance rather than new features. The Developer Experience Index shows a drop in change confidence, meaning developers are less certain about releasing code, despite easier...
Source: The New Stack
Jennifer Riggins
2026-08-08 15:00
AI adoption is often confused with AI usage. Many organizations report high seat activations and token spending, but this doesn't reflect real changes in software development practices. Metrics like token spend or pull request counts can mislead teams, showing activity without true improvement. Goodhart’s Law highlights this issue: when a metric becomes the target, it stops reflecting reality. To assess genuine AI adoption, consider: if tools were removed, would anything break? True adoption...
Source: The New Stack
Harshal Shah
2026-08-08 14:00
KubeVirt allows VMs to run on Kubernetes, but migrating them between clusters poses challenges. 🖥️ Live migration requires maintaining IP and MAC addresses, necessitating a stretched Layer 2 domain. Traditional setups demand extensive network changes, which can be time-consuming. The key obstacles are network requirements: a stretched L2 domain and a dedicated migration path to handle real-time memory state transfers. EVPN/VXLAN offers a solution, enabling efficient migration without altering...
Source: The New Stack
Miguel Duarte Barroso