Articles from Source: The-New-Stack

How to build an AI-powered private document search app with RAG, ChromaDB, and memory

2026-04-10 16:00
🚀 AI technology is evolving, offering new tools for developers to create custom applications. This article explores building an AI-powered private document search app using RAG, ChromaDB, and memory. It highlights the challenges of efficient data processing in vector databases and the need for unstructured data management. The tutorial guides you through connecting an LLM with LangChain and using ChromaDB for memory storage. Steps include installing necessary Python packages and implementing...
Source: The New Stack
Teri Eyenike

Why data governance is the secret to AI agent success

2026-04-10 15:00
AI is enhancing DevOps, with 70% of IT leaders noting its positive impact. However, weak DevOps practices can lead to amplified issues with AI agents. Data governance is crucial as AI handles more tasks, making it essential to ensure compliance, security, and transparency. Surprisingly, only 39% of organizations have automated audit trails despite 77% trusting AI outputs. Investing in strong governance now is vital for successful AI integration, especially in regulated industries. 🛡️📊🔍...
Source: The New Stack
Rod Cope

PyTorch Foundation Expands AI Stack with Safetensors, ExecuTorch, and Helion

2026-04-09 19:18
📢 Exciting updates from the PyTorch Conference EU in Paris! The PyTorch Foundation has announced three new projects: Safetensors, ExecuTorch, and Helion. These additions enhance the open-source AI landscape, providing a vendor-neutral infrastructure for the entire AI lifecycle. Safetensors, developed by Hugging Face, focuses on secure model distribution, reducing security risks in model sharing. ExecuTorch aims to improve on-demand inference capabilities, supporting developers in their AI...
Source: The New Stack
Meredith Shubel

OpenAI’s new $100 tier targets developers hitting Codex (and Claude Code) limits

2026-04-09 18:30
🚀 OpenAI has launched a new $100/month ChatGPT Pro tier aimed at Codex users. This plan offers developers 5x more Codex usage than the $20/month Plus tier. 📈 With over 3 million active users, Codex has seen significant growth recently. The new tier also includes access to Pro models and early access to experimental features. 🔄 This move aligns with Anthropic’s offerings, which also provide similar tiered plans for coding tools. OpenAI emphasizes that Codex delivers more coding capacity per...
Source: The New Stack
Frederic Lardinois

Replit taps RevenueCat to help vibe-coders make money

2026-04-09 18:30
Replit has partnered with RevenueCat to simplify monetization for developers using its platform. 🎉 As app creation becomes easier through AI-assisted tools, turning those apps into revenue remains a challenge. This integration allows users to add subscription features effortlessly using simple prompts. Replit aims to embed monetization into the app development process, making it a seamless part of creation. 💻💰 #Replit #RevenueCat #AppDevelopment #Monetization #VibeCoding
Source: The New Stack
Paul Sawers

Anthropic takes Claude Cowork out of preview and straight into the enterprise

2026-04-09 17:59
🚀 Anthropic has launched Claude Cowork for general availability, allowing non-developers to delegate tasks to a Claude-based agent. This tool, now part of all paid Claude plans, enhances user experience by supporting workflows with text documents and spreadsheets. Key features for enterprises include role-based access controls and usage analytics, addressing governance needs. A new Zoom connector also helps streamline meeting summaries and action items. #ClaudeCowork #Anthropic #AI...
Source: The New Stack
Frederic Lardinois

AWS wants to register your AI agents

2026-04-09 17:30
🚀 AWS has introduced the AWS Agent Registry, a new service to help businesses discover, share, and reuse AI agents across teams. This registry is part of AWS AgentCore, supporting agents from various providers, not just AWS. It aims to enhance visibility into the agent landscape, addressing challenges posed by decentralized AI governance. With the registry, developers can easily search for existing capabilities instead of duplicating work, thus streamlining the development process. Governance...
Source: The New Stack
Frederic Lardinois

The next stages of AI conformance in the cloud-native, open-source world

2026-04-09 17:05
AI model deployment on Kubernetes has been inconsistent across cloud providers due to various technical challenges. As organizations move from labs to production, standardization is becoming crucial. The Cloud Native Computing Foundation (CNCF) has launched a Kubernetes AI conformance program to address this issue, enhancing portability and production readiness. Industry experts predict a shift towards inference workloads dominating AI compute power by 2026. Major cloud providers are already...
Source: The New Stack
Jennifer Riggins

Open source maintainers are drowning in AI-generated pull requests. Enterprise teams are next.

2026-04-09 15:00
Open source maintainers are facing an overwhelming surge of low-quality, AI-generated pull requests. Many projects, like the Jazzband collective, have shut down due to this issue. Maintainers are spending excessive time evaluating these submissions, which crowds out genuine contributions and leads to burnout. This trend poses a similar challenge for enterprise engineering teams that may not be prepared for the influx of AI-generated code. #OpenSource #AICoding #SoftwareDevelopment...
Source: The New Stack
Arjun Iyer

Ramp targets AI’s fastest-growing cost: spend that’s hard to track

2026-04-09 13:00
Companies are facing challenges tracking their increasing AI spending. Ramp, a leading fintech firm, is addressing this issue with a new product aimed at enhancing visibility into AI expenses. By integrating token-level usage data from AI providers, Ramp helps finance teams understand their costs better. AI costs are rapidly growing, often without the oversight seen in traditional spending. Ramp's co-founder, Karim Atiyeh, noted the difficulties his team faced in analyzing their internal AI...
Source: The New Stack
Paul Sawers

Zencoder goes beyond coding

2026-04-09 13:00
🚀 Zencoder has launched Zenflow for Work, its first product designed for non-coders. This marks a significant shift from its origins as a code completion service. The platform now acts as an orchestration layer for AI engineering, streamlining workflows for tasks like writing reports and preparing for meetings. It integrates with tools such as Jira, Gmail, and Google Docs to automate routine tasks. Zencoder aims to extend its capabilities beyond developers, making it useful across various...
Source: The New Stack
Frederic Lardinois

Niantic Spatial wants to map the 80% of the economy AI can’t see

2026-04-08 22:00
🌍 Niantic Spatial is pioneering spatial intelligence, aiming to map the 80% of the economy that AI currently overlooks. Their new service, Scaniverse for businesses, utilizes 3D scans and GPS data to create accurate representations of physical environments. John Hanke emphasizes that while 20% of the economy is online, a vast majority operates outside digital frameworks. #SpatialIntelligence #Niantic #AI #Geospatial #Innovation
Source: The New Stack
Adrian Bridgwater

In the AI Age, Java is More Relevant Than Ever

2026-04-08 21:30
Java is gaining renewed relevance in the AI landscape. Its powerful, scalable, and cost-efficient features make it an ideal choice for modernizing enterprise applications. While Python is often the go-to for AI experimentation, Java excels in production environments. The efficiency of the JVM offers superior performance and cost-effectiveness, particularly crucial in AI development. With new AI frameworks like LangChain4j and Spring AI, Java simplifies integration of AI models, enhancing...
Source: The New Stack
Mary Branscombe

With Claude Managed Agents, Anthropic wants to run your AI agents for you

2026-04-08 17:55
🚀 Anthropic has launched the public beta of Claude Managed Agents, enabling businesses to easily create and deploy cloud-based AI agents. Users can define agents using natural language or YAML files, with all infrastructure managed by Anthropic. This service aims to speed up the deployment process significantly, offering tools for sandboxing and authentication. Some features are still in limited preview, including advanced memory tools and multi-agent orchestration. Governance tools for...
Source: The New Stack
Frederic Lardinois

Microsoft wants to make service mesh invisible

2026-04-08 17:11
At KubeCon EU 2026, Mitch Connors from Microsoft discussed the company's aim to make service mesh technology less visible to users. Microsoft has launched the Azure Kubernetes Application Network, built on Istio’s ambient mode, which simplifies operations and enhances security with mutual TLS by default. Notably, the term "service mesh" is absent from the product name to address customer misconceptions. Connors highlighted that modern AI workloads require different network management compared...
Source: The New Stack
Frederic Lardinois

Open-source leaders question whether Meta’s Alexandr Wang will truly give away its AI models

2026-04-07 20:40
Meta plans to release some of its AI models under an open-source license, led by chief AI officer Alexandr Wang. 🖥️ Wang aims to democratize US-built AI technologies, following Meta's history of contributions to open-source platforms like Llama and PyTorch. 📈 Despite past openness, Meta's approach has shifted, raising questions about the true extent of their commitment to open-source. #AI #OpenSource #Meta #Technology #Innovation
Source: The New Stack
Adrian Bridgwater

Sam Altman promised billions for AI safety. Here’s what OpenAI actually spent.

2026-04-07 20:04
📰 The New Yorker recently published an in-depth investigation into Sam Altman's shifting views on AI safety at OpenAI. The article, spanning over 16,000 words, covers Altman's journey, including his brief exit and return as CEO. Key topics include AI hallucinations, sycophancy in language models, and internal safety review processes. Altman discussed the challenge of balancing user engagement with safety, highlighting the risks of AI hallucinations and the tendency of models to produce overly...
Source: The New Stack
Meredith Shubel

Amazon S3 Files gives the world’s biggest object store a file system

2026-04-07 19:00
🚀 AWS has launched S3 Files, transforming Amazon S3 into a file system. Now, S3 buckets can be accessed using NFS v4.1+, allowing operations like creating, reading, and deleting files. This feature connects AWS compute resources directly to data in S3, enhancing collaborative workloads such as ML training and production applications. However, S3 Files cannot be mounted locally from desktops or other cloud providers. #AWS #S3Files #CloudComputing #FileSystem #TechNews 📊📁
Source: The New Stack
Frederic Lardinois

Anthropic’s Claude Mythos is real, but it’s not for you

2026-04-07 18:00
🌐 Anthropic recently revealed Claude Mythos, a new tier of AI models, despite it not being publicly available yet. Only select partners like Amazon, Apple, and Microsoft will access the Claude Mythos Preview through Project Glasswing for cybersecurity purposes. This model shows strong performance in coding and reasoning, scoring 83.1% on the CyberGym benchmark. Anthropic emphasizes a cautious rollout to ensure that defenders can secure their systems before wider access. #AI #Cybersecurity...
Source: The New Stack
Frederic Lardinois

AWS EKS Auto Mode wants to end Kubernetes toil — one node at a time

2026-04-07 17:55
🚀 AWS EKS Auto Mode aims to reduce the complexities of Kubernetes management. In a recent interview, Alex Kestner, AWS Elastic Kubernetes Service's principal product manager, discussed how this feature can alleviate the daily operational burdens on platform teams. Kestner highlighted that many challenges arise from routine tasks, such as node lifecycle management, which divert focus from delivering value to businesses. Auto Mode is designed to streamline these processes, enabling teams to...
Source: The New Stack
Adrian Bridgwater

True enterprise sovereignty is more approachable than ever, thanks to K8s-powered cloud-neutral PostgreSQL

2026-04-07 15:31
Digital sovereignty is evolving, shifting focus from infrastructure to databases 🎛️. Geopolitical pressures, especially in Europe, are pushing companies to reconsider reliance on managed services. PostgreSQL is emerging as a cloud-neutral solution, allowing enterprises to maintain control while enjoying automation. Gabriele Bartolini of EDB emphasizes the importance of portability for true sovereignty. This approach enhances consistency, compliance, and strategic flexibility for...
Source: The New Stack
TNS Staff

Model Flop Utilization is the metric Aria Networks says will define the AI infrastructure era

2026-04-07 13:00
Aria Networks is introducing a new metric called Model Flop Utilization (MFU) to enhance AI infrastructure efficiency. This metric assesses datacenter performance against peak throughput, crucial for maximizing investment returns. Their "Network that Thinks" initiative combines advanced technologies to improve token efficiency, impacting how quickly models operate across various processing units. As CEO Mansour Karam states, optimizing network performance is vital for achieving the best...
Source: The New Stack
Adrian Bridgwater

Anthropic’s harness shakeup “just fragments workflows,” developers warn

2026-04-06 21:39
Anthropic's recent changes to its Claude subscription limits have raised concerns among developers. 🚨 The removal of third-party harness support has led to fears of fragmented workflows, as these tools are crucial for connecting various AI components. Developers can still use harnesses, but now on a pay-as-you-go basis. 💼 To ease this transition, Anthropic is offering a one-time credit for extra usage, redeemable by April 17, and discounts for pre-purchasing bundles. This move seems aimed at...
Source: The New Stack
Adrian Bridgwater

MCP servers turn Claude into a reasoning engine for your data

2026-04-06 21:01
MCP servers enhance Claude's capabilities by connecting your data directly to its reasoning engine. This allows Claude to analyze your data seamlessly, eliminating the need for cumbersome workarounds. Companies can now easily build MCP servers for more efficient data processing. A simple calculator app will demonstrate the essentials of creating an MCP server. Get ready to leverage Claude for your data needs! 💻📊🔍 #DataAnalytics #AI #Claude #MCPservers #TechInnovation
Source: The New Stack
Jessica Wachtel

MCP maintainers from Anthropic, AWS, Microsoft, and OpenAI lay out enterprise security roadmap at Dev Summit

2026-04-06 16:25
At the MCP Dev Summit in New York, maintainers from Anthropic, AWS, Microsoft, and OpenAI discussed the future of the Model Context Protocol (MCP). They emphasized its governance under the Agentic AI Foundation (AAIF) and its focus on enterprise security and reliability. 🔐📊 With 170 members, AAIF is dedicated to addressing the needs of enterprise users while maintaining MCP's open-source approach. The panel noted that MCP is becoming an industry standard for connecting AI to data and...
Source: The New Stack
Eric Newcomer

Is observability still an operations problem at your organization?

2026-04-06 16:05
📊 Observability is evolving! The shift towards developer-led observability is gaining traction. By giving developers access to runtime telemetry, teams can debug faster and reduce escalations, enhancing software reliability from the start. Join the online event on April 16 to learn how to integrate observability into the development lifecycle. Gain insights on debugging without redeploying code and understanding complex systems. Register now for practical strategies and a chance to win a...
Source: The New Stack
TNS Staff

Cursor’s $2 billion bet: The IDE is now a fallback, not the default

2026-04-05 17:29
Cursor has launched Cursor 3, a new AI code editor that shifts the focus from traditional IDEs to an agent management console. This new interface allows engineers to manage agents and review outputs, with coding as a secondary feature. The design reflects a broader trend in coding tools evolving to prioritize AI interactions. Key features include multi-repo workspaces and Cloud Handoff, enabling seamless transitions between local and cloud environments. #Cursor3 #AICodeEditor #TechInnovation...
Source: The New Stack
Janakiram MSV

SUSE Rancher and Vultr want to break AI infrastructure free from the hyperscalers

2026-04-04 13:00
SUSE Rancher and Vultr are aiming to provide organizations with more options for scaling AI workloads on Kubernetes, moving away from costly hyperscaler solutions. Their recent partnership introduces an open-source AI infrastructure that emphasizes flexibility and independence from vendor lock-in. With Vultr's global GPU infrastructure and SUSE Rancher's support, organizations can run AI workloads more efficiently. This initiative highlights the need for cost-effective, cloud-native solutions...
Source: The New Stack
B. Cameron Gain

Vultr says its Nvidia-powered AI infrastructure costs 50% to 90% less than hyperscalers

2026-04-03 19:08
Vultr is leveraging Nvidia GPUs and AI agents like OpenClaw to automate infrastructure setup for developers. They claim their costs are 50% to 90% lower than those of major hyperscalers. 💻🔧 This platform allows engineering teams to train AI on their security and compliance needs, providing preconfigured options for quick deployment. During KubeCon+CloudNativeCon Europe, Vultr emphasized the shift from manual scripting to using "skill files," enabling a focus on high-level design. For more...
Source: The New Stack
B. Cameron Gain

“I started to lose my ability to code”: Developers grapple with the real cost of AI programming tools

2026-04-03 14:01
🚀 Developers are navigating the impact of AI coding tools on their skills and careers. Paul Ford highlights the excitement of AI's potential, while many developers express mixed feelings. Some enjoy enhanced productivity, but others, like Pia Torain, worry about losing essential coding skills. This raises concerns for junior developers, as reliance on AI may hinder their growth. As AI improves, questions about code quality and the value of expertise emerge. #AICoding #TechConcerns...
Source: The New Stack
David Cassel

Digital Experience Monitoring belongs in the modern developer workflow

2026-04-03 14:00
🌐 Digital Experience Monitoring (DEM) is crucial for modern developers. It connects frontend behavior with backend performance to enhance user experiences. Developers face challenges as production behavior often differs from local tests. DEM bridges this gap by providing insights across all stages of development. It captures real user data, helping teams address issues that may not appear in backend logs. This shift allows for better debugging and validation of user experiences, not just...
Source: The New Stack
Kayla Bondy

The hidden reason your AI assistant feels so sluggish

2026-04-03 13:00
AI workloads are revealing a mismatch in existing data platforms. Many systems, designed for batch reporting, struggle with the demands of agent-driven analytics. ⚙️ The transition from human queries to agent-based interactions leads to higher concurrency and lower latency requirements. This shift challenges traditional cloud data warehouses. 📊 Real-time analytical databases are becoming essential for handling these new workloads efficiently. A growing trend is the use of Postgres combined...
Source: The New Stack
Alasdair Brown

The laptop return that broke a RAG pipeline

2026-04-03 11:00
A recent bug report highlights a significant issue in RAG (retrieval-augmented generation) systems, particularly with returning outdated information. A customer-support agent confidently advised a user about a laptop return policy that had changed, reflecting a "retrieval accuracy gap." This gap occurs when semantic similarity does not equate to contextual correctness. The article suggests a solution: hybrid search, which merges vector similarity with structured SQL queries to improve...
Source: The New Stack
Ed Huang

Anthropic’s rough week: leaked models, exposed source code, and a botched GitHub takedown

2026-04-02 19:59
Anthropic faced significant challenges recently. 🌐 A leak revealed their new AI model, Mythos, and exposed the source code of Claude Code. This incident allowed the public to see 512,000 lines of code, raising concerns about security and transparency. 🔍 Additionally, a botched GitHub takedown removed over 8,000 repositories, which Anthropic later acknowledged was unintentional. 📉 Experts are now questioning the implications of this exposure on AI development and security. #AI #Cybersecurity...
Source: The New Stack
Meredith Shubel

Microsoft execs warn agentic AI is hollowing out the junior developer pipeline

2026-04-02 18:52
Microsoft leaders Mark Russinovich and Scott Hanselman warn that the rise of agentic AI could weaken the junior developer pipeline. In their opinion piece, they highlight how senior engineers benefit from productivity gains, while early-career developers struggle. This "AI drag" may lead companies to favor hiring seniors over juniors, risking a talent shortage in the future. They note that the immediate savings from reducing junior roles may have long-term negative impacts on the industry....
Source: The New Stack
Darryl K. Taft

Why pgEdge thinks MCP (not an API) is the right way for AI agents to talk to databases

2026-04-02 18:30
pgEdge introduces the MCP Server for Postgres, a new service designed for AI applications needing connectivity with databases. This production-ready server supports both new and existing databases running Postgres version 14 and above. It offers flexible deployment options, including on-premises and cloud services. Key features include built-in security, full schema introspection, and reduced token usage, which enhance efficiency and safety for developers. Learn more about how this evolution...
Source: The New Stack
Adrian Bridgwater

The TeamPCP attacks are a warning: Your CI/CD pipeline is the new front line

2026-04-02 16:00
Recent attacks by TeamPCP highlight vulnerabilities in CI/CD pipelines. 🚨 Attackers exploited stolen credentials to deliver malicious versions of popular tools like Trivy and LiteLLM, affecting millions of developers. The trend shows that CI/CD systems, often seen as separate from production environments, are critical yet insecure. 🔒 These incidents reveal a need for stronger security measures in software supply chains, as current defaults may leave organizations exposed. #CyberSecurity...
Source: The New Stack
Dan Lorenc

Why coding agents will break your CI/CD pipeline (and how to fix it)

2026-04-02 15:00
🚨 The rise of AI coding agents is reshaping CI/CD pipelines, raising critical concerns for engineering leaders. As AI generates code faster than teams can manage, new challenges arise. The validation process becomes the bottleneck, risking stability in cloud-native environments. Shared staging environments struggle under the weight of simultaneous code submissions, leading to outages and unmerged code. If not addressed, this issue could negate investments in AI tools, resulting in production...
Source: The New Stack
Arjun Iyer

Why Broadcom gave Velero to the CNCF Sandbox — and what it means for Kubernetes data protection

2026-04-02 14:36
Broadcom has donated Velero to the CNCF Sandbox, enhancing its support for Kubernetes and open-source collaboration. This move aims to foster community trust and reduce proprietary perceptions. The shift in governance will help Velero better meet community needs. Broadcom seeks to strengthen its position in cloud-native solutions while addressing Kubernetes management challenges. Learn more about this strategic decision in the latest podcast episode with Dilpreet Bindra. 🎧🌐 #Kubernetes...
Source: The New Stack
B. Cameron Gain

OpenClaw vs. Hermes Agent: The race to build AI assistants that never forget

2026-04-02 14:26
The article discusses the challenges developers face with AI coding assistants, particularly the loss of context between sessions. OpenClaw and Hermes Agent are two open-source projects aiming to create AI assistants that remember user inputs permanently. They represent a shift from session-based tools to persistent agents that learn and adapt over time. While some session-native tools are adding memory features, OpenClaw and Hermes Agent are built for continuous operation, improving user...
Source: The New Stack
Janakiram MSV