The New Stack
DevOps, open source, and cloud native news with resources and insights for developers
CloudBees CEO: Why Migration Is a Mirage Costing You Millions
CloudBees CEO Anuj Kapur warns that many enterprises chase costly migration projects, calling it a “migration mirage.” A recent survey shows 57% of organizations spent over $1 million on migrations last year, with an average $315,000 loss per project due to overruns. Kapur argues that migration extends timelines, inflates budgets, and demoralizes engineers, while legacy systems often outlive CIOs. He advocates meeting developers where they are and focusing on true modernization rather than wholesale platform replacement.
Tenants: The Missing Backbone of Modern Developer Platforms
Internal developer platforms often collapse under cloud sprawl, compliance pressure, and AI‑driven automation because they lack a first‑class boundary. The article argues that treating a tenant as an explicit logical construct—binding identity, networking, secrets, compute, observability, and compliance—solves these issues....
4 Core Principles for Scaling Your API Engineering Practice
Enterprises are rapidly expanding API portfolios from dozens to thousands, exposing gaps in traditional, ad‑hoc development practices. The New Stack outlines four core principles—treat every artifact as code, maintain a single source of truth, automate all repeatable tasks, and delegate...
Ramp Adds Developer Tools to Platform With AI Coding Assistant
Ramp introduced Ramp Inspect, an AI‑powered coding assistant integrated into its spend‑management platform. The cloud‑hosted tool writes, tests, and visually verifies code, offering instant virtual‑machine environments without local setup. Ramp open‑sourced the Inspect blueprint, enabling other firms to build similar...
The Weekend Our Pipeline Processed the Same Data 47 Times
A data pipeline re‑processed Saturday’s transactions 47 times over a weekend after Airflow retries invoked a flawed fallback that loaded the previous successful date instead of the missing data. The bug let each retry succeed while still ingesting stale data,...
Forks, Clouds and the New Economics of Open Source Licensing
The Business Source License (BSL) has emerged as a hybrid model that lets developers use software freely while restricting commercial re‑hosting, a shift sparked by HashiCorp’s move from MPL 2.0 to BSL for Terraform. High‑profile forks such as OpenTofu illustrate...
Build Cheaper, Safer, Auditable AI with SLMs and RAG
Enterprises moving LLM pilots to production face soaring infrastructure costs, unpredictable latency, and limited auditability. A growing alternative pairs small language models (SLMs) with retrieval‑augmented generation (RAG), delivering CPU‑friendly inference and data‑grounded outputs. The article outlines a modular, agent‑based architecture...
Postman Adds Open Source Documentation and SDK Products
Postman announced the acquisition of Fern, an open‑source developer‑experience company that provides Fern Docs and an SDK generator supporting languages such as TypeScript, Python, Go, Java, Ruby, C# and PHP. The move expands Postman's API platform with customizable documentation and...
Google Cloud: A Deep Dive Into GKE Sandbox for Agents
Google Cloud introduced GKE Sandbox for Agents, a Kubernetes‑native extension that runs untrusted or specialized workloads in lightweight, VM‑like sandboxes powered by gVisor. The solution adds a new Sandbox custom resource definition (CRD) and controller, offering stable identities, persistent storage,...
From Guesswork to Guardrails: Kubernetes Container Rightsizing
Container rightsizing adjusts Kubernetes CPU and memory requests to reflect real usage, eliminating the drift caused by static, early‑stage settings. Oversized requests waste capacity, inflate node counts, and can force larger instance types, while undersized settings risk performance issues. The...
Operations Shift: Assistants to Autonomous Multiagent Systems
Enterprise AI is moving from simple assistants that merely aid human tasks to autonomous agents that execute within guardrails, and finally to coordinated multi‑agent systems that own end‑to‑end outcomes. The article outlines a three‑phase roadmap over the next two years,...
Bryan Cantrill: How Kubernetes Broke the AWS Cloud Monopoly
Bryan Cantrill argues that the 2014 release of Kubernetes disrupted Amazon Web Services' near‑monopoly by giving enterprises a cloud‑agnostic orchestration layer. He notes AWS’s 2014 revenue of $4.6 billion and its aggressive price‑cut strategy, which left rivals scrambling. Kubernetes’ open‑source model...
The Next Era of AI: From Single User to Team Collaboration
The article outlines a transition from single‑user, prompt‑driven AI to multiplayer AI agents that embed themselves in team conversations and retain project context over time. It describes four core agent roles—context, coordination, synthesis, and execution—that work together to surface information,...
From Group Science Project to Enterprise Service: Rethinking OpenTelemetry
Ari Zilka, former Hortonworks CPO and founder of MyDecisive.ai, highlighted at KubeCon 2025 that 23 observability vendors are delivering nearly identical reactive dashboards that only surface problems after failures occur. He argued that CIOs consistently demand lower mean time to...
Stop Drowning in Alerts: Why Signals Are Your Team’s Lifeline
Engineering teams are moving from noisy, threshold‑based alerts to higher‑order signals that embed context, confidence, and actionable insight. Alerts, once sufficient for simple infrastructures, now generate fatigue as they treat every deviation equally. Signals combine correlation, relevance to business impact,...
Why Decoupling Control and Data Planes Is the Future of SaaS
The article argues that separating the control plane from the data plane is reshaping SaaS delivery. In a decoupled model, the vendor provides the control plane as a managed service while customers run the data plane in their own cloud...
Martin Fowler on Preparing for AI’s Nondeterministic Computing
Martin Fowler, chief scientist at Thoughtworks, frames large language models as a move to nondeterministic computing, comparable to the historic shift from assembly language to high‑level languages. He contrasts deterministic binary logic with the probabilistic nature of LLM outputs, emphasizing...
The 3 A.m. Call That Changed The Way I Design APIs
A critical outage at 3:17 a.m. exposed a single‑point failure in a customer‑facing API, costing $14,000 in SLA credits and eroding trust. The incident led the author to create the “3 a.m. Test,” a design checklist that ensures on‑call engineers can quickly...