The New Stack Podcast

In this episode of The New Stack Makers, Microsoft Corporate Vice President and Technical Fellow, Brendan Burns discusses how AI is reshaping Kubernetes and modern infrastructure. Originally designed for stateless applications, Kubernetes is evolving to support AI workloads that require complex GPU scheduling, co-location, and failure sensitivity. Features like Dynamic Resource Allocation and projects such as KAITO introduce AI-specific capabilities, while maintaining Kubernetes’ core strength: vendor-neutral extensibility. 

Burns highlights that AI also changes how systems are monitored. Success is no longer binary; it depends on answer quality, user feedback, and large-scale testing using thousands of prompts and even AI evaluators. 

On software development, Burns argues that the industry’s focus on reviewing AI-generated code is temporary. Just as developers stopped inspecting compiler output, AI-generated code will become a disposable artifact validated by tests and specifications. This shift will redefine engineering roles and may lead to programming languages designed for machines rather than humans, signaling a fundamental transformation in how software is built and maintained.

Learn more from The New Stack about the latest developments around how AI is reshaping Kubernetes and modern infrastructure:

How To Use AI To Design Intelligent, Adaptable Infrastructure

The AI Infrastructure crisis: When ambition meets ancient systems 

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

In this episode ofThe New Stack Agents, Marcin Wyszynski, co-founder of Spacelift and OpenTofu, explains how AI is transforming infrastructure as code (IaC). Originally built for individual operators, tools like Terraform struggled to scale across teams, prompting Wyszynski to help launch OpenTofu after HashiCorp’s 2023 license change. Now, the bigger shift is AI: engineers no longer write configuration languages like HCL manually, as AI tools generate it, dramatically lowering the barrier to entry.

However, this creates a dangerous gap between generating infrastructure and truly understanding it—like using a phrasebook to ask questions in a foreign language but not understanding the response. In infrastructure, that lack of comprehension can lead to serious risks.

To address this, Spacelift introduced Intent, which allows AI to directly interact with cloud systems in real time while enforcing deterministic guardrails through policy controls. The broader challenge remains balancing speed with control—enabling faster experimentation without sacrificing safety. Wyszynski argues that, like humans, AI can be trusted when constrained by strong guardrails.

Learn more from The New Stack about the latest developments around how AI is transforming infrastructure as code (IaC).

The Maturing State of Infrastructure as Code in 2025

Generative AI Tools for Infrastructure as Code

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Woodson Martin, CEO ofOutSystems, argues that successful enterprise AI deployments rarely rely on standalone agents. Instead, production systems combine AI agents with data, workflows, APIs, applications, and human oversight. While claims that “95% of agent pilots fail” are common, Martin suggests many of those pilots were simply low-commitment experiments made possible by the low cost of testing AI. Enterprises that succeed typically keep humans in the loop, at least initially, to review recommendations and maintain control over decisions.

Current enterprise use cases for agents include document processing, decision support, and personalized outputs. When integrated into broader systems, these applications can deliver measurable productivity gains. For example,Travel Essencebuilt an agentic system that reduced a two-hour customer planning process to three minutes, allowing staff to focus more on sales and helping drive 20% top-line growth.

Martin also believes AI will pressure traditional SaaS seat-based pricing and accelerate custom software development. In this environment, governed platforms like OutSystems can help enterprises adopt “vibe coding” while maintaining compliance, security, and lifecycle management.

Learn more from The New Stack about the latest developments around enterprise adoption of vibe coding:

How To Use Vibe Coding Safely in the Enterprise

5 Challenges With Vibe Coding for Enterprises 

Vibe Coding: The Shadow IT Problem No One Saw Coming

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

On a recent episode of the The New Stack Agents, Inception Labs CEO Stefano Ermon introduced Mercury 2, a large language model built on diffusion rather than the standard autoregressive approach. Traditional LLMs generate text token by token from left to right, which Ermon describes as “fancy autocomplete.” In contrast, diffusion models begin with a rough draft and refine it in parallel, similar to image systems like Stable Diffusion.

This parallel process allows Mercury 2 to produce over 1,000 tokens per second—five to ten times faster than optimized models from labs such as OpenAI, Anthropic, and Google, according to company tests. Ermon argues diffusion models better leverage GPUs, with support from investor Nvidia to optimize performance.

While Mercury 2 matches mid-tier models like Claude Haiku and Google Flash rather than top systems such as Claude Opus or GPT-4, Ermon believes diffusion’s speed and economic advantages will become increasingly compelling as AI applications scale.

Learn more from The New Stack about the latest developments around around large language model built on diffusion: 

How Diffusion-Based LLM AI Speeds Up Reasoning

Get Ready for Faster Text Generation With Diffusion LLMs 

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

OnThe New Stack Agents, Gavriel Cohen discusses why he built NanoClaw, a minimalist alternative to OpenClaw, after discovering security and architectural flaws in the rapidly growing agentic framework. Cohen, co-founder of AI marketing agencyQwibit, had been running agents across operations, sales, and research usingClaude Code. When Clawdbot (laterOpenClaw) launched, it initially seemed ideal. But Cohen grew concerned after noticing questionable dependencies—including his own outdated GitHub package—excessive WhatsApp data storage, a massive AI-generated codebase nearing 400,000 lines, and a lack of OS-level isolation between agents.

In response, he createdNanoClawwith radical minimalism: only a few hundred core lines, minimal dependencies, and containerized agents. Built around Claude Code “skills,” NanoClaw enables modular, build-time integrations while keeping the runtime small enough to audit easily. Cohen argues AI changes coding norms—favoring duplication over DRY, relaxing strict file limits, and treating code as disposable. His goal is simple, secure infrastructure that enterprises can fully understand and trust.

 

Learn more from The New Stack about the latest around personal AI agents

Anthropic: You can still use your Claude accounts to run OpenClaw, NanoClaw and Co.

It took a researcher fewer than 2 hours to hijack OpenClaw

OpenClaw is being called a security “Dumpster fire,” but there is a way to stay safe

Join our community of newsletter subscribers to stay on top of the news and at the top of your game.