The New Stack Podcast

Harness Field CTO Martin Reynolds joins The New Stack to talk about what happens after coding agents start opening pull requests faster than anyone can review them. He explains how he first saw the bottleneck during early GitHub Copilot trials, the three ways enterprises are coping with the volume now, and why Harness rebuilt its Code Repository and launched AI Code Review for agent traffic. The conversation also covers GitHub's recent outages, the software delivery knowledge graph behind Harness's reviewer, and how much of the delivery pipeline should stay deterministic. Learn more from The New Stack around the latest in coding agents: AI coding agents can write code, Crafting wants to help them ship it Git real: AI agents aren't just for solo developers anymore Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Traces provide a detailed view of a request’s journey through data, microservices and applications, helping SREs pinpoint where failures occur and resolve issues faster. But while tracing can reduce downtime and developer burnout, collecting every trace creates its own problems. Storing massive volumes of data is expensive, can burden the systems being monitored and makes it harder to find the information that actually matters. The solution isn’t abandoning tracing, but being smarter about what gets retained. Head sampling captures only a portion of traces upfront, while tail sampling evaluates completed traces and keeps those most valuable for troubleshooting. Dynamic sampling goes further by filtering repetitive or nearly identical traces before they overwhelm storage. On The New Stack podcast, Sarah Hudspeth of Chronosphere, a Palo Alto Networks company, explains how teams can build a more effective tracing strategy. She breaks down how thoughtful sampling and observability design can turn tracing from a data-hoarding problem into a practical tool for production troubleshooting. Learn more from The New Stack around the latest in tracing: Sampling: the philosopher’s stone of distributed tracing How OpenTelemetry Works: Tracing, Metrics and Logs on Kubernetes Why Synthetic Tracing Delivers Better Data, Not Just More Data Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

As AI evolves from conversational chatbots to autonomous agents, CPUs are becoming an increasingly important part of the infrastructure equation. In this episode, The New Stack speaks with Bhumik Patel of Arm and Mo Farhat of Google about how CPUs act as an “air traffic controller” for agentic workloads, handling orchestration, data preparation, semantic search, vector databases, code execution and API calls alongside GPUs and TPUs. Smaller AI models, including summarizers and evaluators, can also run effectively on CPUs for specialized tasks. As agents increasingly generate and execute code, secure sandboxing becomes critical. Google’s gVisor and GKE Agent Sandbox provide isolation and scalable environments, with the latter supporting up to 300 sandboxes per second per cluster. The discussion also explores efficiency and cost, with Google highlighting Axion’s price-performance and energy-efficiency advantages across different workload types. Ultimately, the shift toward agentic AI is creating a more diverse compute environment where CPUs, GPUs and TPUs each play complementary roles in delivering scalable, efficient AI applications. Learn more from The New Stack around the latest in CPUs in the world of AI agents: AI Agents Will Eat Enterprise Software, Just Not in One Bite How to ground AI agents in accurate, context-rich data Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Doist CTO Gonçalo Silva says AI is reshaping software development, but success depends on restraint rather than rapid feature expansion. Instead of chasing every AI capability, Doist prioritizes “subtraction over addition,” removing features that fail to deliver lasting value despite development investment. After experimenting with nearly 20 AI concepts, the company found success with Ramble, an AI-powered voice task capture feature, while remaining model-agnostic through rigorous testing and evaluations. Internally, developers use a variety of AI coding tools rather than standardizing on one platform, while Doist OS—a companywide AI assistant with nearly 100 shared skills—helps employees across all functions work more effectively. Silva also outlined Doist’s approach to AI-powered automations, separating AI-driven workflow generation from deterministic execution to improve reliability and reduce token costs. Throughout its AI strategy, the company emphasizes purposeful features, privacy, transparency, and continuous improvement, ensuring AI enhances user productivity without compromising product quality or trust. Learn more from The New Stack around developer productivity: Developer Productivity in 2025: More AI, but Mixed Results Optimizing for Developer Productivity Creates a Winning DevEx Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

As AI coding agents accelerate software development, they also create new challenges for site reliability engineers (SREs), who are increasingly responsible for debugging systems that no single human fully understands. In this episode of The New Stack podcast, Sam Farid and Nate Heinrich of Chronosphere argue that AI agents should also be used for root-cause analysis, helping teams diagnose failures more quickly as model capabilities continue to improve. Rather than immediately purchasing a commercial solution, they recommend organizations first build an in-house AI SRE. The process of documenting systems, dependencies, and operational knowledge creates valuable context that enables AI agents to troubleshoot effectively while improving institutional knowledge. Although Chronosphere offers its own AI SRE platform, the hosts emphasize that building an internal prototype helps teams understand their needs before evaluating vendor tools. As AI-generated code becomes more common, organizations that invest in mapping their systems and leveraging AI for operations will be better equipped to reduce downtime and support increasingly complex software environments. Learn more from The New Stack around AI SREs: 5 ways SRE AI agents are set to augment human capabilities The Future of AI in SRE: Preventing Failures, Not Fixing Them AI Reliability Engineering: Welcome to the Third Age of SRE Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

In this episode with The New Stack Agents, Frederic Lardinois, NVIDIA’s Joey Conway says advances in AI over the past year have dramatically improved the capabilities of local models, making them practical for enterprise and personal use alongside frontier cloud models. Rather than replacing large models, Conway envisions a “system of models” where specialized local models handle routine, cost-sensitive, or privacy-focused tasks, while larger frontier models tackle more complex reasoning. He explains that organizations can fine-tune smaller open models using domain-specific data, creating expert AI agents that reflect the specialized roles found within businesses. NVIDIA supports this ecosystem through open models, training tools, and software such as NeMo, Dynamo, and Nemotron. Conway also highlights the growing importance of agentic harnesses, which give AI models access to tools, memory, and iterative workflows, significantly improving performance and reducing costs. Looking ahead, he expects AI orchestration to become increasingly important, with intelligent routing systems selecting the right model for each task based on complexity, cost, latency, and data governance requirements, enabling enterprises to balance performance, security, and efficiency. Learn more from The New Stack around NVIDIA's latest updates in AI: Palantir and Nvidia want to change who owns government AI Nvidia's best model is now live Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

In this episode, Mark Russinovich, CTO of Microsoft Azure revealed Brain, the AI-powered AIOps system that continuously monitors Azure’s health, detects incidents, identifies root causes, and increasingly automates responses such as pausing problematic deployments and notifying affected customers. Built on Azure Resource Graph, Brain creates a real-time digital twin of Azure, mapping dependencies across hundreds of services, data centers, and regions. Although Brain predates the generative AI boom, years of data engineering, standardized service-level indicators (SLIs), and machine learning laid the foundation for today’s capabilities. Brain combines standardized SLIs, service-specific monitoring, and third-party signals to detect anomalies, while ML models dynamically establish service baselines and correlate outages with software rollouts. Microsoft says automated notifications have reduced customer support tickets by four to six times, with 80–90% of Brain-covered services receiving notifications within 15 minutes, often in under five. The company is also layering LLM-powered agents, called Triangle, on top of Brain to streamline incident routing and eventually enable AI agents to autonomously troubleshoot and remediate outages. Learn more from The New Stack around the latest in Microsoft Azure: Meet Brain, the AI that decides when Azure is officially down Microsoft's pitch to enterprises: Ditch Azure Repos for GitHub, despite its rocky reliability record Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Subquadratic is beginning to back up its ambitious claims with benchmarks and third-party validation for its SubQ 1.1 Small model, which uses its proprietary Sparse Attention (SSA) architecture to dramatically improve long-context performance. Rather than comparing every token to every other token, SSA selectively processes relationships, enabling near-linear scaling while maintaining high accuracy across context windows of up to 12 million tokens. The company reports near-perfect retrieval performance, competitive coding and reasoning benchmarks, and compute savings of up to 1,000x at maximum context lengths. Rather than targeting frontier models immediately, Subquadratic is focusing on enterprise customers that need efficient analysis of massive datasets. The current model was built by replacing the dense attention mechanism in an existing open-weight model and then continuing long-context pretraining. Looking ahead, the startup plans to release a larger mid-tier model while continuing research into "zero attention" architectures that could eliminate attention mechanisms altogether, with the long-term goal of surpassing today's transformer-based AI models in both efficiency and capability. Learn more from The New Stack around cloud spending: The context window has been shattered: Subquadratic debuts a 12-million-token window What comes after attention? This startup says it already knows. Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Harness has introduced Autonomous Worker Agents, a new capability that allows enterprises to replace rigid CI/CD pipeline scripts with AI agents that can deploy applications, run tests, and perform security scans while operating under existing governance, security, and audit controls. Unlike Harness' existing expert agents, which assist developers with coding and pipeline creation, Worker Agents autonomously execute pipeline tasks within customer-controlled infrastructure. Agents are defined using simple Markdown files, draw context from the Harness Software Delivery Knowledge Graph, and run in sandboxed environments with scoped permissions and policy enforcement. Harness also provides built-in audit trails that record prompts, decisions, and outcomes, along with token budgets and approval gates to control AI costs. The launch includes an Agent Marketplace featuring Harness-managed, certified partner, and community-built agents. CEO Jyoti Bansal said production AI agents require far stronger safeguards than coding assistants, positioning Harness' governance and knowledge graph as key differentiators. Looking ahead, the company envisions fully autonomous software engineering, where AI agents manage the software lifecycle while humans oversee high-risk decisions. Learn more from The New Stack around AI software delivery: AI won't speed up software delivery - nothing has How to solve the AI paradox in software development with intelligent orchestration Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

More than two decades after AWS helped usher in the public cloud era, many organizations are reassessing whether a cloud-first strategy still delivers the cost and operational benefits it once promised. While hyperscalers such as AWS, Azure and Google Cloud have built enormously successful businesses, cloud spending has become a growing concern for customers as usage expands and costs continue to rise. On this episode of The New Stack Makers, Summit’s Byron Dill argues that many enterprises have become overly reliant on public cloud infrastructure, using it for workloads that may be better suited to private environments. Rather than treating the cloud as a one-size-fits-all solution, Dill advocates for a more segmented approach that places workloads where they make the most sense based on cost, security and management requirements. The conversation draws parallels to the rapid adoption of AI, where organizations often discover unexpected costs after implementation. Dill explores when repatriating workloads from the public cloud to private infrastructure can reduce expenses, simplify data management and improve control, while examining the costs, timelines and industries best positioned to benefit from a private cloud strategy. Learn more from The New Stack around cloud spending: How to Cut Cloud Waste Without Constricting Developer Productivity AI agents need to spend money — Stripe and iWallet are building the rails Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Gusto is betting that small businesses need more than another AI assistant. The company’s new product, Gusto Cofounder, is designed to act as a proactive business partner that helps owners manage and grow their companies, drawing inspiration from the traditional mom-and-pop partnership that co-founder and CTOEddie Kimwitnessed growing up. Unlike reactive chatbots, Cofounder can take action across payroll, HR, benefits, scheduling, insurance, and accounting workflows by leveraging data already stored within Gusto. Users interact with the platform through text messages or Slack, while a consent framework ensures access to sensitive payroll and employee data remains tightly controlled. Businesses can grant explicit permissions and gradually increase autonomy as trust is established. The platform also integrates with third-party tools such as Google Workspace, enabling it to gather data, perform calculations, run payroll, and communicate results automatically. Kim said the product was built by a five-person team in just eight weeks using Claude Code, which he believes demonstrates how AI is expanding software creation beyond traditional engineering roles. Looking ahead, Gusto plans to add more integrations and eventually enable customers and developers to share reusable, industry-specific business automations. Learn more from The New Stack around how AI is expanding software creation beyond traditional engineering roles: How AI Is Reshaping Software Engineering: Key Takeaways From DeveloperWeek 2025 AI and the Future of Code: Developers Are Key The Engineer in the AI Age: The Orchestrator and Architect Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

WeAreDevelopers, the Berlin-based developer conference founded in 2015, has grown into a major global event, attracting 15,000 developers from over 70 countries each year. In 2026, it expands beyond Europe with new editions in San Jose, California, and Bengaluru, India. Co-founder and CEO Sead Ahmetovic says the conference was created to give developers a stronger voice in an industry where marketers, salespeople, and entrepreneurs often receive more recognition. He believes developers, despite being less vocal, build the products that power the modern world. The event began as a small meetup that quickly gained popularity, filling a gap between highly specialized technical gatherings and broader business-focused conferences. Former GitHub CEO Thomas Dohmke highlights another benefit: giving developers a platform to share the stories behind their work and inspire peers. Discussing the future of software development, Dohmke predicts AI agents will handle much of the coding, while developers focus on managing ideas, prompts, and workflows. Ahmetovic agrees, arguing that developers will remain essential, spending less time typing code and more time thinking, orchestrating, and creating new solutions. Learn more from The New Stack around the latest in developer community growth: How Community Helps Developers Grow Empowering Developers Is Critical to Drive AI Innovation 3 Ways Organizations Can Redefine the Developer Experience Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

At a recent MCP developer summit, The New Stack spoke with Till Döhmen, AI lead atMotherDuck, about the company’s growing role in the evolving DuckDB ecosystem. Backed by investors includingTomasz Tunguz, MotherDuck is commercializing the open-source analytical databaseDuckDBwhile also expanding how employees interact with data through AI agents rather than traditional dashboards. Döhmen emphasized the company’s close collaboration withDuckDB FoundationandDuckDB Labs. Because MotherDuck operates what he described as the world’s largest fleet of DuckDB databases, the startup regularly pushes the database to its limits and feeds insights back to the core maintainers. Rather than forking DuckDB to create proprietary advantages, MotherDuck instead extends the platform through its existing architecture while contributing core improvements upstream when needed. The conversation highlighted the delicate but productive relationship between venture-backed companies and the open-source projects they commercialize, positioning MotherDuck as another example of startups driving both OSS adoption and strong business growth simultaneously. Learn more from The New Stack around the latest in DuckDB: DuckDB: Query Processing Is King DuckDB: In-Process Python Analytics for Not-Quite-Big Data Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

JetBrains is positioning itself as the last major independent AI coding-tool vendor in a market increasingly tied to hyperscalers and foundation model labs. Speaking at Google Cloud Next, JetBrains VP of business developmentMikhail Vink argued that competitors such as Microsoft Copilot, Anysphere Cursor, and Windsurfare all tied to either AI labs or cloud providers. By contrast, JetBrains says its independence allows customers to switch freely between models fromOpenAI,Anthropic, andGoogle Cloudwithout being locked into one ecosystem. That flexibility underpins JetBrains’ broader AI strategy. Rather than building its own foundation model, the company is focusing on orchestration and governance through JetBrains Central, announced in March as a management layer for AI agents, usage controls, analytics, and consumption-based billing. Vink said the company’s profitability, 16 million users, and 300,000 commercial customers from its long-running IDE business have allowed it to remain venture-free and model-neutral. JetBrains argues that as developers increasingly swap between AI models, neutrality may become more valuable than owning the models themselves. Learn more from The New Stack around the latest in AI coding-tools: JetBrains ‘Agentic’ AI Agent Helps Automate Coding Tasks JetBrains: AI agents are about to repeat the cloud ROI crisis JetBrains names the debt AI agents leave behind Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

What began as an internal developer tool atBlockhas evolved into a broader open-source initiative with industry backing. Goose, Block’s AI coding agent, followed a path similar to Amazon’s transformation of internal infrastructure intoAmazon Web Services. After deploying Goose companywide, Block open-sourced the tool under a permissive license, leading to rapid adoption across the developer community. But according to Manik Surtani, Office of the CTO, Block and Co Founder of Agentic AI Foundation, early momentum exposed governance challenges. Although Goose was technically open source, Block retained trademark ownership, creating concerns for enterprises seeking truly independent governance. To address this, the team partnered with the creators ofAnthropicand the Model Context Protocol community to establish theAgentic AI Foundationunder the umbrella of theLinux Foundation. Goose, MCP, and Agents.MD became the foundation’s initial projects, chosen largely to accelerate the launch of the new organization and create a collaborative ecosystem around agentic AI development. Learn more from The New Stack around the latest in open-source AI: Anthropic extends MCP with a UI framework Why the Linux Foundation adopted MCP, with Jim Zemlin and Mazin Gilbert Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

At Google Cloud Next 2026, Fivetran Chief Product Officer Anjan Kundavaram argued that enterprise data systems are unprepared for the scale of AI-driven analytics. Unlike humans, AI agents can generate exponentially more queries, often routing them through the same expensive compute infrastructure. Kundavaram compared it to “using a Lamborghini to mow the lawn.” To address this, Fivetran introduced its “Open Data Infrastructure” vision and a benchmark designed to expose hidden AI workload costs in closed ecosystems. Kundavaram said agents can optimize for cost instead of speed, choosing cheaper compute engines when appropriate — but only in open architectures with multiple options. Closed systems force every query through high-cost paths. He also warned that fragmented data and weak context create a “triple whammy” of poor AI responses, soaring analytics bills, and wasted compute. While many organizations respond by tightening controls, Kundavaram argued the better path is investing in open infrastructure, interoperability, and strong semantic data practices before AI costs spiral further. Learn more from The New Stack around the latest in enterprise data systems: Enterprise AI Success Demands Real-Time Data Platforms AI Agents Are Morphing Into the 'Enterprise Operating System' Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

At Google Cloud Next 2026, Finout co-founder and CEO Roi Ravhon and Google Cloud FinOps lead Pathik Sharma discussed how FinOps is rapidly evolving for the AI era. Ravhon argued that while cloud FinOps had a decade to mature, AI economics are forcing the industry to adapt within a year. Unlike traditional cloud workloads, AI costs are unpredictable because token usage varies even for identical prompts, while advanced reasoning models consume significantly more tokens despite falling prices. Both emphasized that effective AI FinOps requires intelligent orchestration, routing workloads to the cheapest capable models instead of defaulting to expensive frontier models. Sharma noted that AI costs extend beyond APIs to GPUs, storage, training, and organizational adoption. They also cautioned against relying solely on LLMs for operational automation. Deterministic systems, observability metrics, and human approvals remain essential guardrails. Ultimately, both stressed that FinOps is primarily an organizational and cultural discipline, recommending newcomers start with the FinOps Foundation before investing in tools. Learn more from The New Stack around the latest in FinOps: Why FinOps Isn’t About Saving Money FinOps Foundation’s FOCUS 1.2 Expands to SaaS, PaaS Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Managing Kubernetes at fleet scale introduces significant complexity, especially as organizations expand from a few clusters to hundreds or thousands across cloud, on-premises, and edge environments. While GitOps remains the dominant model for declarative management, its traditional one-to-one repository-to-cluster approach struggles to handle multi-cluster realities such as global traffic routing, shared secrets, and unified observability. AsStephane Erbrech, Principal Software Engineer at Microsoftexplains, the challenge shifts from deployment to governance—maintaining consistency, security, and compliance across a vast distributed system without manual intervention. This need is amplified by the rise of AI workloads at the edge, where inference is increasingly decentralized. To address these challenges,Microsoft Azure Kubernetes Fleet Managerenables coordinated, staged rollouts across clusters, allowing teams to validate updates in lower-risk environments before production. Supporting this,Cilium Cluster Meshprovides seamless cross-cluster connectivity, enabling workload mobility and efficient resource use, especially for scarce GPU capacity. Together, these tools help modern platform teams manage lifecycle, networking, and orchestration at scale. Learn more from The New Stack around managing Kubernetes at fleet scale: KubeFleet: The Future of Multicluster Kubernetes App Management Why Microsoft is betting on temporary identities to stop autonomous agents from going rogue Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

In this episode of The New Stack Makers , Matthew O’Riordan, CEO of Ably, explains how infrastructure originally built for human collaboration is now well-suited for long-running AI agents. While Ably initially resisted positioning itself as an AI company, the rise of agents that reason, call tools, and operate over extended periods revealed a natural fit for its real-time communication platform. O’Riordan highlights the limitations of HTTP for these use cases. While effective for short, request-response interactions, HTTP struggles with persistent, stateful experiences—such as handling dropped connections, multi-device usage, or mid-task interruptions. To address this, a new “durable session” layer is emerging, enabling continuous synchronization between agents and users through shared state, presence, and recovery mechanisms. Ably’s solution, AI Transport, augments existing architectures by keeping HTTP for requests while shifting responses to durable sessions. Features like mutable message streams and “live objects” allow seamless reconnection and collaboration. The goal is to provide a drop-in layer that developers can adopt without rethinking their stack—moving beyond traditional pub/sub models. Learn more from The New Stack around Ably and AI Transport: How MCP Uses Streamable HTTP for Real-Time AI Tool Interaction Ably Touts Real-Time Starter Kits for Vercel and Netlify AI Agents Need Help. Here’s 4 Ways To Ship Software Reliably Join our community of newsletter subscribers to stay on top of the news and at the top of your game.

Agentic AI is advancing rapidly, with open-source projects racing to keep pace with real-world deployment. To accelerate progress, the Linux Foundation consolidated key technologies—Model Context Protocol (MCP), Goose, and AGENTS.md—under the newly formed Agentic AI Foundation (AAIF) in late 2025. At the MCP Dev Summit in New York City, Linux Foundation CEO Jim Zemlin and newly appointed AAIF executive director Mazin Gilbert discussed this transition. Zemlin explained that leading both organizations was unsustainable, prompting a careful search for a leader with both technical expertise and collaborative leadership skills. Gilbert now takes on the challenge of guiding AAIF as it shapes the emerging agentic AI ecosystem. While the foundation currently oversees three projects, its broader mission involves defining the future architecture of agent-driven systems—deciding what to build, when, and why. These decisions will influence the trajectory of open-source AI development. The conversation also highlights the importance of open collaboration, funding dynamics, and early adopters in shaping the agentic stack’s evolution. Learn more from The New Stack around the latest in open-source projects and The Linux Foundation: Anthropic Donates the MCP Protocol to the Agentic AI Foundation SAFE-MCP, a Community-Built Framework for AI Agent Security Google Donates the Agent2Agent Protocol to the Linux Foundation Join our community of newsletter subscribers to stay on top of the news and at the top of your game.