The New Stack publishes guide to Kubernetes platforms built for AI workloads in 2027
A new ebook from The New Stack and HPE outlines 10 practical checks for Kubernetes teams managing AI and agentic workloads, moving beyond basic pod monitoring to address platform-wide concerns.
The New Stack has released a new ebook, produced in partnership with HPE, that provides 10 practical checks designed for Kubernetes platforms running AI workloads and agents.
The presence of AI workloads—particularly those leveraging agentic AI—is reshaping the operational questions that Kubernetes teams must address. Traditional concerns like verifying that pods remain operational, measuring time to first token in model inference, or managing GPU capacity during traffic surges now sit alongside broader platform considerations. These questions are evolving into platform-level challenges that exist within the context of upgrades, disaster recovery, and security practices.
The New Stack has published Kubernetes in 2027: Building Platforms for AI, Autonomy, and Scale, a new ebook authored by Chris J. Preimesberger in collaboration with HPE.
Pre-deployment preparation
The ebook emphasizes work that must occur before systems go live. It recommends that teams conduct stress tests on model-serving infrastructure using realistic workload patterns, simulate failure scenarios, and track metrics beyond GPU utilization alone.
Agent access and governance
The guide examines the implications of granting agents access to a cluster, treating this as a significant architectural decision regardless of how straightforward the agent's assigned task may appear.
The ebook calls for clearly designated owners, restricted permission scopes, approval workflows, and comprehensive audit logs that trace tool invocations back to the original task request. It also emphasizes the importance of evaluating the quality of AI-generated outputs, noting that speed provides limited value when results are inaccurate.
Broader platform considerations
Rather than treating AI as justification to abandon established platform practices, the ebook integrates AI concerns with existing operational requirements. The checklist addresses policy consistency across multiple clusters, observability into systems running outside Kubernetes, upgrade and recovery procedures, operational accountability, and mechanisms for developer self-service.
Identifying gaps between teams
The guide highlights specific operational areas where responsibility often becomes unclear:
- Visibility into which clusters maintain a consistent access and network configuration, and documentation of who authorized any deviations
- Ability to diagnose application performance issues by examining dependencies beyond pod-level metrics—including databases, virtual machines, network infrastructure, and external services
- Verification of deprecated APIs and validation of recovery procedures for both cluster configuration and application data prior to each upgrade
- Clarity on certificate renewal responsibilities for ingress controllers
- Ownership of capacity change approvals
- Defined ownership, review schedules, and cost visibility for environments provisioned through developer service catalogs
These questions serve as structured prompts for discussions involving platform teams, application development teams, security teams, and business stakeholders. The ebook is available for download as teams prepare for these conversations.