Skip to Content
Nextra 4.0 is released πŸŽ‰
Disclaimer

Disclaimer

Technical Preview

Agents at Scale - Agentic Runtime for Kubernetes (β€œArk”) is released as a technical preview and early access release. This software is provided as a Request for Comments (RFC) to share elements of our technical approach with the broader technology community, gather valuable feedback, and seek input from practitioners and researchers in the field of agentic AI systems and Kubernetes orchestration.

As a technical preview release, this software may contain incomplete features, experimental functionality, and is subject to significant changes based on community feedback and continued development. The software is provided β€œas is” without warranties of any kind, and users should expect potential instability, breaking changes, and limited support during this preview phase.

API Stability and Migration

This technical preview alpha release does not provide a stable API, and migration of existing data will not be supported until the first general availability launch. However, since Ark functions primarily as an operational layer, the recommended approach is to define your resources declaratively in YAML files or store configuration data in your application layer database, decoupled from Ark itself. This enables you to recreate resources in Ark as needed, allowing the system to reconcile them without data migration concerns.

Security

Ark is architected to support enterprise-grade security through layered controls that can be implemented above and below the core runtime. The out-of-the-box configuration prioritizes developer experience and ease of evaluation, with security layers intentionally left unconfigured to allow organizations to implement their specific security requirements.

β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β” β”‚ Enterprise Security Layer β”‚ ← Authentication, Authorization, β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€ Network Policies, Secrets Mgmt, etc β”‚ Ark Runtime Core β”‚ ← Agent orchestration, Model mgmt β”œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€ Query execution, Memory storage, Evaluations β”‚ Infrastructure Security Layer β”‚ ← Cluster hardening, Node security, β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜ Compliance controls, Monitoring, etc

The default configuration includes:

  • Open cluster access for rapid prototyping and demonstration
  • Simplified service-to-service communication for development workflows
  • Standard Kubernetes RBAC that can be restricted as needed
  • Extensible authentication and authorization integration points
  • Configurable network policies and service mesh integration
  • Pluggable secrets management and encryption capabilities

Ark’s design philosophy enables organizations to implement their required security posture through standard Kubernetes security primitives, enterprise identity providers, network security controls, and compliance frameworks. The platform provides the necessary extension points and configuration options to support defense-in-depth strategies without prescribing specific security implementations that may conflict with existing organizational standards.

Security-Conscious Design

Ark’s architecture and design decisions prioritize compatibility with Kubernetes enterprise security practices. Rather than implementing proprietary security mechanisms, all components are designed to work seamlessly with existing Kubernetes security primitives, enterprise identity systems, and network security controls. This approach ensures that organizations can apply their established security policies and tooling without architectural conflicts or security gaps introduced by the platform itself.

Third Party Vendors, Services and Products

Any reference to third-party vendors, services, or products in this technical preview is provided for informational purposes only and does not constitute an endorsement, recommendation, or approval by McKinsey & Company. Inclusion of such references does not imply any affiliation or partnership, and McKinsey & Company makes no representations or warranties regarding the quality, reliability, or suitability of any third-party offerings mentioned.

Last updated on