Platform Engineering • Ajay Chankramath & Nic Cheneweth • GOTO 2026
Ajay Chankramath and Nic Cheneweth discuss the critical elements of effective platform engineering, emphasizing a product mindset, the foundational role of control planes and API-first design, the common pitfalls of implementing Backstage, and the emerging impact of AI and agents on the platform landscape.
The Necessity of a Product Mindset
Treating an internal platform as a product is a fundamental shift that drives significant long-term rewards in productivity and sustainability. While many organizations have a mature product mindset for external, customer-facing applications, this discipline is often lacking for internal tools. The journey to a product-centric internal platform follows a pattern:
- Standardization: Establish “golden paths” to provide a consistent, well-supported way for the majority of developers to accomplish common tasks.
- Orchestration: Implement control planes to translate developer intent into concrete architectural and infrastructure outcomes.
- Contextualization: Adapt the platform to the specific needs of the business domain, whether it’s financial services or telecommunications.
- AI Application: Use supervised AI to enhance and optimize these established processes, rather than letting AI dictate the foundation.
Control Planes and API-First Architecture
A core principle of modern platform engineering is building everything with an API-first approach. The control plane is the orchestration engine that makes this possible.
- Developer Intent: The goal is to take a developer’s intent, such as “I need a Postgres database,” and translate it into a fully configured, secure, and compliant infrastructure component.
- Kubernetes as a Foundation: The Kubernetes API serves as an excellent foundation for a control plane. It provides a robust, extensible, and battle-tested reconciliation loop that constantly compares the desired state with the actual state, enabling self-healing capabilities. Using Kubernetes as the underlying engine prevents teams from having to reinvent this complex and critical functionality.
- Wrapping Cloud APIs: A control plane allows platform teams to wrap underlying cloud provider APIs (like those from AWS, GCP, or Azure) with organizational standards, guardrails, and conventions. This provides developers with a simplified, standardized interface while still allowing a mechanism for direct access to cloud provider features when necessary.
Backstage: The Promise vs. The Reality
Many organizations adopt Backstage as a developer portal, but the journey is often fraught with challenges. A common failure pattern is viewing Backstage as the platform itself, rather than a UI or a “touchpoint” for the platform.
- Architectural Foundation is Key: A successful Backstage implementation relies on a solid architectural foundation. If the underlying control planes, APIs, and domain boundaries are not well-designed, Backstage can become a “crap magnet,” where business logic is incorrectly placed into UI plugins.
- The Scaling Problem: Cutting corners on the backend architecture to quickly get a single team onboarded creates compounded problems at scale. Without a robust API layer, onboarding more developers and translating their intents into reliable outcomes becomes a significant challenge. The real issues surface not in the proof-of-concept but during the scaling phase.
Domain Design: The Foundation for Scale
When you build a platform with software, you must apply proven software engineering principles, chief among them being domain-driven design.
- Clear Boundaries: A platform should be composed of distinct domains with clean boundaries, much like a microservices architecture. For example, the team managing the underlying Kubernetes cluster (e.g., GKE) should operate independently from the team managing a service mesh deployed on top of it.
- Independent Evolution: These clear boundaries, exposed as well-defined APIs, allow different parts of the platform to evolve and be shipped independently without creating cross-team dependencies. This prevents the platform from becoming a monolithic bottleneck and enables scalability and maintainability.
AI and Agents in Platform Engineering
The application of AI in platform engineering is rapidly evolving, but it requires a cautious and foundational approach.
- Effective Use Cases: The most effective current uses involve tuning Large Language Models (LLMs) with company-specific data (metrics, logs, etc.) to act as a knowledgeable assistant. This AI can help with:
- AI-Enhanced Observability: Proactively identifying anomalies and predicting issues.
- Operational Assistance: Analyzing feedback from clusters to suggest improvements or flag potential problems, acting like an always-on, knowledgeable team member.
- The “Human in the Loop” is Critical: AI is a powerful tool, but its output must be validated. It can hallucinate or provide solutions that are superficial fixes rather than addressing the root cause. If you wouldn’t trust a human who gave you the same output, you shouldn’t blindly trust the AI.
- Warning Against Agentic Automation Without Foundations: A dangerous anti-pattern is jumping directly to agentic AI to generate infrastructure code without having established standards, golden paths, and control planes. This approach amplifies bad practices and creates a fragmented, unscalable, and drifting infrastructure. The foundational work must come first. You cannot build a solid house on a weak foundation, and AI cannot fix fundamental architectural flaws.