Platform Engineering 2.0: your platform was built for a different era. AI just exposed it

The debate regarding the necessity of an internal platform has concluded. According to Google's 2025 DORA research, 90% of businesses now employ an internal platform, with 76% having established dedicated platform teams. The new question for IT leaders is no longer whether a platform should be built but whether the existing platform can withstand forthcoming changes.

The reality for most IT organizations is that it cannot - unless modifications are made. The platform constructed over recent years was intended for human developers to collaborate on containerized applications at a pace comprehensible to humans. However, this era has come to an end. Numerous factors have compounded over the past two years, revealing the platform's limitations.

At the root of every flaw is infrastructure, the developer-centric foundation that was never designed for provisioning GPUs on demand, managing AI agents, or enforcing cost control at the moment of resource allocation. The issue becomes evident when considering a primary example: most developers now utilize AI coding assistants.

The volume of code that can be generated and reviewed through AI-assisted development has surged, causing the bottleneck to shift. It is no longer about writing code but about delivering it. Your pipelines were never designed to handle such throughput, and the developer's role has quietly shifted from author to reviewer and coordinator of machine-generated work.

Another new type of user is emerging: AI agents. These non-human entities have increasingly required platform teams to accommodate them over the past decade. Agents require authentication, token usage regulation and management at an organizational and user level, GPU allocation, MCP compatibility, scoped permissions, non-human identity, audit logging, and strict safeguards on their actions.

Most platforms lack native solutions for these requirements. Cost represents the third pressure. According to Broadcom's Private Cloud Outlook 2026 study, 97% of IT leaders believe some of their public cloud spend is wasted, with 52% estimating waste exceeding 25% of their total public cloud budget. AI infrastructure exacerbates this issue significantly.

GPU instances, inference endpoints, and training jobs dwarf traditional expenditures. Furthermore, the cost associated with every prompt and retry dwarfs existing cost-reporting tools' capabilities. Retrospective FinOps reviews, monthly and quarterly cleanup exercises, are unable to detect misconfigured AI workloads that can exhaust budgets overnight.

Privacy, security, and sovereignty are the remaining concerns. AI introduces new extensive attack surfaces, including shadow AI sprawl, prompt injection, model poisoning, and inference data leaks, which no existing SAST or DAST scans in the pipeline were designed to detect. Additionally, the EU AI Act, US executive orders, and data residency rules impose further compliance requirements.

Platform Engineering 2.0 Evolution, not a Rebuild: The implications of these revelations are clear: none of them necessitate dismantling what has already been built. The foundations - Platform as Product, golden paths, shift-left security, self-service Internal Developer Platforms (IDPs) - remain pertinent. The discipline does not reset with each phase; instead, it evolves.

Platform Engineering 2.0 builds upon these foundations across five pillars, not as a complete overhaul. The first pillar is an AI-native platform. Your IDP transforms into an Agentic Development Platform (ADP), treating AI workloads as primary entities and AI agents as primary users within the same operating model. The second pillar is a multi-persona experience.

Currently, platforms excel in serving application developers while relegating others to a secondary status. This oversight leaves significant untapped potential. Security teams, data scientists, ML engineers, FinOps analysts, business leaders, and agents all require access to the platform, each with their own interface integrated with shared APIs.

The third pillar involves embedded FinOps. The shift is no longer from post-mortem reporting to decision-making at the time of resource provisioning. Every developer becomes a FinOps practitioner without formal training, as the platform highlights costs during action. The fourth pillar pertains to security, which has moved earlier in the pipeline but remains concentrated on developers, with many vulnerabilities surfacing in production.

Shifting security to platform and runtime layers ensures it remains invisible to developers and immutable by design. Lastly, the fifth pillar emphasizes composable design. The future lies not in choosing between build or buy but in composable integration. API-first, swappable building blocks allow for the replacement of individual CI/CD tools or observability stacks without cascading alterations.

The Cloud Native Computing Foundation (CNCF) projects ecosystem has grown from approximately 50 projects in 2018 to over 200 today; rigid platforms cannot accommodate this expanded range of choices. Where to begin: AI-native platform readiness: Platform Engineering 2.0 begins with upgrading the infrastructure foundation. Focus on addressing the platform's weaknesses sequentially rather than attempting to implement all five pillars simultaneously.

The first 12-month target for most organizations should be AI-native readiness: GPU workload support, non-human identity management, real-time cost attribution at provisioning time. To achieve this, start by auditing your current IDP for these three gaps: GPU/accelerator provisioning, non-human identity management, and real-time cost attribution during provisioning.

If all three are absent, your platform is already outdated. The teams responsible for constructing golden paths for developer autonomy now hold the keys to enterprise-wide agentic autonomy, marking the most extensive mandate the discipline has ever faced. The window of opportunity is open, and the necessary foundations are already in place.

The challenge now is to expand these foundations before the platform becomes the bottleneck it was designed to eliminate.