The platform engineering stories enterprise IT leaders actually read.
Volume 341, Issue 4 · 10 articles from 5 issues
Platform-led IoT engineering speeds delivery and cuts cost by reusing standardized capabilities.
Traditional project-based IoT development rebuilds infrastructure repeatedly, creating technical debt and scaling problems. Platform-led approaches use shared foundations for device connectivity, security and lifecycle management, letting teams focus on business-specific features.
Organizations adopting the model report 75% code reuse, 50% lower ownership costs and 40% faster product delivery. Proven deployments span over 100 million connected devices.
Autonomous agents need overlay networking for stable connectivity across distributed endpoints and NAT.
As AI agents become production platform users running across CI runners, cloud functions and laptops, traditional networking assumptions break down.
Overlay networks provide stable virtual identities, encrypted transport, NAT traversal and discovery for agents.
Platform teams should evaluate solutions on deployment simplicity, trust controls, open-source visibility, SDK support and MCP compatibility. The goal is secure, least-privilege agent communication.
Internal developer platforms fail because they prioritize infrastructure elegance over user needs and lack product discipline.
Most platform initiatives fail not due to technical limitations but because they treat developer platforms as infrastructure projects rather than products.
The author argues that platforms must reduce cognitive load, earn trust through positive user experiences, and measure success by adoption metrics and developer satisfaction rather than infrastructure outputs.
Without product ownership, user research, and continuous feedback mechanisms, platforms become expensive systems nobody adopts, often leading teams to build shadow infrastructure instead.
AI has removed code generation as the delivery bottleneck, shifting the real constraint to review, testing, and deployment.
The article argues that AI has eliminated code generation as the limiting factor in software delivery, moving the constraint to review, testing, and deployment.
IDC research cited shows 60% of developers now generate more than a quarter of their code with AI, while acceptance rates without revision have declined significantly, prompting IDC's Adam Resnick to note that the center of gravity has moved from generation to review.
AI-generated code often looks polished while hiding architectural, security, and dependency problems, requiring extensive human oversight.
Organizations are responding with internal developer platforms offering standardized infrastructure, self-service provisioning, and golden paths that embed governance into deployment, with 81% expanding or piloting such platforms.
Platform Engineering 2.0 is not one stack but several composed layers serving developers, data scientists, security teams, and AI agents.
The article argues that Platform Engineering 2.0, driven by AI adoption, comprises multiple integrated layers rather than a single standardized stack.
The experience layer uses Backstage-style portals that serve as front doors rather than complete platforms; the workload control plane is likely Kubernetes, orchestrating applications and AI workloads across environments; the infrastructure foundation handles compute, accelerators, and resource isolation through virtualization or public cloud; and an AI control plane manages gateways, model registries, agent governance, and token consumption.
The author emphasizes that the platform is the coherent product created when those layers are composed around organizational outcomes.
Platform teams should treat autonomous AI agents as first-class users with the same golden paths, RBAC, and audit trails as developers.
As AI agents move beyond code generation into provisioning, deployment, and operational tasks, the article argues platform teams must adapt their infrastructure accordingly.
Agents should get the same golden paths, role-based access control, and audit trails as human developers, exposed through machine interfaces rather than UIs, so they cannot circumvent established processes.
Three governance controls become mandatory: platform-defined authentication with inherited RBAC, complete action auditing, and spending limits, since an agent stuck in a loop burns budget as fast as compute.
AI should remain the execution layer rather than the source of business judgment, operating within human-defined boundaries to prevent policy drift and conflicts between competing agents.
Broadcom's Pankaj Gupta outlines Platform Engineering 2.0 as an operating framework for AI-native organizations built on five pillars.
In this podcast discussion, Broadcom's Pankaj Gupta describes Platform Engineering 2.0 as an evolution of internal developer platforms reshaped by AI's impact on software delivery.
Developers increasingly focus on validating, governing, and safely moving code to production rather than writing it, which means platforms must serve both human and non-human users, including AI agents, with guardrails, permissions, observability, and governance.
The platform audience now extends to security, compliance, FinOps, data science, ML engineering, and business leadership. Five pillars anchor the framework: AI-native capabilities, multi-persona experiences, embedded FinOps, integrated security, and composable design.
Developer trust, not platform adoption, determines whether platform engineering investments deliver productivity gains.
While 80% of large organizations now have platform engineering teams, fewer than 30% achieve measurable developer productivity gains because developers quietly bypass platforms they do not trust.
The real challenge is designing genuine golden paths that are objectively faster than manual alternatives, then measuring voluntary repeat usage rather than onboarding numbers.
Platforms should be treated as products with clear ownership and transparent roadmaps. Teams that earn adoption involve developers early, perfect one workflow before sprawling, and visibly close feedback loops by publishing outcomes and roadmap changes.
Agentic platforms consolidate security, FinOps, and AI into unified enterprise infrastructure.
Agentic development platforms are integrating security, FinOps, and AI infrastructure into a single enterprise substrate, reshaping how platform teams operate.
The sector gained significant momentum at Black Hat 2026, which showcased agentic SOCs, identity solutions for agents, and automated vulnerability remediation.
New reference architectures stack agent infrastructure and path specification layers beneath traditional tooling planes. This lets organizations serve both humans and agents while blending deterministic CI/CD with probabilistic AI skills across the enterprise.
Agent chains ratchet permissions upward because credentials lack built-in attenuation; platforms must enforce it.
As AI agents call tools and other agents in chains, an initial permission grant can persist or widen rather than narrow, creating what the author calls a scope ratchet.
Traditional OAuth standards do not require downstream tokens to be narrower than upstream ones, which enables privilege escalation. The proposed fix is capability tokens with cryptographic caveats that enforce attenuation structurally, at the cost of larger tokens, harder revocation, and key management overhead.
For simple fixed service graphs, scoped service accounts with mutual TLS remain preferable, but dynamic agent chains need stronger controls built into the platform.
More from Platform Engineering · All IT section newsletters · This week