Platform Engineering: The Backbone for Enterprise AI Adoption
The landscape of technology adoption is shifting, particularly concerning artificial intelligence. The era of top-down mandates for developer tooling is fading, as approximately half of all companies now foster AI adoption through a bottom-up approach, empowering teams to experiment with new AI development tools. For organizations slow to embrace this shift, the rise of “shadow AI”—unsanctioned tools used by employees—will soon necessitate formal adoption.
However, this newfound freedom also introduces significant risks and potential for inefficiency. Despite over three years of rapid AI integration, only 60% of organizations have established an acceptable usage policy for AI. Furthermore, while two-thirds of organizations have deployed AI tools into production environments, a similar proportion – 60% – still lack clear metrics to effectively measure AI’s impact. Moreover, the intense focus on the 20% of a developer’s day spent coding has revealed a surprising paradox: AI-generated code, while perceived as a productivity booster, can actually slow down developer throughput and compromise reliability.
This complex environment highlights the growing relevance of platform engineering, which emerged in recent years largely in response to the increasing complexity of modern technology stacks. Today, it offers a compelling solution to many AI adoption hurdles. Much like AI itself, platform engineering is most effective when it addresses the “toil” and other distractions that prevent software developers from delivering value to end-users.
It was therefore unsurprising that the fourth edition of PlatformCon in June heavily featured discussions on platform engineering within the context of AI. An internal developer platform (IDP), it turns out, can establish the ideal guardrails to foster AI innovation without disastrous consequences. Luca Galante, a prominent voice in the field, emphasized this point, stating that while AI captures headlines, “platforms for AI are going to be the backbone of all of this,” enabling the creation of enterprise-grade pathways to production for everything from data science and machine learning to traditional engineering.
The Age of AI necessitates an evolution of the IDP to encompass AI processes. This expansion will facilitate the scalable deployment of proven AI use cases across software development while breaking down data silos. By bridging these gaps, platform engineering is poised to ensure consistency, quality, and security in the delivery of both autonomous AI agents and generative AI applications.
Until recently, AI was largely confined to data science departments. Now, it needs to follow the trajectory of cross-organizational cloud adoption. Patrick Debois, co-author of the “DevOps Handbook,” argued at PlatformCon that AI requires a dedicated platform team. In this new era, the AI engineer emerges as a key change agent, bridging data science’s accelerated path to production by collaborating with the platform team on cross-team cooperation, enablement for data science and application teams, and robust governance.
Debois envisions an AI-enabled twist on the traditional internal developer platform, reorganized to include more AI stakeholders and expand its scope to manage: large language models (LLMs), whether open-source, proprietary, or hybrid; unstructured and structured data, requiring indexing via vector databases; RagOps (Retrieval-Augmented Generation as a Service), an emerging concept integrating third-party data sources; AI Agents as a Service, encompassing memory, state, access control, and exposure to Model Context Protocol (MCP) servers; execution sandboxes for AI agents; comprehensive access and version control across all model inputs and outputs; and a central caching layer to manage costs. All these components would be accessible through the platform’s “single pane of glass,” offering transparency and a unified view. Such a comprehensive, yet expanding, toolset further underscores the necessity of an IDP to manage it effectively. Debois suggests that new AI platform teams should begin by creating prototyping and sandboxing environments for safe experimentation with new AI tools. Once developers gain familiarity, a standardized framework aligned with existing languages and a robust ecosystem for caching, testing, and debugging can lead to more defined “golden paths” for AI development.
Debois also outlined four evolving “AI native developer” patterns: the shift from producer to manager, where developers manage code agents with support from operations; from implementation to intent, where developers express the ‘what’ and AI handles the ‘how’; from delivery to discovery, lowering the cost of experimentation through existing CI/CD pipelines; and from content to knowledge, as AI provides a compelling reason for teams to share knowledge, potentially making knowledge itself a unique value proposition for companies.
Like all product development, platform teams must consider their user base, which for AI extends beyond traditional developers. Ina Stoyanova, Staff Infrastructure and Platform at Equilibrium Energy, emphasized the need for organic expansion of native AI tools. At this early stage of AI, particularly for startups, rapid change makes rigid, permanent platform features a potential waste. By engaging stakeholders, Equilibrium’s platform team identified critical needs for both software engineering and data science teams, including cluster management, compute resources, data resources, data tools, storage, query analytics, and observability. Data science and quantitative analysis teams, however, also had unique considerations not initially on the platform engineering team’s radar. Stoyanova redefined platform engineering for her team as a “curated set of reusable tools, workflows, APIs and documentation that enable internal users to self-service infrastructure, environments and deployment pipelines with minimal cognitive overhead.” This user-centric approach, asking users “What are the tools that you want to use?”, allowed them to build the right solutions without over-investing or hindering the startup’s ability to adapt. Equilibrium Energy also prioritized cost tracking and metrics from the outset, a use case that resonated across both business and technical teams.
Leveraging AI effectively hinges on capitalizing on both structured and unstructured data. The internal developer platform serves as the scaffolding upon which data scientists and machine learning engineers can construct a data strategy to drive AI use cases. At PlatformCon, the Platform Engineering community announced a new reference architecture for AI, slated for publication later this year. This architecture provides a structured mental model encompassing observability, platform interfaces and version control, integration and delivery, data and model management, and security planes. As Luca Galante noted, this goes beyond technical changes, evolving how the industry perceives the platform engineering team itself.
Traditionally, platform engineering teams have comprised diverse roles, from heads of platform engineering to infrastructure, developer experience, and product managers, serving stakeholders including developers, executives, compliance, legal, infrastructure, operations, and security teams. In light of AI, the team structure and role development have expanded to include reliability, security, data and AI, and observability platform engineers. This broader team now interacts with an even wider array of stakeholders, such as site reliability engineering teams, architects, data scientists, and machine learning operations engineers, reflecting an increased granularity of specific needs in the market. There is no doubt that platform teams face an expanded scope in the age of AI.
A key role for modern platform teams is to identify and scale cross-cutting generative AI application delivery use cases, along with selecting appropriate design patterns (e.g., open versus closed generative AI models). However, Majunath Bhat, VP Analyst and Fellow at Gartner, highlighted at PlatformCon that the most pervasive AI challenge for platform teams is security and governance, often intertwined with cost implications. Since product teams may lack expertise in these domains, architects often provide subject matter expertise. To scale applications beyond mere prototypes, Bhat recommends establishing a generative AI center of excellence, or an “enabling team” as per Team Topologies. This team would work closely with product teams and platform engineering experts, providing specialized knowledge that can then scale. Bhat cautions against immediately building a shared platform, echoing Stoyanova’s sentiment: “Unless we understand what the different application needs are, it’s not appropriate for the platform team to assume that they understand what those needs are.” This approach, which might include “complicated subsystem teams” of AI experts, can further reduce the cognitive load on both application and platform teams.
A new paradigm is emerging in application security. While AI-generated code means a greater volume of code to scan and protect, autonomous AI agents can also aid in proactive and autonomous remediation. The internal developer platform is not only a conduit for rolling out new AI tools but also a crucial layer for establishing guardrails, managing role-based access control, and automating security checks. Sónia Antão, Senior Product Manager at Checkmarx, emphasized at PlatformCon that traditional application security cannot keep pace with more code, more contributors, and shorter timelines. She advocates for integrating autonomous AI agents with Application Security Posture Management (ASPM) directly within the Integrated Developer Environment (IDE) for real-time code security. This “intelligent shift left” allows AppSec to gain a clear, risk-aligned big picture of the application landscape, not just catching vulnerabilities early but resolving them faster and with confidence at the speed the business demands. This approach has led to a 25-35% reduction in vulnerabilities and a 69% improvement in response speed.
Generative AI can transform platform engineering into adaptive, intelligent systems that enhance developer productivity, reliability, and business alignment, as Ajay Chankramath, author of “Effective Platform Engineering,” discussed at PlatformCon 2025. He noted that generative AI has evolved from passive assistance to autonomous, intent-aware agents, enabling self-healing pipelines, real-time feedback, and personalized code suggestions. Key influences driving these shifts include Retrieval Augmented Generation (RAG), which grounds AI agents’ answers in real-time, contextual documentation; the Model Context Protocol (MCP), standardizing how LLM agents communicate with external APIs to encourage adoption; and the integration of generative AI within CI/CD pipelines, allowing for intelligent, self-correcting, and self-tuning processes.
Chankramath described an evolution from developers “finding it themselves” to standardized IDPs, and now to embedding AI agents directly into developer workflows. Operations have transitioned from ticket-based approaches to semi-self-service, and now to intent-based autonomous agents. The goal, he emphasized, is not replacement, but elevation: enabling developers and platform engineers to focus on higher-value activities. To support this evolution of AI-driven platform engineering, he offered five recommendations: align AI strategy with developer value streams, treating AI as integrated, flow-native components; always keep human judgment in the loop, ensuring agents propose, not approve, actions; make AI agents collaborative, allowing developers to override, retrain, and recontextualize them; build in observability and guardrails by default, including token trace logs, prompt drift detection, and relevance scoring; and expand AI impact measurement beyond accuracy and latency to include internal customer Net Promoter Score (NPS), sharing all learnings and metrics to prove benefits and increase adoption.
Matthew Vollmer, Head of Blink deep-code research agent at Coder, echoed the sentiment of “Golden Paths with Guardrails,” stressing that the goal is not just using agents, but using them wisely for productivity and safety. This requires providing agents with context (documentation, policies, codebases), responsible delegation (giving tools to senior developers first), and setting clear boundaries through isolated, ephemeral environments with strict access controls and usage limits. Embracing specification-driven development ensures AI agents perform exactly as instructed, avoiding risks or excessive costs. The sweet spot, he suggests, lies in assigning agents “self-contained, well-defined tasks” like small to medium bug fixes. An internal developer platform can facilitate this by onboarding AI agents like teammates. Vollmer shared an anecdote where an engineer described the experience as “pairing with a super fast junior dev who could write code at 100 times the speed of a human junior developer.” By offloading these “grunt tasks” to AI, teams can protect their innovation time, allowing developers to focus on high-value work.
Ultimately, both AI and platform engineering thrive where friction is high. Platform engineering aims to reduce developer cognitive load, a goal AI can significantly advance when implemented correctly. This synergy benefits not just individual developers but the entire software organization. According to the Atlassian State of Developer Experience in 2025, developers are already using AI-saved time to improve code, develop new features, and create documentation. When platform-driven AI adoption is executed effectively, it leads to even more time dedicated to value-driven activities.