{"id":21971,"date":"2025-09-09T15:23:09","date_gmt":"2025-09-09T14:23:09","guid":{"rendered":"https:\/\/www.equalexperts.com\/?p=21971"},"modified":"2025-09-15T16:11:23","modified_gmt":"2025-09-15T15:11:23","slug":"experimenting-to-enabling-how-to-think-about-ai-when-platform-engineering","status":"publish","type":"post","link":"https:\/\/www.equalexperts.com\/blog\/our-thinking\/experimenting-to-enabling-how-to-think-about-ai-when-platform-engineering\/","title":{"rendered":"Experimenting to enabling: How to think about AI when platform engineering"},"content":{"rendered":"

AI is rapidly reshaping how software is built and delivered, and platform engineering is emerging as a critical enabler for scaling its impact across the enterprise. By supporting AI in delivery, platform teams can reduce developer cognitive load, improve governance, and turn early experimentation into reliable, secure innovation.<\/p>\n

Software delivery is changing so radically that approaches from even a year ago can feel outdated. With AI accelerating this shift, teams must constantly evolve just to keep up.<\/p>\n

Curiosity about tools like ChatGPT, Copilot, Cursor, or Windsurf has exploded into something bigger. Teams are no longer just experimenting with AI; they\u2019re scrambling to respond to AI-driven changes in delivery, developer experience, risk, governance, observability, and security.<\/p>\n

At Equal Experts, we’re regularly talking about the key role that platform engineering is playing in an increasingly AI-first world and why enterprise organizations need to enable AI at a platform level to maximize impact.<\/p>\n

The shift we\u2019re seeing<\/h2>\n

Everyone is wrestling with the pace of AI adoption and how to get from the \u201cWild West\u201d of experiments to sustainable, methodical approaches to delivering software with AI. Some of the AI-accelerated software delivery<\/a> work that we\u2019ve done with clients like Travelopia<\/a> and Defra is\u00a0already showing strong results and offering insights into the value of a structured, engineering-centric approach. Our former technical principal, Wes Reisz, recently shared some of his thoughts during a talk at the recent Equal Experts G[=]K25<\/a> event. Watch the talk: AI-assisted software delivery: Leveraging ChOP & LLMs to create more effective learning experiences.<\/a><\/p>\n

A recent conversation with an engagement manager in the US centered on the patterns we\u2019ve seen regarding AI adoption around platform engineering. The conversation ended up revolving around three areas where platform teams are supporting AI work:<\/p>\n

    \n
  1. AI in delivery<\/li>\n
  2. AI inside or AI as a part of the software that teams ship<\/li>\n
  3. AI tooling, a new evolving space where platform teams are being asked to support tools like Model Context Protocol (MCP),\u00a0 Agent2Agent (A2A), and Agent Communication Protocol (ACP)<\/li>\n<\/ol>\n

    Why is AI a platform engineering problem now?<\/h2>\n

    AI is no longer just making itself known in enterprise; it\u2019s already changing how we work. As PlatformEngineering.org<\/a> contributor Luca Galante<\/a> explained, \u201cAI isn\u2019t just knocking on the door; it\u2019s already walking into the room and rearranging the furniture of how we build and run systems.\u201d Google Cloud\u2019s research echoes this, reporting that 86% of organizations say platform engineering is essential to realizing AI\u2019s full value, and 94% consider AI \u201ccritical\u201d or \u201cimportant\u201d to the future of platform engineering.<\/p>\n

    In November of 2024, Anthropic announced the MCP specification<\/a>. MCP is a standard that, when implemented, allows an organization to expose resources, prompts, and tools to databases or other systems directly to an LLM. Already in its first few months, hundreds of these tools have popped up in places like hub.docker.com<\/a>. They are enabling access to hundreds of tools, from Jira and Confluence to GitHub to MongoDB, and they\u2019re landing in the enterprise right now.<\/p>\n

    This isn\u2019t about distant trends. It\u2019s about changes that need to be embraced and planned for right now. If the enterprise doesn\u2019t pay attention to these things, we\u2019ll quickly find ourselves in an era of cost overruns, security concerns, and governance\/compliance issues, similar to the issues we saw early on in the move to the cloud.<\/p>\n

    Platform engineering needs awareness of three AI forces<\/h2>\n

    AI in delivery<\/h3>\n

    AI is now woven directly into the software delivery lifecycle, including in requirements, planning, coding, testing, documentation, and operations. Practices like supervised (and sometimes unsupervised) agentic engineering are enabling teams to encode knowledge into prompts and code and automate what used to be painstaking manual work. Smaller teams are able to deliver faster, with fewer handoffs, and the DevEx platform\u2019s role of removing developer friction and reducing cognitive load must extend to support these delivery modes. The new demand? Platforms must support not just cloud or crosscutting resources but also structure guardrails, services, and even prompts around AI. Done well, this continues to be a multiplier for developer experience and value delivery.<\/p>\n

    AI \u201cinside\u201d the runtime<\/h3>\n

    AI isn\u2019t just supporting delivery. It\u2019s become part of the application runtime. Teams utilizing Retrieval Augmented Generation (RAG), ML models, and prompt-based agent usage directly into the products they are building. Machine learning is a first-class concern of many of the apps we build. Platform teams now face the challenge of supporting not just code, containers, CI\/CD, IaC, but increasingly models, pipelines, new types of monitoring, and all the components required for these AI-infused capabilities.<\/p>\n

    AI tooling<\/h3>\n

    AI tools like MCP (and A2A or ACP) are making knowledge work easier by connecting LLMs to data stores, but their unchecked use runs the risk of creating new forms of shadow IT (and all the risks that come with it). Developers and teams are wiring up MCP servers to automate workflows with access to GitHub, Jira, and Confluence, often without guardrails or governance. This is a realm for platform teams to consider. The danger is in over-permissioned agents, data leakage, and untraceable actions. We\u2019ve seen cases where an MCP agent, given too much scope, closed tickets and merged unreviewed code, which caused production incidents, not through malice but through lack of oversight caused by overaggressive use of AI.<\/p>\n

    The opportunity and the risks<\/h2>\n

    Every major change comes with both a concern and an opportunity. AI represents tremendous potential for enterprise software but can also expose us to more risks that need to be actively mitigated against.<\/p>\n