The author, a CTO, identifies parallels between current AI adoption and past technology shifts, warning against repeating mistakes. Initially, his company experienced uncontrolled AI spending due to employees individually subscribing to frontier models. This led to high, unmanaged costs and a lack of understanding regarding token expenses. Many companies are creating numerous, independent AI teams, each selecting their own models and tools, leading to inefficiency.This situation mirrors the early days of DevOps in 2009, where small, independent teams initially thrived but eventually created a tangled mess of disparate tools and technical debt. To address this, platform engineering emerged, establishing "golden paths" for developers and consolidating tools. Similarly, the rise of cloud computing brought about FinOps to manage escalating costs, as developers often chose the fastest, most expensive solutions.Today, companies are recognizing the need for consolidation in AI tools and models, much like they did with DevOps and cloud. The recommended solution is to build AI capabilities incrementally, starting with a core project and then building subsequent projects upon that foundation. This approach, akin to platform engineering for AI, creates a unified system and avoids the "pile of clothes hangers" — disparate tools and technical debt. By learning from past mistakes, companies can minimize the blast radius of their AI implementations and prevent costly inefficiencies.