AI Mainstreaming in Insurance But Skills Gaps Persist

- Multi-agent AI systems are being adopted in claims processing to reduce handling times from days to seconds and improve accuracy by 30% over single AI models. These architectures feature specialized agents for tasks like intake, documentation analysis, fraud detection, and damage assessment, which work together to automate complex workflows. This approach allows for scalable processing capacity, especially during high-volume events, and continuous improvement as each agent is trained on its specific function. - The transition to agentic AI requires a shift from monolithic systems to a composable, agentic micro-architecture built on cloud-native principles. Integrating AI with legacy insurance systems is a major challenge due to outdated technology, data silos, and a lack of API support. An API-first architecture is becoming essential, enabling modularity and faster, more controlled scaling of digital services by allowing new features to be deployed without altering the underlying core systems. - Insurtech venture capital funding has stabilized around $1.1 billion per quarter after significant volatility between 2020 and 2022. However, the number of deals is decreasing, with global deal volume falling 28% from 500 in 2023 to 362 in 2024, indicating that investors are becoming more selective. This trend is accompanied by a geographic shift in the US, with New York's share of global funding rising to 15% while Silicon Valley's dropped to 10%. - To address the skills gap, some firms are embedding junior AI talent with experienced coaches directly into their teams, focusing on practical application over just hiring senior external talent. While 92% of workers express a desire for generative AI skills, only 4% of insurers are providing reskilling at the necessary scale. Structured learning paths that combine university-level rigor with on-the-job application are being used to build specific competencies for roles in underwriting, claims, and operations. - In agentic AI systems for financial services, an orchestrator or supervisory agent manages workflows by assigning tasks to specialized utility agents. These systems use a "Reason and Act" (ReAct) loop, where agents analyze a problem, execute actions like API calls, observe the outcomes, and then determine the next steps autonomously. This architecture supports explainability and auditing through logged tool calls, retrieved context, and approval workflows. - While overall insurtech funding is stabilizing, a significant divergence is occurring between sectors. In the third quarter of 2025, investment in life and health insurtech saw a 56.8% decrease, whereas property and casualty insurtech funding surged by 90.5%. AI-focused startups are attracting the most significant investments, capturing 74.8% of all funding in Q3 2025. - The "strangler pattern" is a strategy being used to modernize legacy systems without a complete overhaul by creating a modern API layer that gradually replaces old components. This API-first approach allows insurers to connect core systems with external partners and new applications, fostering an open insurance model. This enables the development of modular products, like usage-based insurance, without disrupting the entire system. - A major hurdle in scaling AI is the lack of production-level MLOps skills and robust data engineering to handle messy, real-world data, as opposed to the clean datasets used in pilot programs. For LLM orchestration, best practices include programmatic generation of inputs to enforce user-based data permissions and semantic record-keeping of all inputs, outputs, and reasoning steps for auditability.