October 14, 2025
Agentic AI Enhances Core Systems – Why “SaaS Is Dead” Misleads Insurers
Guest article by Bill Song, CEO and Co-Founder of Peak3, Patrick Dahmen, Managing Partner at Valytics, and Vincent Wild, Business Development Director at Peak3
Agentic AI systems will not replace core insurance systems but will enhance them. In a highly regulated environment, the greatest value lies in hybrid architectures that combine a deterministic core for persistence and compliance with intelligent, modular AI components.
1. “SaaS is dead” – but what does that mean for insurance core systems?
The hype around agentic AI is leading many to wrongly assume that traditional core systems are on their way out. In reality, they are gaining importance because they provide the necessary stability, traceability, and persistence that make AI-driven innovations scalable and regulatory-compliant. The fear that IT systems like policy administration systems could be replaced by AI agent systems was significantly fueled by an interview with Microsoft CEO, Satya Nadella.
According to Nadella, in the era of agents, the assumption that standalone business applications are central is being overturned: business logic moves into an AI layer that performs CRUD operations (Create, Read, Update, Delete) across systems and databases. Once this logic is embedded in the AI tier, policy administration systems become largely interchangeable.
This thesis has circulated widely since the introduction of generative and agentic AI approaches. However, in the insurance sector, it falls short. Insurance operates in heavily regulated markets, where decisions must remain consistent, reproducible, and traceable over decades. Life and health insurance, in particular, rely on long-term commitments; they require data persistence, robust histories, and auditable processes. Systems acting as a "black box" or whose outputs aren't explainable are therefore unsuitable for core processes.
2. Why core systems remain essential
Replacing tried-and-tested policy administration systems – some historic, some modernized, often COBOL-based and regulation-proven – with pure AI agent systems would involve significant operational, regulatory, and reputational risks. At the same time, the technological foundation of these systems has evolved: rather than being monolithic and sluggish, they are now increasingly modular, API-first, and event-driven architectures in which AI components can be securely and seamlessly integrated and orchestrated.
This means: the core system is not disappearing – its role is becoming more clearly defined. It remains the stable, deterministic foundation upon which agentic AI operates.
Insurance companies also operate in regulated markets where decisions must remain traceable over decades. Black-box models without explainability are therefore unsuitable for core processes.
Efficiency through hybrid architectures
For highly standardized routine processes – such as billing, simple contract renewals, or standardized premium adjustments – deterministic rule engines continue to be superior. They are faster, more reproducible, and more cost-efficient because they model clear, auditable logic. Agentic AI delivers value where unstructured data, complex contexts, and uncertainties dominate – for example in fraud detection, complex claims processing, medical document handling, and dynamic customer interactions across channels.
A practical approach is to use AI in a “copilot” role: generating decision proposals, identifying anomalies, suggesting new rules, and prioritizing tasks.
Why core systems stay
Core systems in insurance are more than just booking engines – they are the memory of the organization. They manage contracts, premiums, benefits, coverage, and histories over long periods. This persistence requires deterministic, “sealed,” and auditable paths. Distributed agentic AI components, without clear orchestration, increase the risk of state inconsistencies. This is precisely why the core system remains the single source of truth.
Further, mass scalability, disciplined release cycles, backward compatibility, and regulation-proof migration paths are non-negotiable in insurance operations. They form the stable foundation on which AI innovations can be operated safely and at scale.
3. What does this mean for insurance companies?
Test modularity principles: Restructure the core system to allow AI agents to be easily integrated or replaced. This particularly involves the use of MCP (Model Context Protocol) – the “API for AI models” and new industry standard that enables AI models to access external systems or data sources via standardized connectors, and thus execute entire business processes.
Deploy AI pilot projects selectively: Identify low-risk use cases (e.g., fraud triage, voice/chatbots) and gather experience.
Strengthen governance and compliance-readiness: Embed transparency, auditability, and explainability into all AI decisions.
Use cases for hybrid architectures in insurance
Many leading insurance groups are already actively shaping the transition to hybrid architectures. Examples include:
Claims: In document processing, AI models can extract and summarize unstructured information from assessments, medical reports, and correspondence, linking it to policy logic. AI generates coverage and liability suggestions, flags inconsistencies, and prioritizes cases with potential fraud risks. Human adjusters review the AI suggestions, and the core system executes payments, reserves, and notifications deterministically. This reduces processing times, improves justification quality, and gradually increases the straight-through processing (STP) rate.
Underwriting: For complex risks, AI can prefill data from internal and external sources, uncover contradictions, and improve documentation consistency. Acceptance criteria suggestions are clearly reasoned and traceable. Final underwriting decisions rest with the underwriter, and the core system reliably implements the policy, including pricing logic, clauses, and limits.
Customer service and broker support: AI assistants can aggregate inquiries, summarize case histories, and suggest next-best actions. They support the creation of regulatory-compliant documents and records. The core system ensures that all transactions, consents, and deadlines are properly captured and adhered to. Customers and distribution partners experience a faster, more consistent service.
4. Conclusion: Agentic AI as an accelerator for modern core systems
Agentic AI is not the end of policy administration – it is its accelerator. Those who combine flexible core systems with intelligent peripheries gain in speed, quality, and resilience.
However, in many cases, legacy static core systems are more of a hindrance than an enabler – especially when it comes to meeting future demands.
About Peak3
Whitepaper
Whitepaper
Leveraging AI to automate processes and mitigate fraud
