The insurance industry is in the middle of an AI gold rush, and not all that glitters is built for actuarial work. As agentic AI moves from hype to deployment, pricing teams face a decision that will define how they work for the next decade: adopt the right tool for the job, or inherit a black box at the heart of their most regulated process.
Not all AI is the same, and for actuaries, the distinction matters more than in almost any other profession.
Actuarial work carries a uniquely high bar on three dimensions that most professions do not have to balance simultaneously:
Most professions can afford to experiment with AI tools that are occasionally wrong or opaque. Actuaries cannot. That is why understanding the different generations of AI matters so much here, and why not all AI tools belong in an actuarial workflow.
To understand what agentic AI actually is, and why it represents a genuine shift rather than another buzzword, it helps to understand how we got here.
Level 1: Predictive AI
Predictive AI brought actuaries beyond Excel. These multivariate models predict everything from loss ratios to churn rates and underwriting tiers. This is where machine learning first made its mark in actuarial work, and where Akur8 led the transformation in insurance, replacing legacy tools with ML-powered pricing that combined statistical rigor with full transparency. What once required weeks of manual effort became faster, more accurate, and explainable by design.
Level 2: Generative AI
Generative AI expanded those capabilities from prediction to production, generating reports, summaries, code, and analysis from natural language inputs. But this generation of AI came with well-documented risks. As regulators and practitioners have noted, generative AI often operates with opaque decision-making processes and a degree of unpredictability, making it vulnerable to errors, hallucinations, and misuse. For actuaries, whose outputs carry regulatory and financial weight, that unpredictability holds a fundamental barrier to adoption.
Level 3: Agentic AI
Agentic AI is where the real transformation begins. These are systems that do not just respond to prompts; they plan, decide, and execute multi-step tasks with a degree of autonomy. The shift is significant: from answering questions to completing missions. According to McKinsey's State of AI 2025 report, 62% of organizations are already at least experimenting with AI agents, and 23% report scaling agentic AI systems within at least one business function. Yet in any given function, no more than 10% of respondents say their organizations are scaling agents broadly, which means most of the field is still figuring out how to make it work in practice.
Artificial intelligence is everywhere in insurance right now. Every platform claims to be "AI-powered”. Every vendor promises transformation. But when it comes to pricing, a critical question is being overlooked: does your AI actually understand how pricing models work?
Pricing is a statistical and actuarial science problem and requires deep knowledge of model structures like GLMs and GBMs, complex interactions, overfitting risks, and the regulatory frameworks that govern how those models can be used. Generic AI tools have no frame of reference for any of this.
Generative AI often operates with opaque decision-making processes and a degree of unpredictability, making it vulnerable to errors, hallucinations, and misuse. The result is a black box sitting at the heart of your most sensitive business process: an AI that acts with confidence, but cannot explain its reasoning in the actuarial terms that regulators, auditors, or your own professional standards require. For an actuary whose certification requires them to stand behind every assumption in a model, that is not a tool. It is a liability.
The problem becomes more acute with the wrong agentic AI. The risk is not just an incorrect answer. It is that without the right human checkpoints, a wrong answer propagates through a workflow before anyone catches it. A flawed assumption made early can compound across iterations, affecting model structure, variable selection, and ultimately the rates that reach policyholders. Add regulatory requirements around model explainability and filing documentation, and the stakes become even clearer: an agentic AI operating without actuarial grounding is a black box sitting at the heart of your most sensitive business process.
Agentic AI in pricing should not remove actuaries from the picture. It should simply free them from the tasks that slow them down, so they can spend more time on decisions that require their expertise. Done right, agentic AI takes on the execution; while the actuary manages judgment.
Here is what that looks like in practice:
According to McKinsey's State of AI report, AI agent adoption in the insurance industry is scaling fastest in risk, legal, and compliance functions (16% of respondents) and knowledge management (16%), all of which fall squarely within the scope of actuarial work. Yet the functions where actuaries arguably have the greatest impact: product and service development (tariff design, pricing) and strategy (capital modeling, profitability), show adoption rates of just 0-2% in insurance. That gap is not a sign that agentic AI is not relevant to pricing actuaries. It is a sign that the right tools have not existed yet. For actuaries who hold themselves to a high bar on transparency, auditability, and statistical rigor, generic AI agents have simply not been fit for purpose. That is exactly the gap Akur8 Agents are designed to close.
Most tools on the market are generic large language models repurposed for enterprise use, lacking any grounding in actuarial logic, statistical rigor, or regulatory context. Akur8 Agents are different by design. They run directly on Akur8's platform infrastructure, so outputs are not generated from guesswork but grounded in the same ML-native foundation that powers Akur8's pricing platform. Your prompts and data stay secure, never reused or fed into model training. Every action is visible, every result is traceable, and nothing changes without your approval. This is not AI adapted for actuaries after the fact. It is AI built from the ground up, curated by Akur8's actuarial and product teams, around the real workflows, edge cases, and standards that actuarial work actually demands.
Ludovico is Head of AI Transformation at Akur8. He completed a PhD in physics before entering the insurtech industry. He had been 5 years at Akur8 leading the Product & Engineering teams in Pricing. Now he is building Akur8's internal and external AI capabilities.
