Summary

• Akur8 Agents can guide users through its pricing platform and automate repeatable workflows without taking control of actuarial decisions.
• The underlying mathematical and statistical modelling remains within Akur8’s established actuarial engine rather than being handed wholly to a general-purpose language model.
• The launch demonstrates one approach to regulated agent deployment by bounding automation inside specialist software and preserving expert approval.

Insurance software provider Akur8 is putting AI agents into actuarial pricing, but its implementation keeps the underlying mathematics and final decisions with specialist software and human actuaries rather than handing the process wholesale to an autonomous model.

The Paris-founded company launched Akur8 Agents for pricing at the end of September, adding guidance and workflow automation directly to the platform insurers use to analyse risk and build pricing models.

Akur8 says the agents can help users navigate the platform, automate repeatable tasks, surface information, and apply predefined actuarial skills, while actuaries retain control of analyses and decisions. Its existing actuarial and statistical engines continue performing the underlying mathematical work.

That separation provides a useful example of how agentic AI can be introduced into regulated professional work without assuming that every calculation and judgement should be delegated to the same probabilistic system.

Automation moves around the model

Insurance pricing involves considerably more than fitting a model to historical claims data, because actuarial teams also prepare information, validate assumptions, segment portfolios, compare models, document decisions, examine regulatory constraints, and translate technical outputs into rates that commercial and underwriting teams can use.

Software has automated parts of that process for years, while machine learning has widened the range of models actuaries can test. Agents add a different layer because they can coordinate multiple steps and respond to user instructions rather than carrying out one predefined calculation at a time.

Akur8’s approach keeps that orchestration separate from the engines responsible for actuarial mathematics. A language model can interpret an instruction, retrieve information, or decide which workflow action should come next while a specialist statistical system remains responsible for producing the numerical model.

That boundary is particularly relevant in insurance because pricing decisions can affect portfolio profitability, regulatory compliance, competitive position, and customer outcomes across large numbers of policies. A plausible explanation from a language model is not an adequate substitute for a reproducible calculation.

Insurers also operate under requirements around model validation, documentation, fairness, governance, and auditability, which make complete delegation to an opaque autonomous system difficult to defend. Keeping the established modelling engine underneath the agent provides a clearer line between generative assistance and the calculations on which prices depend.

Control becomes part of the product

Akur8 has made human review explicit in the product design, with its current agent material describing visibility into agent actions and approval before changes are executed. That places governance inside the workflow rather than treating oversight as a separate policy document.

The language reflects a broader change in the market for enterprise agents because early products often competed on the amount of human work they could remove, while systems moving into finance, healthcare, insurance, and other regulated sectors increasingly compete on how precisely that autonomy can be bounded.

Buyers need to know which actions require approval, what information an agent can access, how changes are logged, whether outputs can be reproduced, and where deterministic software takes over from generative reasoning. Those controls influence whether an agent can move from a demonstration into work that affects customers or financial outcomes.

Akur8 already has a sizeable deployment base on which to test the model, saying more than 3,000 actuaries use its platform across more than 350 customers in over 40 countries. Company figures do not yet show how much time the new agents save, how often users reject proposed actions, or whether pricing cycles become materially shorter.

Those measures will determine whether agentic actuarial software becomes more than another interface layer. The benefit depends on automating enough repeatable work to free specialist capacity without creating a second workload in which actuaries spend the saved time checking unpredictable automation.

The architecture is nevertheless notable because it avoids the assumption that maximal autonomy is always the desired end state. In Akur8’s implementation, the agent sits around specialist software, established statistical engines retain the mathematical work, and qualified professionals remain responsible for the outcome.

That division of labour may prove a more practical pattern for agents entering regulated work, where the commercial opportunity comes from reducing repetitive process rather than removing the expert whose judgement the regulation and the business still require.