Agentic AI in the insurance industry: How to successfully shift from 'chatting' to 'acting'

Over the past two years, generative AI has taken the insurance industry by storm. But anyone still discussing whether and how GenAI should be used in 2026 is missing the real revolution: the Agentic Shift.

In this blog post, I will show you why the move to the Agentic Enterprise is primarily a business and governance issue, why your organisations need to become ‘agent-ready’ now, and why you need a strong partner at your side to achieve this.

From assistant to actor: what's behind Agentic AI

After the ‘predictive’ AI models of the first wave and the GenAI chatbot wave since 2023, one thing is clear: generating text is no longer enough. The hype is subsiding and the question of real business value is becoming louder.

The next stage in AI evolution is Agentic AI: an agent has a specific goal, breaks this goal down into individual steps, independently calls up systems, tools, APIs and data sources, makes predefined decisions within clear guidelines and documents its actions in a traceable manner.

The short formula for 2026 is: LLM + tools + tasks = AI agent.

Instead of ‘Can you summarise the policies for customer X for me?’, in future it will be: ‘Please prepare a proposal for a contract adjustment for customer X based on previous behaviour, claims history and our current campaigns – and put a finished recommendation in the CRM and output system for me.’

This will make agents the new ‘middleware’ between business and IT: they will translate business objectives into technical actions – and back into usable results.

Why this is a game changer for insurance companies – especially in claims and benefits assessment

In claims and benefits management, insurers differentiate themselves from the competition through speed and quality.

A well-designed agent can automatically prepare the initial assessment of insurance claims, process standard cases almost to completion, relieve the burden on claims adjusters so that they can concentrate on complex, contentious cases, and at the same time learn from each case within clear governance boundaries.

In combination with generative approaches that analyse unstructured data (expert reports/images), massive efficiency and quality levers are created along the claims value chain – from claims reporting and coverage assessment to fraud prevention.

The real hurdle is not technology – it is organisation, culture and governance.

But the most exciting (and challenging) part comes now: the agentic shift is not purely an IT project, but an organisational and cultural transformation.

We are moving towards a world in which humans and agents work together as a hybrid workforce. This raises key questions: Who actually ‘manages’ an agent? Who is responsible when an agent performs a process step? What does a role landscape with ‘agent owners’ look like?

One thing is clear: in five years, traditional clerks will increasingly become agent managers, with agents taking over data searches, preliminary checks and suggestions. Humans will focus on quality control, decisions in borderline cases and continuous ‘fine-tuning’ of agents. This is not a job reduction, but a skill shift. It must already be reflected today in a modern, technology-oriented HR approach – with new role profiles, learning paths and career models.

And the most important success factor: employees must understand agents as new team members from the outset, not as competitors. This requires active change management and training opportunities.

Governance by design – not as an afterthought compliance patch

As soon as agents act autonomously, the question immediately arises in the insurance world: ‘How do we explain to the auditors or the Financial Services Authority why the agent has granted or rejected cover here?’ The answer cannot be: ‘That's how the model was designed.’

Governance by default – i.e. relying on the standard settings and policies of hyperscalers or tools – is not enough. These cover basic security and infrastructure aspects, but not your technical rules, roles, approvals and liability issues. Governance must therefore be consciously incorporated into use cases, processes and architecture and must not just ‘run along’.

The answer is therefore governance by design. In concrete terms, this means:

• Transparent decision-making logic: audit trails, reasoning logs and confidence scores make decisions verifiable and explainable.
• Controlled autonomy: Risk classes for use cases – from ‘assistance’ to ‘highly critical decisions’ – as well as clear guardrails for when an agent can make its own decisions and when human-in-the-loop is required.
• Rules and processes: Governance in line with the EU AI Act, data protection and BaFin expectations, and processes for monitoring and incident handling.

This creates controlled autonomy: agents act, but always within a framework defined and accounted for by the company. Regulation does not become a brake on innovation, but rather a competitive advantage: those who can demonstrate verifiable AI processes build trust and gain speed because projects are not stuck in endless loops with legal and compliance teams.

Becoming agent-ready: processes, data & cloud, and the role of connectors

The uncomfortable truth is that AI solutions in insurance rarely fail because of the model, but because of the historically grown IT landscape.

To ensure that agents are capable of acting and can be cleanly integrated, you need an agent-ready architecture with three cornerstones:

Clearly defined processes and connectors

End-to-end, agent-ready processes mean that you know and document your processes from start to finish and translate them into agent workflows – including the necessary decisions, data accesses and system calls.

Smart, reusable connectors to inventory and claims systems are what really make agents capable of acting. They ensure secure read access to all relevant contract and case data and allow clearly regulated write access (e.g. setting statuses, triggering workflows).

Clean data models & knowledge management

Agents need consistent, business-driven data models, central data platforms and knowledge repositories (e.g. cleanly managed Sharepoints) that they can access in a targeted manner. Clear rules for access rights, deletion and retention ensure that actions remain scalable, secure and audit-proof.

Cloud & hyperscalers as enablers

Only with a modern cloud landscape can you use existing hyperscaler agent platforms – and build scalable, secure agent solutions on top of them, instead of getting stuck in isolated solutions and on-premises limitations.

But building the perfect target architecture is a process that takes years. To remain operational in the meantime, we can build smart ‘workaround connectors’ that rely on regular data exports (e.g. to a cloud data lake) and a few clearly defined return channels to inventory and damage systems – this allows agents to read and write today, while your IT department works quietly in the background on the long-term API strategy.

How to get off to a good start: focus use cases and mini PoCs

Perhaps the most important recommendation: please don't start by trying to build ‘agents for everything’. Successful organisations start with a focus, with clear, manageable use cases and a realistic proof of value.

• Sharpen your use cases: Select one to three clearly defined use cases along the value chain (e.g. a specific claims/legal protection case) and evaluate them according to business value, feasibility, data availability and regulatory risk.
• Pilot with guard rails: Build a mini PoC for these use cases with clear KPIs (e.g. processing time) and embed governance by design from the outset – including human-in-the-loop principles.
• Scalable platform instead of individual cases: Transfer successful agents to a platform with reusable building blocks (connectors, agent blueprints) and establish an agent lifecycle from idea to operation.

What adesso brings to the table: Business, people, technology – the bridge builder

The agentic shift hits the sweet spot where business, people and technology must interact – and this is precisely where adesso positions itself with a focus on the insurance industry.

Business: In-depth insurance and regulatory expertise

We know the specialist processes of insurers in detail – including their regulatory embedding. We translate agentic concepts into concrete product and process strategies that fit in with your business goals. We support you in developing target visions for the agentic enterprise – including prioritising the most value-adding use cases.

People: Change, skills & future operating models

We support you in developing hybrid operating models – with clear roles for humans and agents. We help build AI and agentic competencies in specialist areas, IT, governance and HR. We support you in creating acceptance, security and enthusiasm among employees.

Technology: Architectures, Hyperscalers & TrustworthyAI

We design and implement cloud and data architectures that connect your existing core systems with LLMs, agents and new digital services – including hyperscaler integration. With our Trustworthy AI and compliance approaches, we combine technical possibilities with the requirements of the AI Act, BaFin and data protection. We work with you to build connectors and bridges so you can quickly become operational, and we bring practical experience from GenAI projects in the insurance industry – from knowledge agents and input management to automated text and document processes.

In short, we help you turn ‘We need to do something with agents’ into a concrete, viable agentic transformation programme – with a clear focus on value contribution, governance and scalability.

Conclusion: 2026 is not the time to ‘try things out,’ but to take action

In 2026, the question will not be whether you use agents, but how well they interact with your core systems, your employees and your regulatory requirements.

My plea:

• 1. Think holistically – towards Agentic Enterprise
• 2. Start focused – with clearly defined use cases
• 3. Build the foundation – processes, organisation, cloud and governance by design
• 4. Identify early adopters – use multipliers from the specialist areas for mindset, acceptance and the scaling of successful agent use cases.
• 5. Bring strong partners on board – for business, people and technology