Credit: Andrei Hasperovich/Adobe Stock
For the last two years, the conversation around artificial intelligence has mostly revolved around the breakthroughs and maturing of generative AI.
As months go by and models evolve they more and more become a trustable destination for knowledge and reasoning augmentation.LLMs draft emails, summarize documents, break down problems. Tools layered on top of those constantly-improving models generate images, code software and publish marketing copy.
The insatiable demand for inference computing is growing as announcements from the big technology players pile on about record investments, infrastructure deals, chips and services partnerships. As frenzy builds up, the debate has sounds strangely familiar: Is AI overhyped? Is it a bubble? Will it disrupt, replace, displace jobs? Those questions miss the real shift already underway.
Generative AI was only the long-awaited opening act. The next phase, AI with Agency, also more commonly known as agentic AI, is where the transformation accelerates. For the industries of risk management, the difference between the two matters more than most people currently realize.
Generative AI: The augmentation layer
Generative AI is fundamentally an assistive technology, meaning it helps humans create content faster and process information more efficiently. Claims adjusters can summarize reports in seconds.
Underwriters can review documentation more quickly and more thoroughly leaving no stones unturned. Customer service agents can draft relevant and fact-verified responses almost instantly. The overall quality of an industry's output from now on stems from its ability to integrate artificial intelligence in workflows at the right time. All of that is already valuable. Productivity gains are real. Announcements around layoffs enabled by AI-driven efficiencies are welcome news moving the markets these days.
But generative AI still relies on humans to drive the workflow. It produces output based on prompted guidance and people decide what happens next with the output. In other words, generative AI alone accelerates work and helps with tasks. It doesn't run tasks. Not at this level. Agentic AI changes that equation.
Agentic AI: The operator layer
Agentic AI systems are designed not just to generate content but to execute tasks toward defined goals. Instead of answering a question, an AI agent might complete a multi-step workflow.
For example, a claims agent might currently ask an AI tool to summarize an accident report or perform a risk assessment. With an agentic AI layer added, the system could analyze the report, verify coverage, retrieve policy data and customer activity, flag anomalies and prepare a preliminary decision in the form of a recommendation, all this before a human has to review the output file and take accountability through a decision or continue executing on a course of action.
In underwriting, agentic systems could monitor incoming data, evaluate risk signals, cross-reference regulatory requirements and recommend policy adjustments in real time. These systems don't replace professionals.
They function more like digital team members assigned to specific operational roles. The prime candidate for task automation is what we commonly characterize as the "busy work" of individual contributors. The difference between the two layers may feel subtle, but it represents a fundamental shift in how work gets done.
Why the risk-management industries will feel this shift quickly
Insurance is built on structured workflows: underwriting decisions, claims processing, compliance checks, fraud detection and customer servicing.
Those processes are mostly case-by-case situations, they generate large volumes of data and rely on repeatable decision patterns. It is the exact environment where agentic systems thrive. For example, imagine an ecosystem of specialized AI agents working together where one agent reviews policy language, another analyzes claims history, a third evaluates fraud signals and a fourth prepares regulatory documentation. Each agent handles a narrow task but collaborates with the others to move a case forward.
This "multi-agent" approach is already emerging across industries, where groups of deployed AI systems coordinate well-defined and delimited tasks much like teams of experts within departments inside a company. When deployed responsibly, it can significantly reduce operational friction without eliminating the ultimate need for a human near the end of the value chain. Human consciousness instills purpose, impulse and defines guardrails upstream of an AI Agent's action and it is also looped back downstream for compliance, scalability and accountability.
The real risk isn't the technology
The biggest mistake organizations make right now isn't adopting AI too quickly. It's adopting it without clarity in the goals being pursued.
Many companies are deploying generative AI tools simply because competitors are doing it. They add chatbots or copilots but fail to start from updated job descriptions for the impacted roles and they often omit redesigning the workflows around them. The result is a patchwork of tools that produce impressive demos but limited business impact. Agentic AI forces a more disciplined approach.
Before building an AI agent, organizations must answer basic questions:
What job is this system responsible for? That's purpose.
What does it know? That's knowledge and awareness.
What is it capable of? That's giving it arms and legs.
And where does human oversight remain essential? That's supervision and reinforcement.
Without those answers, AI becomes noise in a company instead of infrastructure.
Humans remain the operating system
There's a persistent narrative that AI is about replacing people. In reality, the next phase of AI is about amplifying decision-makers, not eliminating them. Insurance professionals bring judgment, ethical reasoning and contextual understanding that machines simply do not possess.
Agentic AI systems work best when they handle structured tasks and surface insights, while humans retain responsibility for interpretation and accountability. Think of it less as blind automation and more as intentional operational leverage.
The conversation businesses should be having
The AI bubble debate focuses on whether the technology is overvalued. That's the wrong question. The more important shift is structural: businesses are currently moving from AI-powered tools that generate answers to Agentic-AI systems that execute real work.
Generative AI changed the expectations we have when interacting with machines through inference computing. Agentic AI is next and will change how organizations operate. The companies that succeed in this transition won't be the ones with the best AI tools, they'll be the ones that redesign their roles, upgrade their skills and adapt their workflows thoughtfully, deliberately and with humans still firmly in the loop. For now.
Nicolas Genest is the Founder and CEO of CodeBoxx.
Opinions shared in this piece are the author's own.
(Photo Credit: Andrei Hasperovich/Adobe Stock)
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.