.png)
Key Takeaways
- Automation, AI, and agentic systems are three distinct technologies, not stages of one evolution. Each has a different strength and a different failure mode for claims correspondence.
- Automation is reliable for fixed, repeatable correspondence tasks (like a status letter every 30 days), but breaks when a new exception or regulation appears.
- AI works backward from a desired outcome, making it well-suited to the messy inputs claims teams deal with daily (such as adjuster notes, varying jurisdictional language), but it still needs human review.
- Agentic systems decide their own next steps rather than following pre-set rules, and chaining multiple autonomous decisions together compounds the risk of error. Fully self-directed agentic workflows remain too unreliable for claims correspondence today, which is why agent-based tools like Kyber’s Review Agents keep a human in the loop rather than letting the system act independently.
- Kyber combines automation (guardrails, triggers, audit trails) with AI (flexible drafting and editing) to give claims teams both speed and compliance.
Automation. AI. Agentic systems. These words dominate insurance conferences and boardroom conversations, but most people lump them together as if they’re stages in a single evolution. They’re not.
As Kyber’s CEO, Arvind Sontha, often explains, these are three distinct technologies with different strengths. Automation provides tight control, AI delivers flexibility, and agentic systems aim for self-direction but remain too unreliable for compliance-critical work. For claims leaders, understanding the differences isn’t just technical trivia. It’s the foundation for making the right bets as you modernize your workflows.
Automation: Controlled and Predictable, But Brittle at the Edges
Automation in claims refers to a system that performs the same action every time a set condition is met. It is most effective when applied to repetitive, structured, and document-heavy tasks where consistency is the top priority, and that reliability is its strength. In practice, use cases for automation in claims correspondence can look like:
- Sending a letter on a fixed schedule (e.g. a status update every 30 days)
- Trigger an acknowledgment the moment a claim is logged
- Apply the same formatting and delivery rules to every letter
Automation follows fixed "if X, then Y" rules, making it reliable for repeatable tasks, but it breaks when a new exception or regulation appears that wasn't programmed in.
As Arvind, CEO of Kyber, puts it, automation is what you need “for certain pieces of the process, like status letters that have to go out every 30 days.” It ensures compliance where the rules are clear and repeatable.
But the same rigidity that makes automation dependable also makes it fragile.
“You’d try to pipe in all sorts of delay reasons and the next thing you know, a new regulation pops up, the whole system breaks because it can’t handle it.”
The moment the environment changes, whether a new jurisdictional requirement or a novel exception, the rules fall apart.
In claims, where exceptions are constant, this brittleness is a real limitation. Automation works best when the task is straightforward and variance is the enemy. It falters when nuance, interpretation, or judgment is required.
AI: Flexible and Outcome-Driven
AI is a system that generates an output by working backward from a desired outcome, unlike automations which follow a fixed set of rules. In claims correspondence, it can be used to:
- Draft a letter from unstructured adjuster notes
- Apply the correct policy and jurisdictional language automatically
- Adjust to new situations without needing new rules written
It works well when inputs vary and judgment is needed. It still requires human review before a letter goes out.
“AI says, I know the outcome I’m looking for. Let me take the data I have and synthesize it.”
That shift is powerful in claims, where the inputs are messy: adjuster notes in free text, policy language that varies by jurisdiction, regulatory rules that change state to state. Instead of brittle logic, AI can pull all of this into a coherent, compliant letter.
Take status updates. With traditional automation, you’d need to program every possible delay reason into the system. One new exception, and the rules collapse. With AI, the system can understand the intent, keep the policyholder informed with the right context, and generate a letter that meets the standard without needing a new rule every time.
The benefit for claims teams is resilience. AI doesn’t eliminate oversight, but it reduces the burden of constant template rewrites and rule maintenance. That means faster turnaround, fewer errors, and more time for adjusters to focus on complex claim decisions instead of paperwork.
Agentic Systems: Ambitious, Workforce-Like, and Not Ready for Compliance
An agentic system is AI that decides its own next steps to, instead of being told what to do. Agentic systems reason about what tasks to do in the first place, chooses the right tools, and executes
If automation is about rigid rules and AI is about flexible outcomes, agentic systems aim for something different: self-direction. Think of it less like a tool and more like a workforce.
In insurance, this would mean the system looking at an open claim would decide on its own:
- Which letter to send
- Which channel to send it through
- What compliance notes to include
“Agentic isn’t just chaining together LLM calls. It’s a core model reasoning through a task, deciding which steps are required, and choosing when and how to execute them.”
That’s a radical shift. Instead of programming workflows, you’re asking the system to design the workflow itself. It’s the dream of a digital workforce that manages itself.
But the challenge is reliability. A single model call might be 70 percent accurate. String a series of them together and failure multiplies. “If I give you a model and say it has a 70 percent success rate, and now you chain five calls together, you multiply failure. The whole thing breaks down.”
This compounding error problem makes agentic risky in claims. One missed step isn’t a small mistake. Rather, it can mean a regulatory violation or a lawsuit.
Still, the analogy to a workforce is instructive. Just as you would measure, monitor, and govern a team, agentic systems will eventually need similar oversight frameworks. They might prove powerful in areas with lower stakes or where experimentation is safe. But in claims correspondence, where accuracy and auditability are mandatory, agentic remains more aspiration than reality.
So What? Rethinking the Toolbox
The easy mistake is to see automation, AI, and agentic systems as a single staircase. Automation was yesterday, AI is today, and agentic will be tomorrow. But the reality is not sequential. These are three different technologies with different strengths.
For claims leaders, that means the question isn’t “what’s next” but “what’s right.” Automation still has a role where repeatability and control matter most. AI is already delivering results in the messy middle, where outcomes need to be clear but the paths to get there are variable. And agentic, while promising, remains too unreliable for compliance-critical work.
The mindset shift is simple but powerful: stop treating technology as a maturity ladder and start treating it as a toolbox. The winners in this next phase of insurance will be the leaders who know which tool to pick, and when.
How Kyber Fits
This is exactly how we’ve built Kyber. We don’t fit neatly into the industry’s shorthand of “AI” because we’re not just one thing. We combine automation and AI to give claims teams both speed and compliance.
On one side, Kyber uses automation to enforce the guardrails: when letters are triggered, how data is mapped, and where audit trails are captured. On the other, Kyber uses AI to flexibly draft, edit, and personalize correspondence without brittle rule sets. Together, it means adjusters get drafts instantly, leaders get compliance by default, and teams scale without adding overhead .
That’s why when people call us “an AI solution,” we usually say: we are, kind of. But we’re also automation. And above all, we’re a better tool for humans. Kyber is built so adjusters, managers, and compliance teams can focus on claims, not paperwork — with the confidence that every letter is fast, accurate, and audit-ready.
FAQs
What is the difference between automation and AI in claims correspondence?
Automation follows fixed rules and performs the same action every time a condition is met, like sending a status letter every 30 days. AI works backward from a desired outcome, synthesizing variable inputs like adjuster notes and jurisdictional language into a compliant letter. Automation is reliable but rigid; AI is flexible but still requires human review.
What is an agentic system in insurance claims?
An agentic system is AI that decides its own next steps, rather than following a fixed rule set or a defined outcome. In claims, that would mean the system deciding on its own which letter to send, which channel to use, and what compliance notes to include.
Are automation, AI, and agentic systems stages of the same technology?
No. They're three distinct technologies with different strengths, not a sequential staircase. Automation isn't a precursor to AI, and AI isn't a precursor to agentic systems. Each fits a different kind of task.
Why aren't agentic systems used for claims correspondence yet?
Agentic systems chain multiple AI decisions together, and errors compound at each step. A single model call might be 70% accurate; chaining several multiplies the chance of failure. In claims, one missed step can mean a regulatory violation, so agentic systems remain too unreliable for compliance-critical work.
Is Kyber an AI platform or an automation platform?
Both. Kyber combines automation to enforce guardrails, such as when letters trigger and how audit trails are captured, with AI to flexibly draft, edit, and personalize correspondence. That combination is what gives claims teams speed and compliance at the same time.
.png)
.png)
.png)
