The reason behind all this I run Simplifai.
Our solution is AI Agent that handles Claims end to end. It's not a chatbot to assist you find next best answers, it's not a point solution to optimize one little task. It's powerful technology to "think" and "do" the work. It's designed to executive on every single task (we counted 30 different types) in order to handle claim from intake to closure. It's powered by Simplifai Platform - complex enterprise-grade technology, which is a mix of workflow management, integration, insurance industry knowledge, AI models, analytics, AI Skills and UX/UI to do a configuration of AI Agents.
Insurance Carriers, MGAs, TPAs and Brokers buy our solution, pretty much anyone who handles claims. Very few use AI Agents full capacity from Day 1. Reaching end to end state is a journey of 6,12,24 months - it all depends on maturity, priorities and appetite of the company, and little bit of politics.
There are 3 applications of AI Agents: automate work (something every insurer get used to do and talk about), improve the way people do the work (something that start to emerge in 2022 with release of LLMs) and extend human capacity (do the work end to end).
While the first application is the most common and well known, other two still confuses many, but slowly start to manifest into requests from our prospect and customers.
Simplifai is a proof that AI Agent can do ALL the work that human can, but we have not reached the state where we can handle any claim of any complexity without extensive configuration, which became one of our visionary goals.
As we been working really hard together with our customers there is something interesting started to emerge.
We realized that we are building a new type of software, driven by insurance need. It's not a core system, not a point solution, not a copilot bolted on top of claim system.
Allowing AI Agents handle part or all of the work pushes us to build a single operational AI Layer where AI agents do the work, and people direct and govern it - The AI Operational System or as I call it internally Agentic Insurance System.
We did not invent this idea. You did. Every carrier, TPA, and MGA that asked us for one more capability is helping to define the blueprint. No one just defined the final state yet and accepted the reality of where we all going.
It might take few years or 10, I don't really know, but I have much stronger sense - the age of traditional software is going to end and replace with Agentic Software - with humans in different role.
Let me show you the picture.
From tools people use to agents that work For thirty years, insurance software worked one way. The system held the data. People navigated screens. Vendors charged per seat, because they were selling human attention. The software was passive. It waited for a claims handler or an underwriter to do the work.
That logic is breaking. Andreessen Horowitz, in its 2026 thesis, argues the system of record will finally start to lose primacy as models read, write, and reason across operational data directly. Foundation Capital puts it harder: systems of agents will collapse the old enterprise stack, dissolving the boundaries between the database, the workflow, and the analytics into one layer. Microsoft has already repositioned its business platform "from systems of record to systems of action."
Strip away the language and it is one idea. Software stops being a tool people operate. It becomes an agent that acts. The person stops doing every step and starts supervising the work.
That is the AI OS for insurance. Agents do intake, triage, checks, drafting, and routing. People set the goals, handle the exceptions, and own the decisions that matter. McKinsey calls this working "above the loop." The new interface is not a screen full of forms. It is a command-and-govern console. In a regulated industry, that oversight layer is not a nice extra. It is the point.
You are pulling this into being Software companies did not dream up the AI OS and push it on you. You pulled it out of us. One feature at a time.
Automate this intake step. Add a fraud check here. Summarize this claim file. Connect that third-party data source. Give underwriters a workbench that pre-fills the submission. Each request looks small. Each one is a brick. Stack enough of them and you have a wall. Stack all of them and you have a building. That building is an agent-native operating system, whether anyone called it that on the statement of work or not.
A vendor roadmap is just the sum of what customers demand. And every demand now points the same way. Less manual work. More autonomous execution. Under human control.
This is not a secret plan. It is a response. Andreessen Horowitz has described the mechanism for years: a vertical product starts with one sharp wedge, becomes a system of record, then absorbs more and more of the workflow around it. Federato's CEO Will Ross said the demand out loud: "The rising demand for true AI-native capability has surprised even us." The pull is real, it is documented, and it is coming from the buyer.
So when a vendor ships something that looks like a full operating layer, do not assume overreach. Look at your own feature backlog. You wrote the spec.
Both ends of the market are converging on the same layer Watch where the software is moving, and you see the same destination approached from two directions.
From the core side, the systems of record are adding agents. Duck Creek launched an insurance-native Agentic AI Platform in April 2026, with an orchestration layer that mixes automation, autonomous execution, and human-in-the-loop control across policy, claims, and billing. It cites BCG's estimate of up to $80B in annual US impact. Guidewire has announced 50 new AI agents across upcoming releases, an Agent Studio to build custom agents, and shipped its first core agentic capability, an Underwriting Assistant, in its Olos release.
From the AI-native side, the point solutions are becoming platforms. Federato raised $100M in November 2025 to grow from underwriting intelligence into a full lifecycle platform. Sixfold raised a $30M Series B in January 2026 to build an autonomous AI Underwriter. Its CEO frames the endgame cleanly: underwriting operations run on the system, and strategy runs on people.
The tell is in the middle. Guidewire, a core vendor, strategically backed Sixfold, an AI-native insurgent. The incumbent and the challenger are building toward the same layer. Sometimes they just merge into it.
Core systems are adding AI. AI is adding core. They meet at the AI OS.
The money already moved If you want to know where an industry is going, follow the capital.
Per Gallagher Re, working with CB Insights, global insurtech funding rose 19.5% in 2025 to $5.08B. That was the first annual increase since 2021. Two-thirds of it, $3.35B across 227 deals, went to AI-centered companies. The concentration is accelerating. In the fourth quarter of 2025, AI-centered firms took 77.9% of insurtech funding. In the first quarter of 2026, they took 95.2%, or $1.55B across 68 deals.
Gallagher Re's global head of insurtech, Andrew Johnston, put the trend line plainly: over time, AI and insurtech may become synonymous.
Investors are not betting on a feature. They are betting on a new category of software. Almost all of the money is now on one side of the table.
Why insurance is the obvious target The reason vertical AI investors circle insurance is not sentiment. It is math.
Software historically sold against the software budget. AI sells against the labor budget, which is far larger. Bessemer describes vertical AI as reimagining systems of record as systems of work, reaching a line of the P&L that software never touched. Sequoia's Julien Bek is blunter still: for every dollar spent on software, six are spent on services, and insurance is one of the largest services markets there is. It is pure intelligence work.
Insurance fits the profile better than almost any other industry. It is document-heavy, workflow-heavy, labor-intensive, and regulated. Those four traits used to be reasons technology moved slowly here. Now they are exactly why the prize is so big. There is more manual work to automate, and more structure to automate it against.
This is real, not a demo Skepticism is fair. The industry has seen a decade of pilots that never left the lab. So look at what is running in production, and read the caveats honestly.
Aviva has deployed more than 80 AI models in claims. McKinsey reports it cut liability-assessment time by 23 days, improved correct routing by 30%, reduced complaints by 65%, and saved over £60M in a year. Allianz built a seven-agent claims system in Australia, "Project Nemo," in under 100 days, and its transformation lead reports an 80% reduction in claim processing and settlement time. Sixfold reports Zurich North America saved up to two hours per submission across more than 200 underwriters. One MGA, working with FurtherAI, cut submission clearance from about 32 minutes to one.
Now the honest part, because credibility depends on it. Most of these figures are company-reported, not independently audited. Read them as direction, not proof. The Allianz 80% applies to a narrow claim type, spoiled food under AUD$500, and Allianz stresses that payout decisions are never automated. A human stays in the loop.
And adoption is early. ScienceSoft found in late 2025 that only about one in three insurers had even a single AI agent in production, and true end-to-end automation is still rare. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, mostly from cost, weak controls, and hype outrunning value.
None of that contradicts the thesis. It confirms the timing. The direction is set. The execution is uneven. That gap is where the winners and the laggards separate.
Slow, then total Regulated industries adopt technology the same way every time. Slowly, then comprehensively.
Slowly, because trust has to be earned. The path runs from human-in-the-loop, where a person checks every action, to human-on-the-loop, where a person supervises the exceptions. That curve takes time, and it should. Governance is the gate. In the US, more than half of states have adopted the NAIC model bulletin on the use of AI by insurers, which requires a documented governance program and vendor oversight. In the EU, the AI Act treats AI for pricing and risk assessment in life and health as high-risk, with human-oversight and documentation obligations landing in August 2026. This is why the AI OS has to be built around a control-and-govern surface. Regulation makes supervision a requirement, not a preference.
Then comprehensively, because the economics compound and expectations reset. Celent found 22% of insurers plan an agentic AI solution by the end of 2026. Gartner projects that by 2028, at least 15% of everyday work decisions will be made autonomously by agentic AI, up from zero in 2024, and that a third of enterprise software will include it. By 2029, Gartner expects agentic AI to resolve 80% of common customer-service issues on its own.
The sequence is legible. High-volume, lower-severity, rules-bound work goes first. Intake. Triage. Simple straight-through claims. Submission clearing. Complex underwriting, nuanced claims, and policy administration follow as trust builds. Disruption takes years because trust and governance take years. It becomes near-total because, once the work is done by agents, no carrier can afford to do it by hand.
McKinsey found that insurance AI leaders have produced roughly six times the total shareholder return of laggards, a wider gap than in almost any other sector. In insurance, waiting is not neutral. It is a choice with a price.
The choice in front of you This is not a call to panic. It is a call to decide, with your eyes open, before the decision is made for you.
You have two honest paths, not one. Replace what you have with an AI-native system built for agents from the ground up. Or keep the core you have and add an AI operational layer on top of it, an agent layer that does the work while your system of record holds the truth. Both lead to the same place. The wrong move is the third option, which is to keep buying disconnected point tools and copilots that will be legacy the moment the operating layer arrives. Even worth, trying to build things on your own - please don't become a statistic.
The US insurance industry employs about three million people, per the Insurance Information Institute. This shift does not erase that workforce. It changes its job. From doing the work to directing it. The carriers that plan for that transition will keep their best people and multiply them. The ones that treat AI as a bolt-on will spend the decade re-buying the same thing three times.
Where Simplifai stands We build the Agentic Insurance System. AI agents for claims first, then underwriting, policy administration, and service, under human oversight and governance. We built it toward an operating layer on purpose, and in the open, because that is what our customers kept asking us to build. Not in one grand demand. In a hundred small ones. Automate this. Connect that. Give my team control over all of it.
We did not decide the industry should move here. The industry decided, and it is telling every vendor the same thing. We are just choosing to say it out loud, and to build for it deliberately instead of pretending each feature is the last one.
The AI Operating System for insurance is not a prediction. It is a direction the whole market is already walking. The only real question is whether you build it on your terms, or inherit it on someone else's.
