The application layer

NorBot Systems is the application layer for agriculture. We turn AI into measurable results inside real farm and agribusiness operations, starting in Ontario.

I’m a CPA, so I’ll start with two numbers.

That is about one business in five overall, and fewer than one in twenty in agriculture. Agriculture is moving, and it is still the lowest of any sector. I don’t read that gap as agriculture being slow. I read it as a layer that has barely been built for it yet. This piece is about that layer, and why I started NorBot to build it.

AI is a five-layer cake

Jensen Huang, the CEO of NVIDIA, describes AI as a five-layer cake.2 Energy is at the bottom. Above it come chips, then the infrastructure that turns chips into “AI factories,” then the models. At the top are applications, “where economic value is created.”

The bottom four layers are being built at enormous scale. In Huang’s words: “We are a few hundred billion dollars into it. Trillions of dollars of infrastructure still need to be built.” The top layer, where AI meets a real business and changes its results, is the thinnest.

There is a reason. The examples Huang gives include drug discovery, industrial robotics, legal copilots and self-driving cars. Most of those can be built once and sold to a whole market, the way a chip is. A farm doesn’t work like that. Two dairies down the same road run differently, and an application that changes a business’s results has to be built around how that business works.

For a farm or an agribusiness, that layer isn’t a chatbot or a dashboard. It is a system that knows the operation, proposes the next move and ties the result back to the numbers. In agriculture, that layer is almost empty.

The gap is translation

The gap isn’t technology. The models can already do the work. I wrote about when that changed in April, in The Capability Inflection, a brief for Ontario CPA partners: from late 2025, AI could finish long, multi-step jobs reliably instead of only suggesting the next step.

What is missing is someone who understands both the AI and the business. Three questions decide whether an AI project pays off, and each one comes from a different world.

• Which outcome matters. A dairy, a grain farm, a greenhouse and a packing plant each make their money in a different place. Point AI at the wrong place and you get a tidy report that changes nothing.
• Where the data lives. In spreadsheets, accounting software, equipment, PDFs and people’s heads. Someone has to bring it together before anything can be learned from it.
• How to prove the return. Every figure has to tie back to the record it came from, so that anyone can check it.

Those are questions about the business, the systems and the books, and the answers have to fit together. That is what I mean by translation: turning an operation’s question into a working system, and its answer into a decision someone can defend. It takes people who have worked on both sides.

I have. I spent a decade lending to producers and agribusinesses across Grey, Bruce, Perth and Huron counties. I know the industry, how its businesses make money, and how to calculate and justify a return. I have also learned AI from the inside: what it can do now, and how to build real software with it. NorBot is those two kinds of knowledge in one company: we understand AI capabilities, software development and architecture in depth, and we understand agriculture and what drives its financial results.

Start with one workflow

We start small on purpose. A pilot is one workflow with a clear return, automated or improved, and live in weeks on your own data. It is a quick win. It is also a test: of the data, of the fit, and of whether we can prove the return.

If it works, the pilot becomes the first module of an AI-native operating system for the whole organisation. By AI-native I mean designed around AI from the start, not a chat window added to old software. A copilot helps with a task. An operating system learns the operation, works through its tools and brings back the decision. Fragmented data comes together, people are supported, and every number is traceable.

Then we run it with you. Every month we operate the system and improve it with you, and as the business is ready, more of the operation joins it. Every system is built for one business, on its own data, around how that business works.

Why not start bigger? A new system should start beside the old one: build one workflow, run both, and retire the old path when the new one wins.

The organizational singularity, on the farm

Salim Ismail, who wrote Exponential Organizations, calls the point where a company’s old operating logic breaks and has to be rewritten around intelligence the organizational singularity.3 In his telling, organisations move from human-centred operating models to AI-native intelligence systems. It is what I am building toward for agriculture.

In my words: an operation that senses what is happening, decides what to do next, acts through its tools and learns from the result, with people accountable for every decision.

The last part is the one people skip. The system proposes, shows its numbers and its reasoning, and a named person approves the decision. People approve decisions, not tasks. Ismail puts it plainly: “A human must always stand behind certain decisions.” And when there isn’t enough to answer with, the system says so.

We don’t advise on AI. We build the system around your operation, on your own data, and we stay to run it and improve it with you.

What this means for a producer or an agribusiness

If you run a farm, an agribusiness or a producer organisation, this is what working with us looks like.

• You start with one workflow, not a transformation. You choose the work where a better answer is worth the most. We make it work on your own data, in weeks.
• You see the reasoning. The system proposes and shows the numbers behind it. A person decides. When it doesn’t know, it says so.
• Your data stays private. We never name a client or show client data, and that includes this website.
• You can measure the return from the first workflow. We would rather prove one result than promise a dozen.
• You work with the founder, in person. Ontario first, with regional relationships.

None of this means handing your operation to a machine. The people who run it stay in charge, with better information.

Everything we build for you is held to one standard: it runs on your own data, and you can check its numbers.

The models are ready. What is left is the layer that makes them work for you.