Where the durable value in enterprise AI is

Some of the recent data puts about a 3% gap between the frontier models and the open weight models. Whatever the exact number is this month, the direction is clear enough: there is very little left to distinguish them.

I think that changes the question every business owner should be asking. If cognition is going to be roughly the same for you and your competitor, then the model is a strange place to build an advantage. So where does the durable value sit? My answer is that it sits in two things you already own. Your workflow, and your data.

Where the model fits

I want to be fair to the model, because it does matter. It gives you extraordinary cognitive capability and it lets you deliver an enormous amount of work very quickly. That is real, and it is worth having.

But it is not a differentiator. The capability is converging, and the gap keeps narrowing. What is hard, and what nobody can buy off a shelf, is knowing what good looks like inside your own business. That is the thing I would be protecting.

Workflow is accumulated knowledge, not a list of tasks

The word workflow makes it sound like a series of steps. That undersells it badly.

A workflow is closer to the crystallised, accumulated knowledge of how your business does something. For a lot of businesses, that is the differentiator. I used to do a lot of work with a hire company, and the way they delivered service in the branch was everything that separated them from their competitors. It was not a document. It was how they did things, built up over years.

Which is why handing a workflow to AI is harder than most people expect. People understand their own process broadly, and they have an intuitive feel for it. Translating that into something AI can execute is still really difficult, because all the small grey decisions a person makes without thinking have to be understood properly first, almost deterministically, before you can hand them over.

That leads to a point worth sitting with. AI is inherently non-deterministic. Run the same prompt twice and you can get two different answers. So you have to work out where you genuinely need a decision or cognition, set the constraints around those points, and link each step to the next one. The model is probabilistic. The envelope around it has to be deterministic. Most of the real difficulty in putting AI into a business process lives in that envelope, not in the model.

Your data is the part nobody else can reach

The second half is data, and this is the one I would call the real value.

Your internal data is not accessible to anyone else. How you do things. What you know about your customers. How long a process takes. What a transaction really costs. Why one solution is worth more than another to a particular client. None of that is publicly available, and all of it is what grounds AI in your business rather than in general knowledge.

We learned this building 1KE. Early on we could produce reasonably detailed proposals, and they were fine. Once we started grounding the product deeply in how each customer operated, what their services are, how they like to work, even the specific words they use for their own capabilities, the outcomes changed standard entirely. They became high fidelity, and clearly differentiated from anything you would get out of general knowledge.

Where to start

If you are working out where to put your effort, I would do it in this order.

First, look through your business for the workflows that genuinely differentiate how you operate today. Include the things you have never thought of as workflows. Plenty of businesses have processes that are the differentiated part and have never been written down as such.

Second, find and govern the data that underpins those workflows, and think about how to improve it. Then build the controls that make probabilistic intelligence dependable enough to be part of how you run.

The reason the order matters is that AI is a multiplier. Whatever quality you have, or lack of quality, gets multiplied. Whatever clarity you have around your workflows, or lack of clarity, gets multiplied too. You end up with a differential either way, and it is worth deciding which direction that differential runs before you scale anything.

So the take is this. The model provides the cognition. The workflow is where your process and your difference live. The data holds what you uniquely know. AI will magnify both of those things for you, and it cannot generate either of them.

Watch the full conversation with Sidney on YouTube: https://youtu.be/nkv2tQuomdw

Subscribe to The AI Take for one AI lesson a week, and what it changes for your business.

About 1KE:

1KE is the presales autopilot for technical sales teams. It turns discovery calls, emails, and technical documents into structured requirements and customer-ready solutions in minutes, so teams can respond while the deal is still warm.

Learn more → https://1ke.ai

Michael → https://linkedin.com/in/michaelchanter/ · https://x.com/1keceo

Sidney → https://linkedin.com/in/sidneyminassian/

Follow 1KE → https://linkedin.com/company/1ke-ai/

‍