# Unknown Thesis
> As models improve, the quality and meaning of a company's information become the bottleneck for differentiated work. Information fragmentation destroys that meaning. Business software will therefore converge into one living model of the company.
For decades, software was built to be operated by humans. With the rise of AI, this premise is changing. Through agents, software can now work towards outcomes defined by humans.
A large part of the SaaS industry was built around specialized business needs: CRMs to help manage customers, HR for people and hiring, project management to manage work and track progress, finance, and many others. Each tool supported a specific set of business requirements and these specialized tools were necessary to run any meaningful company. These software systems required their data models and user interfaces to be specifically built for these tasks.
The result: fragmentation by design. Companies ended up maintaining a growing number of SaaS subscriptions to support their operations.
By the end of 2025 this constraint began to meaningfully lift. LLMs became capable of reasoning across a much wider range of business operations. The first "company brains" appeared. They prove a demand for shared context and meaning, but the information is still a copy: applications still own the data models and define what information means. Fragmentation of business information and ontology continues to persist, one layer removed.
Agentic workloads become differentiating when the agent understands the business. Agents need to know the meaning of information and how things relate. This becomes difficult when each application holds only a certain part of the picture.
To unlock the next generation of business software, we believe that business information will need to unify into one continuously evolving model. Information and its contextual meaning will no longer be tied to individual applications but will become a foundational layer of the organization itself. On top of this layer, a new generation of agentic software will emerge.
Unknown is building this layer.
## Old Architecture
Traditionally, SaaS software closely mirrored departments of organizations. Engineering departments would use task tracking tools, infrastructure monitoring and developer tools. Sales and Marketing had their pipeline and outreach tools. Customer Support had their own systems to track customer issues and ensure customer satisfaction. Management would assemble a company-wide view by merging information coming from each department.
Each system understood one part of the business very well. But none had an understanding of the business as a whole. The same customer, employee, project or decision could appear across multiple systems, each with different information and its own context.
Departments and ultimately humans made this architecture work. They would connect the dots, cross-reference, correlate and ensure that the company could move towards its goals.
This was a reasonable tradeoff. Specialized software allowed individual teams to be much more efficient, but the company as a whole remained fragmented. The more software a company adopted, the more coordination was required to make all of those systems work together.
This architecture assumed the operator would always be human.
## The Break
The obvious approach to allow agents to do meaningful work is to let them access information as they perform work. Connectors, MCP and CLIs unlock this to some extent. But access to information is not the same as understanding what that information means.
For example, a decision about a customer made five months ago may have been shaped by a support issue, discussed in Slack and affected revenue later on. An agent doing work today needs more than just access to the underlying data. It needs to understand how these things relate, what was true at the time and why a decision was made.
Some of this can be inferred when needed, but inference has limits. A name could refer to a customer or an employee, and that same customer could also be a partner. What these things mean is specific to every business. This ambiguity only increases as companies grow and produce more information.
History creates an additional problem. What was true six months ago, or why a decision was made, cannot always be reconstructed from the information that is available in the current moment. That understanding has to be captured as the business evolves.
The difference is not access to information. It is whether the meaning of the business has to be rebuilt for every task, or can accumulate as the company evolves.
This is the break from the old architecture: when humans were the operators, much of this understanding could remain implicit. When agents become operators, it needs to become part of the software itself.
## New Architecture
Business software will reorganize around company-owned information rather than individual applications. Humans and agents will work from the same understanding of the business, while models and workflows on top can change depending on the work that needs to be done. Information will not be aggregated after the fact, but will originate in this information layer. That is what separates this architecture from memory systems layered on top of the stack: not a better copy of the business, but its origin. The systems that remain, like payments or communications, connect at its edges.
Work will become less tied to predefined vertical workflows. Instead, an agent can determine what information and capabilities it needs, carry out the work across systems, and involve a person when judgment or approval is required. The interface can be created specifically around that moment rather than forcing the work through a fixed application.
Consider a customer renewal. Today, understanding why an account is at risk might require looking through the CRM, support tickets, product usage and previous commercial decisions. In the new architecture, those are not separate pieces of context that someone has to assemble manually. They are part of the company’s shared understanding of that customer. An agent can work from that context, determine what needs to happen next and present the relevant information to the person responsible. The person decides in minutes, with the full history in front of them, instead of spending time collecting relevant data and building up a potentially incomplete picture.
The application is no longer the center of the architecture. The company’s information is.
## What Changes
Not all software will change in the same way.
We expect the biggest change in software whose main purpose is to organize company information and give people a way to work with it. CRM, project management, internal operations and many other business-supporting applications fall into this category. As agents take on more of this work, companies that maintain one consistent model of their information will be able to differentiate. These companies will have less need for applications purpose-built for every business function.
Other software provides value beyond the application itself. Those continue to provide protocol level value. Payment systems move money, cloud infrastructure runs workloads, communication networks connect people, and payroll or tax systems carry regulatory and legal responsibilities. User interfaces will also continue to be needed, but in a much more just-in-time fashion. These capabilities will remain as underlying infrastructure, even if agents become the primary way companies interact with them.
The distinction is important. We do not believe software disappears. We believe the application stops being the default way every business function needs to be packaged and operated.
## Unknown
Unknown is building the information layer that will drive AI-native companies.
Our mission is to give every company a living understanding of itself: its customers, people, products, relationships, decisions, history and current state. That model should belong to the company, not be spread across the applications it happens to use.
The value of this layer does not depend on a future generation of agentic software and is decoupled from intelligence development. We are building it first to preserve a company’s understanding of itself: what exists, how things relate, what changed and why.
## Ownership and Independence
Applications, models and agents will change many times over the life of a company. Its accumulated understanding of itself should not have to change with them. The information layer will become the persistent layer of the company, and therefore the layer where ownership matters most.
Ownership needs to mean more than being able to export a set of records. The meaning around the information, its relationships, history and structure need to remain portable as well. A company should be able to change the applications and models it uses without losing the understanding of its business that has been built up over time. Portability means that the model and its meaning exist in a documented and exportable form, never only inside our product.
This is why we believe the information layer should be independent from the applications and models that operate on it. Application providers benefit as more of a company stays inside their applications. A model provider benefits when more work runs through its models. The information layer should have a different incentive: allow the company to choose and replace either without having to rebuild its understanding of itself.
If a company cannot take its accumulated understanding with it, then that information did not truly belong to the company in the first place.
## Why Now
Companies have always wanted software that could adapt to the way they actually operate. Excel showed how valuable that flexibility could be, but also its limits. As a business grows, it becomes increasingly difficult to keep its information and context structured and up to date. Traditional SaaS companies solved this by defining structure and meaning in advance, at the cost of forcing the business into many separate systems.
AI changes this tradeoff. It can help continuously interpret new information, connect it to what already exists, resolve inconsistencies and evolve the structure as the business changes. Maintaining a company-specific understanding of all of its information no longer needs to depend entirely on people manually keeping the information coherent.
In practice, meaning is captured where the work happens instead of being reconstructed afterwards. New information is connected to the entities it concerns, changes are recorded with their history, and where information conflicts, the system resolves what it can and surfaces the rest. Nothing is silently overwritten; history is never lost.
This is the core bet behind Unknown. AI does not only make it easier to work with company information. It makes it possible to continuously maintain the structure and meaning around that information as the company evolves.
## Where It Starts
We expect this architecture to emerge first in startups.
New companies do not have decades of software decisions to unwind. They can build around agents from the beginning, keep their information in a shared company-owned layer and avoid adopting many of the applications that became necessary for the previous generation of companies.
A small startup may not yet have an information problem at large scale, but an AI-native startup can have many agents at work from its first years. As agents begin working across support, email, sales, product and operations, they need shared context much earlier in the life of the company. Starting early therefore means that the company's history and meaning can build up from the beginning instead of being reconstructed after it has already been scattered across systems.
Concretely, a new company stores its business-relevant information and its meaning in the layer from day one. Its customers, decisions and work artifacts, queryable both by founders and agents.
This is similar to how cloud infrastructure first took hold. Startups were able to adopt a new architecture without having to replace years of existing infrastructure, and over time that architecture became the default way new companies were built.
We believe the same will happen here. New, AI-native companies won't replace the old stack. They will simply never adopt it.
## What Would Make Us Wrong
Our thesis depends on the belief that agents need a persistent understanding of the business. We would be wrong if agents can get enough context from existing systems to do their work well, and if history, relationships, provenance and decision rationale do not matter enough to justify a separate information layer.
We could also be wrong about ownership. Companies may prefer to keep their information and its meaning inside existing software platforms. If portability, shared context and independence do not matter enough to justify a separate layer, the architecture we describe will not emerge.
Finally, the thesis could be right while Unknown is still wrong. A large, existing software or AI company may manage to build and maintain one living model of the company well enough. If capturing the true meaning of business information does not require a dedicated company, Unknown does not need to exist.
These are the assumptions we are betting on. If they prove false, we will update our thesis.
---
Every company will need a living understanding of itself that machines and humans can work from. We are building the layer that makes this possible, starting with companies being built AI-native from day one.
Nikolai Onken & Brock Whitten
Amsterdam and Vancouver, 2026