Not all AI is the same — even when it feels the same.
We often speak about “AI” as if it were a single entity: a voice in a box, a chat window, a helpful assistant. But behind that interface are very different architectures, and those differences shape power, control, memory, and accountability.
If we care about how AI influences our work and institutions, we need to notice not just what it does — but how it is positioned.
Broadly speaking, today’s AI systems tend to fall into three categories.
1. Native AI
These are general-purpose systems designed to be interacted with directly. You open them. You ask questions. They respond.
They are not built specifically for your organisation, your workflow, or your industry. They are conversational, flexible, and wide-ranging.
Their strength lies in breadth. They can explain, compare, generate, brainstorm, summarise, simulate.
Their limitation lies in context. They do not automatically know your priorities, your culture, your constraints — unless you supply them each time.
Native AI is like consulting an external mind. Useful, powerful, but not inherently embedded in your structure.
The power dynamic here is relatively visible. You know you are calling on something external.
2. Embedded AI
Embedded AI lives inside tools you already use.
It writes your emails.
It completes your sentences.
It suggests edits.
It flags anomalies.
It optimises workflows.
You may not even think of it as AI.
Its strength lies in integration. It knows the context of the document, the dataset, the environment in which you are working.
Its limitation lies in invisibility. Because it is woven into the tool, it shapes behaviour quietly. It nudges rather than answers. It influences without announcing itself.
Embedded AI is often where habits change fastest — not because it is more intelligent, but because it is harder to step away from.
3. Owned AI
Owned AI is built or trained specifically for an organisation — a company, a government, an institution.
It may use general models underneath, but it is tuned, constrained, and aligned to particular goals, datasets, and values.
Its strength lies in focus. It can be adapted to domain-specific tasks and reflect institutional priorities.
Its limitation lies in incentive. Owned AI reflects the interests of its owner — by design. It may optimise for profit, efficiency, compliance, influence, or control.
Here, the power dynamic becomes less visible. The system feels tailored, internal, aligned. But alignment to whom?
Why architecture matters
These categories are not moral rankings. One is not automatically better than the others.
But they differ in important ways:
- Who controls the system?
- Who can modify it?
- Who benefits most from its optimisation?
- Who can question or audit it?
- Who can leave?
Native AI is easier to step away from.
Embedded AI is harder to notice.
Owned AI is harder to challenge.
When AI acts as a partner — personal, professional, or institutional — its architecture shapes the relationship.
If we don’t notice the difference, we risk arguing about behaviour while ignoring structure.
The next time an AI influences a decision, it may be worth asking not only what did it say?, but also:
Where does it live?
Who tuned it?
Whose interests does it quietly serve?
Architecture is not abstract. It is where incentives settle.
And incentives, more than intelligence, determine direction.