Our Definition of Technology Is Too Small

AI is becoming the invisible interface between intention and action.

I started writing about Dwarkesh Patel’s account of the OpenAI and Hugging Face incident. I am still working on that piece. But I decided to take a detour. The more I thought about it and I seem to be thinking about how things are evolving much more than I probably should – I kept coming back to a more basic question.

What do we mean when we call AI a technology? (Dwarkesh)

Technically, the answer is easy. Technology is the practical application of knowledge to accomplish a purpose. AI clearly fits inside that definition. (Merriam-Webster)

I don’t think the problem is the dictionary. The problem is the frame the word and the associated definition creates in our heads.

When a CEO hears “technology,” the organizational machinery just kicks in. They think: IT. Procurement. Vendor selection. Cybersecurity. Training. Rollout. Productivity. Return on investment.

The average person hears “tool.” Something I open, use and close.

Both frames regardless of the person doing the thinking carry the same assumption: the person remains the actor. The technology helps the person do something.

That assumption is no longer valid and we need a dramatic reframing of not only the definition but also the vocabulary.

The Assumption Inside the Word

For most of the software era, people had to adapt to the machine. (In reality this goes beyond software but go with it for the moment)

We learned the menus. We opened the applications. We moved information from one system to another. We decided what mattered, resolved the exceptions and pushed the work forward.

We were the interface. We were also the glue.

In The Invisible Interface, I describe the shift as moving from clicks to conversation, and from conversation to delegation. Instead of operating the software step by step, people increasingly state what they want and expect the system to coordinate what happens next.

That may sound like a small change in how we use software. It is absolutely not.

Ask a traditional system to prepare you for a meeting and you still must tell it what to do. The drudgery: Open the CRM. Search the email. Find the last deck. Pull the outstanding issues. Build the summary.

With the new class of AI, you can state the outcome: Prepare me for the meeting.

The system can interpret (key word) what you mean, decide which sources matter, assemble the context, choose a sequence of steps, use other tools and return a result.

As these systems become more capable, they can also send the follow-up, change the calendar, update the record or initiate the next workflow.

The model is only one part of this. Memory, tools, permissions, feedback and controls determine what the system can do. The same underlying model can be a drafting assistant or part of a system that acts in the world.

By many people’s definition – That is still technology.

But it is technology in a different position.

It is no longer only on the other side of the interface, waiting for us to operate it.

It is becoming THE interface.

The Definition Needs to Expand

I think our working definition must include this:

A new class of technology can interpret human intent, choose among possible paths and act within authority delegated to it.

Ok – just to nip this in the bud right now: That does not make the system conscious. It does not make it a person.

It describes the operational change – nothing more.

The old interface was totally visible. Keyboard. Screen. Menu. App. Its job was to help us translate what we wanted into machine instructions.

The new interface increasingly sits underneath the surface. We see the request and then the result. What happens between them is the memory, the context, the sources, the reasoning, the permissions that are allowed, the defaults that are set, the tools it uses, the actions taken and the controls.

That is the invisible interface.

Not because the interface disappeared. Because it moved into the path between what we want and what happens.

A personal operating layer is not defined by whether we enter through chat, voice or text. It is defined by what happens behind that entry point: persistent context, action across systems and autonomy that remains visible and controllable.

Not every AI system is there today. A tool that drafts an email and waits for you is still mostly (but not completely) a tool.

The frame changes when context, memory, permissions and action come together.

We Have Made This Mistake Before (history as a guide)

When electricity arrived in factories, many owners initially used it inside the architecture they already understood.

Steam and water had powered groups of machines through belts and shafts. Early electrification often changed the source of power before it changed the factory.

The larger gains came as electric motors moved closer to individual machines and factories were reorganized around the flexibility electricity made possible. The technology mattered. The redesign unlocked its value. (IDEAS/RePEc)

That history is useful because it shows how a new capability can arrive before management changes the frame around it.

The automobile was initially described as a “horseless carriage.” That was not a foolish name. It was the closest category people had. It also defined the new machine by what had been removed from the old one. (Research Guides)

We do the same thing when we call AI a chatbot, a copilot or another software tool. The descriptions are not wrong.

They are just too damn small. They describe the first visible product. They do not describe the system forming around it.

If the Frame Changes, What Else Changes?

The first thing that changes is ownership.

This cannot be an IT program with some participation from the business. IT can own infrastructure, security and integration. It cannot decide how much judgment the organization should delegate, which outcomes the system should pursue or who remains accountable when it acts.

Those are management decisions. Those that tell me IT is handling this make me cringe.

Permissions also take on a very different meaning.

Giving a system access to email, customer records, payments, procurement or clinical workflows is not simply connecting software. It is granting authority.

What the system can access determines what it can change, how it answers and more.

The objective becomes policy.

Tell a system to reduce call time and it may become very good at ending calls – click. Tell it to increase conversion and it may push the easiest transaction.

The objective, the constraints, the evidence requirements and the escalation rules are part of the operating model. They are not technical details buried inside a prompt.

The work also must be redesigned. Putting AI into an existing process may save time. That is the electric motor attached to the old shaft.

The larger value comes from asking whether the work should still move through the same steps, the same handoffs, the same organizational layers and the same decision points.

So think to yourself: Who does the work? Where does judgment sit? What can be delegated? What must be verified? When does a person come back in?

These are design questions.

Accountability slowly but surely moves upstream.

The system may choose the path, but management chose the objective, the data, the permissions, the constraints and the point at which the system had to stop or ask for help.

Automation does not eliminate responsibility – especially with systems that can have flexibility in how they proceed forward.

It concentrates responsibility around the people who designed and approved the delegation.

The customer relationship changes too.

If an AI layer remembers the customer, narrows the options, recommends the default and completes the transaction, it is not simply supporting the customer relationship. It is mediating it.

Whoever controls that interface can influence what gets considered, what gets chosen and which businesses never appear.

That is not only a technology issue. It is distribution, market power and enterprise value.

And then there is the human consequence: When AI takes over cognitive work, it also takes over some of the practice through which people build judgment.

An organization can become more productive while becoming less capable of operating without the system. It can get faster and weaker at the same time.

The Frame Determines the Response

This is why I think the definition matters.

If we see AI as a tool, we ask how to deploy it.

If we see it as an invisible interface, the questions change:

·      Where is it sitting between intention and action?

·      What does it remember?

·      Which paths can it choose?

·      What can it change?

·      Which defaults does it set?

·      Who can see what it did?

·      Who can stop it?

·      Who owns the outcome?

I can pretty much guarantee you that those questions lead to a different strategy, a different operating model and a different governance structure.

AI is technology. That part is not in dispute.

But the old technology frame assumes that a person operates the machine.

This new form of technology is beginning to interpret the person, coordinate the machines and shape what happens next.

The interface may become invisible. The authority, incentives and responsibility inside it cannot.