AI Can Open the Door. But Who Watches the Lobby?

AI can write code, answer support tickets, summarize documents, and automate pieces of work that used to take hours.

That does not mean it understands the full business role it is stepping into.

The doorman fallacy is a useful way to explain where companies get AI wrong. They see the visible task, automate it, and accidentally remove the hidden judgment that made the whole thing work.

Short answer: The doorman fallacy is the mistake of defining a job by its most visible task while ignoring the hidden value around it. In AI adoption, this happens when businesses replace people because AI can imitate one task, without accounting for context, judgment, risk, and customer trust.

The problem is not AI.

The problem is treating “the door is open” as proof that the doorman was unnecessary.

Jump Ahead

• What is the doorman fallacy?
• The hotel with the doorstop
• Why this matters for AI
• Programmers do more than write code
• The form was the door
• Where businesses get AI wrong
• How to use AI without leaving the door wide open

What is the doorman fallacy?

The doorman fallacy is usually credited to Rory Sutherland, who wrote about it in Alchemy.

The idea is simple.

A business looks at a role and only sees the obvious task.

• A doorman opens doors.
• A cashier rings up orders.
• A support rep answers questions.
• A programmer writes code.

If that is all the business sees, automation looks like an easy win. Replace the visible task, reduce the salary, and call it efficiency.

But the visible task is often not the real value.

The doorman is not valuable because human hands are better at pushing a door than a hinge, sensor, or automatic opener. The doorman is valuable because he is part of the hotel’s operating system.

He greets guests. He notices who belongs there. He helps with luggage. He hails taxis. He recognizes regulars. He discourages problems before they reach the front desk.

He makes the building feel like a five-star hotel before the guest ever checks in.

Opening the door is the visible task.

Judgment is the real value.

The hotel with the doorstop

Imagine a hotel trying to save money.

A consultant walks into the lobby, watches the doorman for a few hours, and writes the easiest recommendation in the world:

Remove the doorman. Install an automatic door.

Or even cheaper:

Use a doorstop.

Now the door is open all day.

Problem solved.

Except the business problem was never “the door is too closed.”

The real business problem was managing the transition between the outside world and the hotel experience.

That transition includes security, welcome, status, information, assistance, and trust.

A doorstop can keep a door open.

It cannot read the room.

It cannot spot the guest who looks confused.

It cannot tell the difference between a regular customer, a lost tourist, and someone about to cause a problem.

It cannot make the hotel feel expensive.

It cannot protect the brand.

This is how businesses accidentally destroy value in the name of efficiency. They automate the part they can see and remove the part they never measured.

Why this matters for AI

AI is making the doorman fallacy easier to commit.

Not because AI is bad.

I use AI constantly.

The problem is that AI is very good at visible output.

It can produce code. It can produce copy. It can produce summaries. It can produce designs. It can produce support responses.

That makes it tempting to assume the person doing that work was mostly an output machine.

But in a healthy business, good people are rarely just output machines.

They are filters. Translators. Pattern recognizers. The people who know when the request is technically possible but strategically dumb.

That is the part that does not show up cleanly on a spreadsheet.

And that is the part AI can miss when nobody experienced is guiding it.

Programmers do more than write code

A lot of companies are about to learn this with programmers.

The surface-level view is simple:

Programmers write code. AI writes code. Therefore, AI replaces programmers.

That is the doorman fallacy wearing a hoodie.

A good programmer is not just typing syntax into a screen.

A good programmer is asking:

• Should this feature exist?
• What happens when a customer uses it wrong?
• What breaks if this gets popular?
• What data should we avoid collecting?
• What happens if the API fails?
• What does sales need?
• What does support need?
• What does marketing need to track?
• What does the owner think they asked for versus what they actually need?

That is not just code.

That is business judgment translated into a working system.

AI can help write the code faster.

Great.

But someone still has to understand the consequences.

The form was the door

I have had clients ask for “a simple form.”

Five fields. Submit button. Done.

On the surface, that is a tiny programming task.

But after a few questions, the real issue is almost never the form.

It is lead routing. Spam filtering. CRM handoff. After-hours notifications. Duplicate submissions. Attribution. Follow-up ownership.

It is also what happens if the email notification fails, or if two salespeople call the same prospect and make the company look disorganized.

The form is the door.

The programmer is the doorman.

AI can generate the form quickly. Maybe in 30 seconds.

But if nobody asks the business questions around that form, the door is just propped open.

Leads may come through.

So will spam, confusion, bad handoffs, missing data, broken reporting, and support headaches.

That is the difference between output and implementation.

Where businesses get AI wrong

Businesses usually do not get into trouble because they use AI.

They get into trouble because they use AI to skip thinking.

They replace a person, role, or process by asking only one question:

Can AI do this task?

That question is too small.

A better set of questions looks like this:

• What invisible work is happening around this task?
• What judgment is being applied?
• What exceptions does this person handle?
• What risks do they quietly prevent?
• What customer experience do they protect?
• What breaks if this is done faster but with less context?
• Who is responsible when the automation gets it wrong?

This applies to programmers, support teams, operations people, marketers, project managers, sales assistants, and anyone else whose work looks simple from far enough away.

The further leadership gets from the actual work, the easier it is to underestimate it.

That is why AI automation decisions should not be made from the spreadsheet alone.

How to use AI without leaving the door wide open

The answer is not “avoid AI.”

That is the wrong lesson.

The answer is to pair AI with the people who understand the system.

Use AI to speed up the visible task.

Use humans to protect the invisible value.

For software and web work, AI can help draft code, build prototypes, summarize requirements, generate tests, document workflows, and speed up repetitive pieces of development.

But the business still needs someone who understands architecture, security, SEO, analytics, customer experience, maintainability, and what happens after launch.

A website, form, automation, or AI workflow is not done because it works once in a demo.

It is done when it works in the real business.

With real customers. Weird edge cases. Tracking. Ownership. A plan for what happens when something breaks.

AI can open the door.

Sometimes beautifully.

But someone still has to watch the lobby.

The real AI advantage

The companies that win with AI will not be the ones that fire the doorman.

They will be the ones that give the doorman better tools.

That is the practical opportunity.

Let AI handle more of the repetitive work.

Let experienced people handle more of the judgment.

Let programmers move faster, but do not confuse faster code with better business decisions.

Let support teams use AI, but do not remove the human path when trust matters.

Let marketing teams use AI, but do not let it publish disconnected slop that weakens the brand.

The businesses that understand this will get leverage.

The businesses that do not will end up with a doorstop where the doorman used to be.

And they will wonder why the lobby feels different.

What to do next

Before replacing a role, workflow, or vendor with AI, map the invisible work around the visible task.

Ask what the person or process is protecting, not just what they are producing.

That is where the real value usually lives.

If your business is adding AI to your website, marketing, content, support, or internal workflows, the question is not just “Can AI do this?”

The better question is:

Who is making sure this does not leave the door wide open?

Sources and inspiration

• Rory Sutherland, Alchemy, for the original doorman fallacy concept.
• The Conversation: The doorman fallacy and careless AI adoption
• Sahil Bloom: The Doorman Fallacy