When AI can build everything, what is left for the consultant?

At 11:21 p.m. during my summer holiday, I was designing my house with Claude.

Not because I am good at it. Honestly, I do not know what I am doing.

But it produced sketches at a quality that helped me understand the challenges in what we were drawing. It was available at exactly the moment when the idea and energy were there. No supplier would work at that hour, at that pace, and at that price.

That is one of the biggest threats I see from AI to my own work: not just its quality, but its availability.

For the past seven years, a large part of my work has been helping companies become more efficient through IT. Put very simply, the work has been split in two: First, understanding the customer's needs and helping them make the right decisions. Then turning those decisions into software.

But what happens to that business model when the customer can begin on their own at 11:21 p.m.?

What and How

We have split the process into a What spec and a How spec.

Right now, my focus is almost entirely on What: What should be built? Why should it exist? What value should it create? Which decisions need to be made?

How is about how the solution is implemented. Here, AI has already shown that it can take over far more than I thought possible only a few years ago. At the same time, I have two years of experience, principles, and technical specifications that can give it direction.

Agentics is my practical laboratory for that question. A place where I can challenge my own assumptions and test AI-native software development in practice. Not another tool layered on top of Azure DevOps, Jira, requirements meetings, and traditional development processes, but a platform designed for a future in which AI delivers much of the work.

And this is where my dilemma begins.

They are not replaced. They are passed over.

Architects and building consultants will not necessarily be replaced by AI. But they will be passed over by people like me who would previously have called them.

The same thing will happen in software.

Through a guided AI process, a customer can work out what they want. Not after three workshops and a requirements document, but from the sofa while the idea still feels urgent or brilliant.

It is a very short step from:

This is actually what I want.

to:

Can you build it?

We all know how the AI will answer. And how many people will turn down that offer?

The crucial difference is that my spare time does not take money out of my bank account. I can invest evenings and weekends, learn along the way, and get surprisingly far without approving a quote first.

AI is not the only disruption. The work is moving from paid supplier hours to hours the customer is willing to spend themselves.

The final 20 percent

So I owe you an honest account of the economics.

I have only scratched the surface. I may be 80 percent of the way there, and from my own vibe projects in software, I know exactly how expensive the final 20 percent can become.

An architect would probably have caught my mistakes much earlier. I spent an entire day correcting structural errors so the house could actually be built — in the belief that I had now found all of them.

It may be like building your first bathroom yourself. Only afterwards do you discover why experience, craftsmanship, and the right decisions mattered. The second time, you may gladly pay a professional to do it properly.

But software is not a bathroom. The mistakes are not always visible, and it may take a long time to discover that you built the wrong solution. By then, it may already have changed workflows, data, and the organisation.

Even when building becomes almost free — tokens still cost money, after all — building the wrong thing may become more expensive than ever.

The expensive part is the handover

There is another cost that is easy to overlook.

What really costs money in a traditional project is not only the hours spent building. It is the handover: all the time spent transferring knowledge, context, and decisions from the customer to the supplier.

If the boundaries are not clear enough for the supplier to make decisions independently, every detail must be clarified. That is why two quotes for the same job can look wildly different. One supplier has included every clarification hour. The other is willing to make decisions and adjust along the way.

The customer sees the price difference and thinks the first supplier is expensive. After all, the customer can already get far on their own.

This is a major unsolved problem. How do we give an expert enough context and sufficiently clear boundaries to take over without starting again?

My hope with the house is that the distance I cover myself feels like handing over a relay baton at full speed. Not like something an architect later has to redo.

The consultant's new product

I still believe customers will want help making the right decisions. That is, in reality, a large part of what they have always bought: the experience that spots consequences before they become expensive.

But I do not know how much of the implementation business will remain.

Perhaps it is not the consultant who disappears, but the consultant's current product.

If producing code is no longer the primary scarcity, value moves elsewhere. From implementation to judgment. From being able to build the solution to understanding the consequences, challenging assumptions, and helping the customer choose the right solution.

Perhaps the consultant of the future will not primarily sell hours and code.

Perhaps we will sell judgment, consequence awareness, and accountability. Perhaps our most important ability will be taking the baton at speed: understanding the choices the customer and AI have already made, spotting the mistakes, and safely carrying the work through the final 20 percent.

The question is whether customers will understand the value before they build their first AI-generated bathroom.