Sticky Is Almost Dead

I recently had lunch with an old friend who is undoubtedly one of the most talented and intelligent people I know.

Whenever we get together, we exchange ideas. Sometimes they are product ideas. Sometimes they are business ideas. Occasionally, they become something more than that.

This particular lunch was no different.

He showed me a product he has been building, and I was extremely impressed. I have no doubt that he will be successful. As I drove to meet him, though, I found myself thinking about an idea of my own.

When we sat down, I told him something that I had become increasingly convinced of:

Sticky software is what?

For decades, software companies have relied upon the idea of "stickiness."

I worked for several organizations where executives openly discussed the fact that their software didn't need to be dramatically better than the competition because customers simply couldn't leave.

Once software became embedded within an organization, replacing it was enormously difficult.

Users had to be retrained.

Data had to be migrated.

Processes had to be redesigned.

Integrations had to be rebuilt.

Entire departments had to change the way they worked.

A competitor could arrive with a product that was faster, easier to use, less expensive, and technically superior, and it still might not matter because the effort required to switch was simply too great.

That was the moat.

That was stickiness.

I believe artificial intelligence has destroyed it.

During lunch, I proposed an idea to my friend.

Imagine a tool that could gather information from user manuals, websites, marketing materials, application interfaces, support documentation, and, where available, the source code itself. AI could then transform all of that information into a complete set of user stories describing exactly how the product worked.

Human experts could refine those stories to better fit their organization's needs.

Another layer of AI could then convert those stories into prompts, specifications, and ultimately an entirely new implementation of the application.

Not copied code.

Not reverse engineering.

Not intellectual property theft.

A completely new implementation based solely upon an understanding of what the product was intended to do.

My friend and I spent a great deal of time discussing the idea, and as always, it was a fascinating conversation.

The following day, I decided to test my theory.

I selected a product that I thought was exceptionally well designed. Admittedly, it was not especially complex from an engineering perspective, but it was obvious that the creators had devoted an enormous amount of time to understanding their customers.

Countless hours had probably been invested in market research, usability studies, feature prioritization, testing, iteration, and refinement.

Importantly, there was no proprietary hardware involved. It was simply an application.

I asked AI to examine the website and user guide and convert them into user stories. I then transformed those stories into a format suitable for AI-assisted development.

Less than two hours later, I had reproduced most of the product's capabilities.

I sat back and stared at the screen.

Months or years of work had effectively been compressed into a few hours.

The code was entirely different.

There had been no reverse engineering.

I had even added a few improvements.

At that moment, I became convinced that the economics of software had changed forever.

If your product consists solely of software, you should be paying very close attention.

Large organizations may simply decide to create their own versions.

Competitors may introduce faster and less expensive alternatives almost overnight.

Recurring revenue streams may become much harder to protect.

The cost of competition has fallen dramatically.

Of course, none of this means established companies are doomed.

Brand recognition will matter more than ever.

Customer trust will matter more than ever.

Technical support will matter more than ever.

Execution will matter more than ever.

But I increasingly believe there are only two truly defensible moats left.

The first is hardware.

The second is the model itself.

The first moat: hardware

For now, hardware remains extremely difficult to reproduce.

Supply chains are difficult to build.

Manufacturing processes are difficult to optimize.

Regulatory approvals are difficult to obtain.

Specialized components are difficult to source.

Physical systems simply move more slowly than software.

If your software depends upon highly specialized hardware that you control, you probably possess a substantial advantage.

At least for the moment.

The best moat: the model

The more interesting possibility is that the true moat of the future is not the application at all.

Perhaps the moat is the intelligence itself.

Today, many companies are building applications that rely upon publicly available large language models. In many cases, these products are little more than an interface sitting on top of a model that everyone else can access as well.

That is an uncomfortable reality.

If everyone has access to the same intelligence, anyone can build a competing product.

The user interface can be copied.

The workflows can be duplicated.

The features can be reproduced.

The code can be rewritten.

But what happens if the intelligence itself is unique?

Suppose your organization possesses proprietary data that nobody else has.

Suppose you have spent years learning how to train, tune, and evaluate models for a very specific purpose.

Suppose your model understands the subtle progression of Alzheimer's disease, the cognitive effects of traumatic brain injury, the unique characteristics of a rare industrial process, or the complexities of a specialized scientific field.

Then the application itself becomes far less important.

The model becomes the product.

That is when software becomes sticky again.

Not because customers are trapped.

Not because data cannot be moved.

Not because migration is expensive.

But because the intelligence itself cannot easily be reproduced.

That is a fundamentally different kind of moat.

As I thought about this further, I realized something even more interesting.

Perhaps software was never the product at all.

Perhaps software was merely the container.

Perhaps the true product has always been knowledge.

The companies that thrive in the coming decade will not necessarily be the companies with the most beautiful interfaces or the largest engineering teams.

They will be the organizations that possess the deepest understanding of their domains and the ability to convert that knowledge into intelligence.

At Voice of Reason, this idea is central to everything we do.

Our mission has never been simply to build an application. Our mission has been to understand memory, cognition, communication, independence, and the subtle ways in which artificial intelligence can help people living with dementia, traumatic brain injuries, and learning disabilities.

The software can be rewritten.

The screens can be redesigned.

The workflows can be reproduced.

But deep understanding is much more difficult to copy.

There are, however, several concerns that remain.

First, it is unfortunate that cloning has become so easy. There is tremendous value in understanding customers, identifying their needs, testing ideas, and creating products that genuinely improve their lives. Those investments deserve protection.

Second, we must ask whether innovation itself will suffer. If years of work can be replicated in hours, what incentive remains to take enormous risks?

Or perhaps innovation will accelerate because experimentation has become cheaper than ever before.

I genuinely do not know the answer.

Finally, entrepreneurs should recognize that the world has changed.

For years, companies believed that once customers adopted a product, they would remain there indefinitely.

Artificial intelligence has changed that equation.

I increasingly believe that only two moats remain.

Hardware.

And the model.

Everything else is temporary.

Sticky software is dead.

Sticky intelligence may just be beginning.