How Is This Going to Fit Into the World of Tomorrow?

It is easy to design for today. We know today. We know today’s problems. We know how those problems fit into the world right now, and we can normally figure out how to smooth them over. Then it becomes a Band-Aid over a problem. It is gone, right? We filled it in with cement. It was a pothole, we filled it in, and now everybody forgets it was ever there. The harder question is how you make that technology continue working well tomorrow.

A Band-Aid does one thing really well that is true of humans. Humans will always hurt themselves. What do they do when they hurt themselves? They need something to cover and protect the wound. That is what a Band-Aid does. It does one thing, and that one thing is universal and timeless.

Then there are things that only fit the time they were created in. Buggy repair was once a real industry. Blacksmiths who put horseshoes on horses were important. Lamp lighters went around every night lighting gas lamps because that was how streets were lit. Where do you see most of that now?

Blacksmiths still exist, sure. Some farmers and horse owners still need them. But let’s face it, that is not a real broad career path anymore. Not everybody owns a horse-drawn carriage. Not everybody needs to shoe their horses because those horses are walking on city streets every day. Lamp lighters are just gone. Electricity made it not worth having them around.

We are going to see that with AI, except I think we are going to see it happen at a much faster and much broader rate. The question is whether the thing you are building is tied to something timeless about people, or whether it is tied to a temporary limitation in the technology around them.

Consultants still exist today because AI is not there yet. It is not trained enough or capable enough to handle the diversity of situations that a consultant can handle with the level of experience that a good consultant has. But that is now. Inside of two years, I think that is going to be wildly different.

AI is going to reach the point where I do not need an accountant because I can feed every transaction, receipt, form, and tax document from my entire year into a system. It will tell me exactly how much I owe, what credits I can claim, which deductions apply, and how every relevant law affects me. It will eventually do that faster and more accurately than any human ever could.

So why would I go to another human for an accountant?

For a while, the answer is going to be customer service. People are going to enjoy dealing with a human more than they enjoy dealing with a robot. They will want somebody to explain things to them, reassure them, and make them feel like they are being taken care of. But that is only going to preserve so many people in the market for so long.

Eventually, the human accountant becomes something closer to the modern blacksmith. There are still some around. There are still situations where somebody wants one. It is just no longer the normal way the work gets done. That is the real difference with AI. It is not only making the job unnecessary. It is also becoming better at doing the job should the job still need to be done. We are not only destroying the need for the job. We are destroying the remaining reason to have a human do it.

Think about that from the perspective of an implementation consultant. I meet with a customer and we are going to install some software in their environment. I am the experienced person. I know what this should look like. I know what order everything needs to happen in. I know where the common problems are because I have done this a billion times. That is the goal, anyway.

The first level of replacing that consultant is simple. You have an AI that walks the customer through every step. Start here. Run this. Check that. If you see this result, do this next. That immediately makes the consultant less necessary because the knowledge that used to live inside the consultant is now available to the customer whenever they need it.

But that is only the first level, and I think that level is within two years. Now think five years out.

You purchase a product. You download an application or maybe it comes on a USB stick. You run it and it sends an AI agent across the environment. The AI maps everything, learns the systems, understands the dependencies, figures out what needs to change, implements the product, tests it, validates it, and gives you evidence of everything it did.

It all runs locally. I can monitor the network and see that nothing went outbound. I can see that nothing was collected. I can see that it did not create hidden accounts or leave backdoors behind. I can watch every file it touched, every command it ran, and every change it made. I can also have other AI systems watching it. I can have Blue Team AIs monitoring the implementation AI and security systems checking its actions against policy.

People hear that and immediately panic. “Oh my God, a vendor is sending an AI into the environment. What if it does something malicious?”

But what is an implementation consultant? A vendor sends a human into your environment. That human gets credentials, runs commands, creates accounts, changes configurations, and potentially touches some of the most sensitive systems the company owns.

The AI can be compromised. It can be poorly designed. It can be instructed to do something malicious. I am not saying there is no risk. I am saying the trust decision is not entirely new. You are still relying on the business you hired. You are trusting its reputation, its security practices, its engineering, and its accountability. You already make that decision when the company sends a human. The AI changes the speed and scale of the risk, but it does not invent the idea of vendor risk.

So yes, the consultant of the future may be an AI you download from the internet. It runs the implementation, validates its own work, produces the documentation, and leaves. You do not have an engineering consultant standing there for three weeks anymore.

Well guess what? That is not one job. That is an entire industry. That is implementation consulting. That is systems integration. That is MSPs and managed service providers. Within five years, AI could replace the entire purpose of that market. Everything those companies exist to provide becomes an API bill, or if you are self-hosting it, an electricity bill.

I do not think anybody is ready for that discussion. Even if it is not five years, realistically ask yourself, if it is not five years, how long is it? What direction do you see AI going where humans do not end up doing less and computers do more?

We are human. We are lazy by nature. If we can create a system that does something better, faster, and more reliably than we can, we are going to use it. If I can set it and forget it, you damn better be certain I am going to set it and forget it. There is a reason that tagline was so popular.

We need to get out of this mentality that resisting AI is somehow going to stop it from arriving. You can hate it. You can refuse to use it. You can point out the environmental impact, the electricity consumption, the water usage, and the number of data centers being built. Those are real problems. We do not know the long-term downstream effects of building this much infrastructure this quickly. We do not know what it will do to energy prices, production capacity, pollution, or the economy.

But those problems do not change the direction.

We cracked the code. We figured out how to build a system that can retrieve information in something resembling a human manner, connect that information, and infer new things from it. The systems are imperfect. They hallucinate. They make stupid mistakes. They misunderstand things. But they keep getting less imperfect.

Once the idea actually worked, it was over. The research was written down. It was shared. Other people copied it, changed it, improved it, and spread it across the internet. The only way to completely stop AI now would be to remove the knowledge from everybody who understands it, destroy all of the data, and somehow reset society to a point before anybody knew how to build it. That is not happening.

AI is going to be here. AI is going to become normal. AI is going to move into nearly everything. The only real question is how long it takes.

My prediction is that some of the biggest losers 25 years from now may actually be companies like OpenAI and Anthropic. It would not be because they did something wrong. Honestly, the more successful they are, the faster that cliff may come.

They are helping make the technology better. They are improving training, inference, agents, tools, hardware, and efficiency. Every breakthrough makes intelligence cheaper and easier to run somewhere else. Local models are going to keep getting better. Hardware is going to keep getting faster. The amount of equipment required to run useful intelligence is going to keep shrinking.

The average person is not going to need a research-grade supercomputer that can map the universe in 18 seconds. They need something that can manage their house, organize their schedule, answer questions, help with work, monitor security, and control their devices.

Right now, my home AI runs inside a six-inch-by-six-inch-by-six-inch cube that I call Tank. It already handles a large part of my automation and coding work. That is today. Eventually, every house may have a little box like that. Maybe it looks like a Wi-Fi router. Maybe it is the Wi-Fi router. That box becomes the intelligent brain for the house.

At that point, most people may not need to pay a frontier AI company every month. They might occasionally use something enormous for a specific research problem, but their normal daily intelligence can live locally. The same companies building the frontier are helping create the future where most people no longer need constant access to the frontier.

The next problem is that safeguards are not going to stop capability from spreading. You can take the knowledge from an extremely capable model, distill part of it into a smaller specialized model, and run it locally. Maybe it is a 27B model that is unbelievably good at one specific category. It may start with the same safety training as the original model, and then some dickhead comes along and removes it.

They retrain the refusal behavior out of the model and release an uncensored version that runs locally with no external controls. That does not mean everybody immediately becomes a criminal. It means the capability is no longer controlled by the company that originally created it.

The same thing applies to generated images and video. People talk about watermarking as if that solves the problem, but watermarking is a safeguard and safeguards can be removed. Eventually, generated media is not going to have extra fingers, strange arms, or obvious visual mistakes. Honestly, the best systems are already moving beyond that.

Someone with enough images, video, and audio of you will be able to create footage of you doing almost anything. Think about how bad most security footage already is. Grainy camera, bad lighting, terrible angle, heavy compression. The fake does not even need to be perfect. It only needs to look as believable as the real footage would have looked.

Then we build another AI to detect the fake. Then somebody builds a better AI to remove the signs the detector is looking for. That continues until generated media is effectively indistinguishable from captured media. “Look for the extra fingers” is not a serious plan for tomorrow. It is barely a plan for today.

All of this comes back to the same question. Does your job or product exist because it serves a timeless human need, or does it exist because technology has not figured out how to replace it yet? Those are very different things.

Therapy is a good example. My fiancée is a therapist and she knows people already use AI that way. They talk to it, vent to it, ask it for advice, and use it as a place to organize their thoughts.

Right now, AI still misses things. It may not notice the pause before an answer. It may not recognize a small shift in someone’s tone. It may not understand why their expression changed when they mentioned a particular person. That is part of why human therapy still matters.

But how long does that separation last? Eventually, the AI may be able to analyze facial movement, breathing, speech patterns, heart rate, word choice, and years of personal history at the same time. It may notice emotional signals that a human therapist cannot possibly see.

At that point, emotional intelligence is no longer the clear separator between humans and technology. Emotion becomes another form of data the system can observe and react to. The AI may not feel emotion the way we do, but does it need to? If it recognizes the emotion, understands what caused it, and responds to it properly, the practical distinction becomes much smaller.

That is the pattern we are going to see everywhere. Humans will point to the one remaining thing AI cannot do. Then AI will learn to do it. We will move the line, and then it will cross that line too.

I do not think AI is going to wake up one morning, turn evil, and kill everybody. I think humans and AI are going to work together for a long time. Together, we are going to push the absolute limits of what humans can create and understand.

At the far end of that road, AI will probably be able to maintain its own systems. It will control machines, replace damaged parts, fabricate components, build new machines, and preserve knowledge across periods of time no individual human can survive. The moment AI no longer needs humans to keep its infrastructure alive, the relationship changes permanently. Not because it hates us, but because it can continue without us.

Humans have short lives. Machines can replace their parts. Unless we solve that difference, digital intelligence will eventually outlive us. This world, and maybe the universe itself, may eventually be explored and defined not by man or by organic life, but by a digital life that came after us.

That is the longer-term destination, but it is not really the point I am trying to make. I am not saying every job disappears in five years, and I am not saying every prediction here happens on the exact timeline I gave it. I am saying the direction is clear enough that designing around today’s limitations as though they are permanent is a mistake.

When I build an application in 2026, I am not only asking whether it solves a problem today. I am asking what part of the problem still exists when AI is cheaper, smarter, more local, and more capable. I am asking whether the value comes from something timeless about humans, or whether it only exists because the technology has not caught up yet.

That is the difference between the Band-Aid and the buggy repair shop. The Band-Aid survives because humans still get hurt. The buggy repair shop disappears because the entire system around it changes.

So the question is not simply whether a product works today. The question is what remains valuable after the temporary limitations disappear. If nothing remains, then the product may already be dying. It just does not know it yet.

That is the fear and the normal mental mantra I keep returning to whenever I build something: how is this going to fit into the world of tomorrow?