How to Get Ahead in the Next 2 Years of AI

The person AI makes look good, right up until it doesn't

We are going to see a lot more people building software they do not understand, running businesses they do not know how to operate, and generating reports and charts they cannot explain. The strange part is that these people might look more productive than someone who took the time to learn things properly, because the tools have become that good.

But I think most people misread what is happening. AI is not just making us faster. It is changing which skills are valuable, how people learn, how businesses operate and eventually how we organise our lives. After years of running tech businesses, teaching people to code and building with AI every day, I have a view on where this is going and what is worth learning. And no, I do not think everyone needs to become a software engineer. It depends on your goal.

The paradox: AI makes it easy to stop learning

AI makes it extremely easy to get things done. It also makes it easier than ever to stop learning. Anthropic published an experiment in which developers worked with an unfamiliar Python library. One group had an AI assistant, the other did not. When tested afterwards, the AI group averaged 50%, compared with 67% for the group that worked without it.

It is a small study, and it does not prove that using AI makes everyone learn less. But it matched something I had been thinking about. When I taught myself to code, I spent hours debugging stupid things. It was frustrating. Today Claude Code or Codex would solve the same problem in seconds, which is amazing. Yet I suspect a lot of my understanding came precisely from those hours of struggle. That is where things clicked.

There is a big difference between producing an answer and understanding why it is correct. AI is pulling those two apart, and it is creating two kinds of people: those who use AI to expand what they understand, and those who use it as a substitute for understanding anything. When everything works, they look equally capable. When something breaks, the difference is obvious. We wrote a whole piece on not becoming dependent on AI for exactly this reason.

Where do juniors learn now?

Stanford published updated research this year showing that employment for 20 to 25 year olds in highly AI-exposed jobs was about 19% below where it would otherwise have been, compared with less exposed jobs. Younger workers are facing a different market, and there is a logical reason.

Think about how anyone becomes a senior engineer. They start as a junior, fix bugs, ship small features and slowly build judgement until they can handle big systems. Now a senior with Claude Code or Codex can handle a big chunk of that junior-level work themselves. Great for productivity. But if AI takes over the work juniors used to learn from, how exactly does a junior become a senior?

That is one of the most interesting problems in the AI economy, and it is not limited to programming. It applies to marketing, operations, research and design too. The practical answer for you is that the learning no longer happens by accident on the job. You have to do it deliberately. I go into this further in AI took the bottom rung.

The handful of concepts under every AI tool

People make learning AI far more complicated than it needs to be. There is a new tool every week, and everyone is trying to guess which one will win. I spend a lot of time with these tools, and I started seeing them differently once I noticed that underneath, they are built from a surprisingly small number of ideas. Understand those, and you can categorise any new tool and pick it up quickly. There is a logical order:

Why this makes every new tool make sense

Once you see those building blocks, the products stop looking like magic. Some are moving towards AI teammates with their own computer, connected to your apps and remembering your context across sessions. Open-source agents such as Hermes approach the same problem but let you configure the instructions, tools, memory, skills and scheduled tasks yourself. Different products, different levels of control, same underlying questions: how does the agent know what to do, what context does it need, what can it access, how does it remember, and how do we know we can trust its output?

That is why I find this period exciting. We are moving from chatbots that give clever answers to AI that operates as part of a real system. Someone who deeply understands the concepts can move between tools without taking a new course every time. Tools, interfaces and models will change. The ideas transfer. If you want the plain-English version, start with how AI actually works and our explainer on context engineering.

Path one: go deep technically

Despite all the talk about AI replacing programmers, I am very bullish on people who genuinely understand software engineering. As the cost of building software falls, businesses will build software that used to be too expensive to build. All of it still has to work. It still needs databases, security, authentication, integrations, testing and maintenance. Demand for people who understand how these systems really work is not going away. If anything, it goes up.

The US Bureau of Labor Statistics still projects software developer employment to grow by 15.8% between 2024 and 2034. The opportunity is to be a developer who understands not just programming but AI agents, context engineering, harness engineering and how to build reliable systems around models. That is a much deeper career than using AI to write code. It is exactly what our AI-Native Software Development Programme is built around.

I will be honest about the cost: this path is a grind. Learning to build software properly is not easy. That is precisely why it stays valuable.

Path two: be technically competent in your own field

Not everyone wants to be a software engineer, and that is fine. A marketer, accountant, lawyer, teacher, doctor or business owner does not need to build an agent framework from scratch. But I do think a basic level of technical competence is becoming a necessary skill, the same way our parents had to adapt to the internet, mobile apps and Excel.

That means understanding how data moves between applications, what APIs and integrations do, how an agent makes decisions, why context matters, what mistakes AI systems commonly make and how to check their work. With the agent tools available now, people without programming backgrounds can coordinate increasingly complex work, but only if they understand what they are directing.

So one group will specialise in building and engineering these systems. The other will be competent at understanding, directing and applying them in their own field. Both paths benefit from the fundamentals of software. That is the honest reason I keep recommending people learn some coding even if they never plan to be a developer.

Same tools, very different outcomes

Access to intelligence is becoming abundant. Everyone will have it. But I am not convinced everyone benefits equally. I think the gap gets wider, because having access to intelligence is not the same as knowing what to do with it.

Two people can have exactly the same models. One generates endless content, copies other people's ideas and automates work they do not understand. The other explores new fields, tests ideas, digs into hard problems and builds things they could never have built alone. Same technology, completely different result, and the difference compounds. Someone who understands software uses AI to build more ambitious software. Someone who understands business uses it to test more ideas. Someone curious uses it to learn in months what used to take years.

The real divide over the next few years will not be between people who use AI and people who do not. It will be between people whose capabilities expand because of AI and people who become dependent on it without developing much themselves. If you want to start on the right side of that line, the free trial is a low-stakes way to learn the fundamentals by building something real.