Is AI just the next layer of abstraction?

2026-10-16

I have seen several comment threads where code written by AI agents is described as "just the next layer of abstraction". That does not quite capture what is happening these years.

Yes, we have been through steadily increasing degrees of abstraction, with all the languages and tools we have had access to as programmers until now. So far the statement holds.

I wrote my first programs on a Commodore 64, where (often German) computer magazines printed endless rows of data that were incomprehensible to most people. The level of abstraction was low - real machine code typed in by hand - and if you got one number wrong, the program/game simply didn't work. No help at all in finding the error.

I went through BASIC, COMAL80, Turbo Pascal, several object-oriented languages such as C++. Then we got "managed code" in the form of the Java and .NET platforms. That gave us garbage collection, exceptions, better debugging and much more. Design patterns also arrived, as a shared "language" among developers on top of the object-oriented layer.

All of it layer upon layer, in a stack of ever-increasing abstraction.

Agentic code - the final abstraction?

With AI agents writing the code for us, we move completely away from writing code in a programming language. It all happens in natural language.

You can even get code written by copy/pasting images or drawings of what you want made, directly into your agent.

That is the ultimate abstraction over all previous programming languages.

And this is a change of far greater significance than the earlier shifts.

It is somewhat like the carpenter or bricklayer on the construction site being replaced by a robot that does the craft. And then all carpenters and bricklayers would have to be retrained as building designers or architects - if they still want to be part of building something.

Farewell to the "programmer"

In other words, we are saying goodbye to the role of programmer and to the part of the developer role that consisted of the actual programming.

This will challenge much of what we as developers have learned over the last 30-40 years. For example: If I no longer write the code myself - does it matter what it looks like? Does it matter which language I choose for the code to be written in? Do principles such as DRY, SOLID, KISS, DDD, CQRS, design patterns etc. still matter?

Just as it does matter how the house is built, the principles still matter a great deal for how good the result is and how much maintenance it requires.

The code IS still there, even if we don't look at it.

So we still need to know what a good result looks like and what users actually need. We still need to be concerned about the quality of both the design and code.

We need to know what to ask for.

I suspect it will also become more expensive to use the best AI agents. And with a tailored design and a deliberately chosen architecture around a large solution, it becomes easier and cheaper to maintain.

From developing systems to producing solutions

Perhaps we will get even better opportunities to deliver more solutions that give better results, faster. Because we can focus more on the desired outcomes, rather than on keeping a compiler happy or writing code that the next developer also finds neat.

Personally, I have thrown myself at tasks I would not have been able to take on before - they would have been month-long projects before AI agents came along. And it has been with a focus on defining the solution rather than writing the code.

Decisive skills and mindset in the future

Technical knowledge and the overview to decide how the good solution is built, in the specific situation, becomes even more decisive for success going forward.

Beyond that, in my opinion there are a number of non-technical skills that will also become essential going forward:

  • The ability to read and understand code that an agent has written. Precisely in order to keep quality on track. And to know where a change needs to be made.
  • Communication and relationships. Focus on building the right thing becomes decisive. That requires good communication skills and the ability to create and maintain relationships. Not something you learn on a traditional IT education.
  • Integrations and connections. When individual solutions become easy, the focus shifts to the connections between systems and data. That requires overview, domain knowledge and decisions.
  • Keeping focus and concentrating. Not exactly abilities that in these years are highlighted as things we are getting better at; quite the opposite. (See e.g. the article about film students who can no longer hold their attention long enough to watch a full-length film).
    It can be surprisingly demanding to sit for a whole day keeping a "conversation" going with your AI agent. It is a new way of working, which we probably have not quite gotten used to, or seen the full consequences of, yet.
  • The joy of building - as opposed to the joy of writing the code itself. These have previously been one and the same thing - but agent-written code changes that radically.

So systems written by AI agents are not just "the next layer of abstraction". It is a new way of working, where the principles we have learned before are shaken, and where new personal and professional qualities become valuable going forward.

If you became a developer because you simply loved writing code, you are going to be challenged.

It becomes essential to find the joy in delivering more solutions, faster. Hopefully solutions that make other people's workday easier or better, or products that users love to use (games, webshops etc.). That alone may well become the biggest motivating factor going forward, even though we no longer write the code ourselves.