Something is happening in software development that is easy to miss if you are only watching what average teams are doing. The most enthusiastic adopters of AI augmented development are not juniors trying to catch up or mid level devs looking for an edge. It is the extreme seniors. People like Linus Torvalds, who has talked publicly about using AI to help write code.
This might seem counter-intuitive, but it actually makes perfect sense if you zoom out and tilt your head a little to the left… AI is not mainly a typing accelerator. It is an execution layer. The people who already live in intent, architecture, and judgment get the most leverage from a faster execution layer.
Skill Allocation Creates an Interesting Adoption Curve
Juniors and principals tend to love AI quickly. A lot of mid level developers shrug at it.
Juniors have a coding skill gap. Code is a skill. The reason they are junior is because they haven’t skilled up that particular skill as high as it needs to be to move quickly. AI closes that gap and they feel it immediately. Their output jumps because the bottleneck was always translation from idea to code.
So why principals? They already know how to code, so why do they care about this at all? Is it just faster typing?
No. Not even a little bit. It is actually way more interesting (and the reason why I wanted to write this)… Architects and principals delegate. They think in systems. They must communicate clearly about not only what the code should look like syntactically, but what it needs to DO. What problem it must solve. How it should be broken down. We spend most of our time reviewing work for correctness and coherence.
AI fits that style perfectly.
We can describe what we need, get a draft fast, and then do what they are best at, which is evaluating and steering.
Mid Level Developers Are Getting Shafted (sorta)
Mid level developers sit in the middle. They have worked hard to become solid implementers, and that is a real accomplishment. But if your identity and value are mostly “I write good code,” AI feels like a mild improvement, not a superpower. Meanwhile the biggest AI leverage often comes from skills that mid level devs have not fully had to build yet, like writing crisp specs, decomposing messy ambiguity, and making architecture level tradeoffs. So the reaction is often, “This is not that useful,” when the real story is, “My next level is not more coding, it is more clarity and judgment.”
It is an uncomfortable transition because we’ve put so much focus on “hard skills” and we place so much of our self worth on lines of code shipped. We’re in a new paradigm. It is about to get wild.
The (New) Skills That Matter
We spent years pretending soft skills were optional. Communication was the polite thing you did after you proved you could code.
AI flipped that. If you can explain what you want with precision, AI turns into force multiplication. If you cannot, you will spend your day debugging confident nonsense. The people who invested in communication, problem framing, and context transfer are winning because AI rewards clarity in a very direct way.
The highest leverage skills in AI native development are still human skills.
Clear specification. Can you say what you want, why you want it, what constraints matter, and what success looks like?
Problem decomposition. Can you cut a big messy thing into pieces that can actually be built, tested, and iterated?
Judgment and pattern recognition. Can you look at output and know if it is correct, maintainable, and aligned with the real problem?
Systems thinking. Can you keep the whole architecture coherent while the parts move fast?
AI is excellent at generating artifacts. Humans still own intent, structure, and evaluation.
Why Extreme Seniors Feel Like Cheaters
People like Linus have been operating this way for decades. Hold the architecture. Specify what needs to happen and why. Delegate execution. Review quickly. Accept or reject based on experience. AI just replaced the delegation target.
The loop is the point. If you can run that loop cleanly, AI makes you frighteningly fast.
A Simple Skill Shape
Below is an illustrative example, not a scientific measurement. It captures why AI can feel transformational for juniors and principals, but less impressive for a lot of mid and senior level developers. The mid and senior level developer is strong at coding (sometimes even stronger than a principal), but the next level up is dominated by specification, decomposition, judgment, and systems thinking.
If you are a junior, AI helps you ship while you are still building your mental models. If you are a principal, AI helps you execute intent at speed. If you are mid level, AI is often a mirror. It shows you what your next growth area actually is.
Process Has to Move Too
Most companies are trying to bolt AI onto 2015 era Scrum ceremonies and wondering why it is not working. If your process assumes slow execution and scarce iteration, AI will feel like chaos. If your process assumes clear intent, tight feedback loops, and fast review, AI will feel like a superpower.
This is not “move fast and break things.” It is “move fast and keep your standards.” The only way that works is stronger specs, tighter review, and better judgment.
We Are Still Early
The tools change every few months. Patterns are still forming. The only reliable posture right now is systematic experimentation with high standards. Try things, measure outcomes, keep what works, and be honest about where it fails.
Looking Forward
AI is not replacing developers. It is shifting the center of gravity of the job away from typing code and toward deciding what should exist, why it should exist, and whether it is actually good.
I have been coding since I was four years old on a Commodore 64. After decades across insurance, banking, aviation, manufacturing, telecom, music, and entertainment, this is the biggest shift in how software gets built that I have ever seen.
Last week I watched a mid level developer dismiss AI as a toy because it “only saves typing,” and then I watched a principal use the same tools to ship an entire vertical slice in an afternoon by treating AI like an execution layer and themselves like the reviewer in chief. That gap is real, and it is going to get wider for a while.
If you are early career, use AI to learn faster, but do not outsource your understanding. If you are mid level, treat AI as a signal about what to level up next. If you are senior or principal, you already know the game. AI just gave you a faster board.