I've been writing code for 17 years, and recently, for the first time, I felt with real clarity that I might not make writing code my core competitive advantage anymore.
AI has been changing too fast over the past few years. At first I treated it as a smarter search box. Then I found it could write code, analyze problems, call tools, even complete entire tasks on its own. For me, this wasn't just exciting tech news — it changed how I work every single day.
I went through a few very clear shifts in how I saw things.
First time: From GPT to GPT-4 (2023 to 2024).
My first impression was that it had common sense now. It could understand context. Things that used to require accumulated experience and searching through documentation suddenly got faster. For example, I helped my wife modify her R program entirely relying on DeepSeek and GPT — I'd never written R myself but still got it done. But it still struggled with complex code. Back then I didn't think it would replace programmers at all.
Second time: DeepSeek R1 (early 2025).
This was a bigger shift. What I saw wasn't an AI that "had common sense" — it was an AI that could think.
There was a fairly complex regular expression. It didn't just give an answer. It would try, get it wrong, backtrack, try a different approach, and keep going. And it actually solved it.
In that moment I had one feeling: wow, this can actually solve hard problems. The leap from "has common sense" to "can think" was far more important than solving one regex. I didn't feel like I was about to be replaced. What hit me was how fast it was progressing — fast enough that I couldn't imagine what it'd be like in another year or two.
Third time: Agents got hands (September 2025).
This time it wasn't about being smarter — it was about being able to act. I could hand a task to it, and it would go run commands on a server, read logs, modify code, debug, then come back and tell me the results.
It went from "I write code, AI assists me" to "I hand the task off and let it do the work." But it was still far from reliable. Anything slightly long-running required constant prompt supplementation, context additions, making it write a plan first, then steering it back when it went off track.
Even so, I already sensed something: people might really be able to write less and less code themselves.
Fourth time: Claude Opus 4.6 + skills and Agent memory (late 2025 to 2026).
After Claude released Opus 4.6, I felt like large language models had crossed a singularity. It was already very much like a senior engineer — able to understand complex intent, maintain direction across longer contexts. Add skills (essentially scripted prompts that codify fixed steps and experience) and Agents with long-term memory, and it was no longer a smart but forgetful assistant. It had experience, tools, and memory.
What really unsettled me was one time when I was driving. I was thinking about all these changes the whole way, and suddenly a thought popped into my head: I might really not be making a living by writing code anymore.
That moment wasn't excitement, and it wasn't anxiety. It was more like my heart missed a beat.
Because what I thought about wasn't one particular scenario, not "will I be replaced." It was something more vague — I suddenly realized that the craft I'd spent 17 years honing might no longer be the scarcest thing around. Not useless, just not enough.
So my approach has changed. Before, when I ran into a problem, my first instinct was to open the editor. Now my first instinct is to think the problem through clearly, then hand it to AI. I spend more time on "should this even be done" and "how to break it into steps for AI" than on "how to implement it."
Honestly, I haven't fully adapted to this shift yet. Sometimes AI writes code and I can see at a glance how to make it better, but I hold back. Not out of laziness — I've found that when I spend my time on "being clear about what I want" rather than "writing it myself," the results are actually better. Once you've experienced that, there's no going back.
After writing code for all these years, the most honest thing I can say comes down to one thing: I can't keep treating coding as my moat. Things have to change.
Top comments (0)