Of course I wrote this post with AI.
That one line would eliminate a select few readers (not that I expect many anyways). However, having harbored that hatred myself, I understand how using AI can feel nauseating.
I have overcome the nausea. I no longer feel repulsed by using AI for meaningful work. I still feel repulsed by the peddlers of AI who, like snake oil traders before them, believe and preach that AI is the means and end of all.
The change was gradual. But this is the first month where I have mostly written prompts and generated code, rather than writing code directly. Looking back to find the other end of the spectrum, I need not go far. Less than six months ago, in May 2025, I complained to colleagues about receiving AI-generated pull requests. I hated being a human compiler. At that point, I did not use AI myself.
And then the last two months happened. What I witnessed wasn’t incremental improvement - it was a phase change, a categorical leap in capability. While mainstream companies haven’t caught up, those of us working with these tools daily see the evidence clearly: we’ve stopped writing code line by line. The timeline for disruption isn’t measured in years. It’s months.
Software engineering has shifted from a craft of writing code to a practice of creating through intelligent systems and owning their output. This is happening now - not in some distant future. And success requires abandoning attachment to traditional coding practices while doubling down on architectural thinking, testing rigor, and context management.
The new job description
Engineering work has split into two distinct functions.
First, you’re a creator. You define system intent and architecture. You make decisions. Before, most of my time went to typing syntactically correct code, even though the real thinking was always creative - designing systems, solving problems. Now that burden is gone. I can create bigger systems in less time. My creativity finally gets the time it deserves.
Second, you’re an owner. A large language model is just that - a language model. As the tax-paying, liable employee, I am responsible for proofreading and vouching for what it produces. The value I bring is working code, not unverified AI slop. The craft is no longer writing - it’s creating and owning.
Since November, I haven’t typed a single line of code. Yet I’m operating at maximum productivity. My entire career’s accumulated knowledge now works at breakneck speed. This isn’t theoretical - it’s my daily reality.
The surprising new bottleneck
Here’s something I didn’t expect: human-to-human knowledge transfer is now the constraint. When AI and I produce a massive pull request in hours, it takes another engineer longer to understand the changes than it took us to generate them. The time for human comprehension exceeds the time for AI generation. This fundamentally inverts the traditional workflow.
What I’ve learned about working effectively
A few things have become essential.
Connect everything. Link all your information sources to AI tools for direct querying. Document obsessively. Maintain organized research notes and context documents. Understanding how language models generate text determines how effectively you feed them context. And build for ownership - when investigating critical issues with AI, you make leaps of faith without time to verify every claim. Your systems must have clear, testable interfaces so you can stand behind the output.
What matters now
Some things remain critical: system design with testable interfaces, fundamental software architecture principles, domain knowledge and judgment. These haven’t changed.
But other things? They no longer matter as much as they used to. Code style and readability conventions. Line-by-line code comprehension. Traditional notions of “clean code”. I know this will be heresy to some (it would have been to me, not long ago).
The divide
This is the watershed. Engineers who focus solely on writing code may find their skills less valued. But for those willing to think architecturally and embrace creating through AI systems, we’ve been handed superpowers.
The profession is splitting into before and after. We’re standing at the divide right now. Software engineering is shifting from a writing profession to a creating and owning profession. Those who adapt to creating through AI and owning the output will thrive. Those who don’t will be left behind - and it’s happening in months, not years.