A Definition of AI-Native Software Engineering

So, after trying out a few different terms to capture the shift in software engineering caused by LLM coding models, I believe AI-Native Software Engineering (or Coding) is the term that best captures the shift.

This term is needed as the other common term - Vibe Coding - has become a pejorative and indicative of a haphazard approach. Other terms, like AI-Assisted development, Agentic Coding, or Spec-driven development, capture part of the process but not the goal. AI-Native captures the idea that you have something that's fundamentally different, process-wise, than before.

Looking back to 2024, at first it seemed like LLMs were just going to help engineers write code faster through interactive prompting. Then, at the end of 2024 and the start of 2025, there started to be a bigger change: these models were able to run unattended for longer and longer and come back with something.

Figuring out how to optimize that something has been the challenge many developers have been working on since Q3 2025, with lots of different takes. The ideas that impacted me most were Kiro's take on Spec-Driven Development, Gas Town & City's craziness, and finally Chad Fowler's reimagining of the whole software concept with The Phoenix Architecture.

At first, a lot of the discussion in development communities seemed to revolve around the idea of code review. How do we review all the code that AIs write? Do we even have to do so? Code review, paired with CI, has for the past decade-plus been the central quality-control mechanism on which good, maintainable software has been constructed. Imagining a world without pull requests to review was both scary and enticing: what senior dev wouldn't love that mental focus and time back?

But as 2026 has progressed, the public dialogue has shifted to a grudging acceptance that all code can no longer be manually reviewed at the speed and volume at which LLMs can write it, although folks still seem split on whether that's a good thing or the downfall of society as we know it.

All that brings us to now and a succinct definition of AI-Native Software Engineering:

A Software Engineering process where LLMs write and review the near totality of code and features in a codebase that still results in efficient, well-architected, and maintainable code.

There are lots of different ways to slice this pie - I've written about my approach most comprehensively in the The 3x3 Rule of AI-Native Software Engineering - just like there have historically been many ways to write software in the past.

And just like in the past, there is no silver bullet that will "solve software" in all of its forms. In my year or so experimenting with AI-Native Coding I've found that the ship still needs a captain. Although the need for steering comes with less frequency than it did in 2025, course corrections are still needed and charting the initial destination in the form of an appropriate architecture is still essential.

Individual software developers are no longer each a captain of a single ship, but rather everyone's got their own fleet now and the job is to get all the boats safely to port without turning the wheels yourself.

While there's a lot of doom and gloom in the software industry, I think the English expression "May you live in interesting times" applies. Having been coding since I got an Apple IIc (compatible) in 1986 as a surprise present at age 7, I find this new age of software development profoundly interesting and the biggest sea-change I've seen in my lifetime.

Would I have chosen this to happen in my lifetime? Probably not. The uncertainty it's causing in a field that previously was a ridiculously solid career choice is unnerving. But given it's here and the returns are quite obvious and visible, I think it behooves developers to find a way to become Centaurs instead of the reverse.

The software industry is going to advance with or without me. AI-Native Software Engineering is the lens I'm using to figure out how to advance with it.