I'll never forget logging in post-2025 holiday break and seeing specs committed to the repo at work. Admittedly, at first, I was confused. I had been an avid Copilot user for users at this point and oft-celebrated its ability to predict my function names (often naming them better than I could). But, up until that point, AI felt like a surface-level tool that helped out in some instances but ultimately hadn't changed anything that much.
However, around that point, things changed. If I had been paying more attention, they had been bubbling up for a while, but it was as if a switch flipped. Seemingly overnight, everything was agentic. The expectation went from some AI usage to full-on implementation of agentic workflows into everything.
Department heads were prototyping ideas, product managers writing full PRDs with Claude, designers rapidly skipping wireframing steps and jumping right into full-fledged designs in hours instead of days. I sat back, watching in horror, as the soul of the career I loved so much was seemingly sucked out leaving behind only a shrivelled up corpse of what it once was.
I'd be lying if I didn't say that this year has been equal parts exciting and a huge struggle. There's a huge conflicting feeling watching an AI agent take your prompt and spit out a working prototype in mere minutes. It's equal parts exciting to see the barriers from idea-to-existence topple down while also immediately losing the sense of satisfaction that programming gave. Being given a problem, giving hours of your time to it to break the problem into equal parts and find solutions for each piece and finally getting over that slump... in my 36 years of existence, there are few things like it. So how do I find that again?
Reclaiming the soul of engineering
Code is no longer the barrier, so it begs the question, what is? The answer, at least for me, is the product. In 2026, AI still notoriously designs poorly. Constant monospaced fonts, repeated gradients, everything with a rounded border, etc. They all look the same and it lacks a certain panache.
I used this as an excuse to not use AI in the past. But, after spending many, many hours trying to align myself with the future, I realized that I'm still the secret sauce for AI's success. It can write code, it can statistically predict the next token in an AST, but it cannot make good product. So much of that requires an actual human testing, iterating, and subjectively implementing what it knows to be good.
And that's where my place is. I can elevate a product requirement ask from the sales team (or otherwise) in a working product spec. All the things that the sales teams and AI don't consider: easy access to actions, filtering capabilities, saving state in modals as you navigate, implementing good keyboard shortcuts.
The code is still quick to generate–but the well-written product spec is what truly sets it apart.