With the recent acceleration of AI models like Opus 4.6 and coding agents like Claude Code, I have been thinking about how my role is changing and wanted to share some thoughts.
So much code Link to heading
I still remember the first time I installed GitHub Copilot. The amazement I felt when I started typing a function name and before I could even finish it would fill in the implementation. While rarely perfect, it was good enough most of the time. It felt unfair. I actually had to pull up my first GitHub invoice to anchor when this happened (it was August 2022).
The next step function was when I moved to Cursor. Their tab auto-complete was almost reading my mind. I became absolutely hooked. I even moved away from using PyCharm as my primary Python editor and over to Cursor, solely due to that tab completion. I would start changing a function, and not only would the auto-complete finish my thought, it would then hop to the N other places that needed to be updated based on the change. It’s hard to believe, but this was February 2025. It seems like so much longer ago.
I’ll admit I was late to the agentic coding paradigm, I felt like I would always come up with better code since I had context of the repo I was working on, or I better understood the big picture, or whatever reason I’d tell myself. I was happy enough having AI auto-complete for me. But I did start outsourcing writing tests to agents. I’m honestly not even sure of the models I was using (maybe Sonnet 3.7?), but I know that I never cracked the $20/month limit on Cursor. While it was good enough for tests, I still didn’t find it good enough for anything important. I was more effective with the auto complete, or very focused changes.
And then late 2025 happened. For me it started with Gemini 3. This was the first model I remember hearing a lot of buzz about, murmurs it was actually really good. And it really was, I started writing more prompts into Cursor and having it generate better code. I was still using it in very targeted ways however, not fully trusting it. But things really started to change for me over the two weeks of holidays I took around Christmas.
As a fun little side project, I was working on a small mobile app, seeing what was possible. As a solo dev, I didn’t want to bite off more than I could chew, so my research told me the best option was to use the web-app-in-an-app concept, with the latest and greatest framework here being the Ionic Framework. I’d be able to build a cross platform app that looked good enough. And that was exactly what I started with prior to the break. I was able to get my idea working, but the UI looked like a developer built it, regardless of whatever prompts/libraries I tried. So I decided with a bit of time over the break, I would re-write it in React Native. Why not, that should at least make it look better. And, I made the decision I was going to get out of my comfort zone and delegate most of the work to Cursor, and Gemini 3. Around the same time, I was also hearing that Opus 4.5 was really good, but it was so expensive, I didn’t feel like I could/should use it as I’d blow through my $20/monthly budget. Not even a few days in, I realized $20 wasn’t going to cut it, so I went all-in and upgraded to the $200/month version (which gives $400 in credits, at least at the time). I figured with $400 in spend, I would have the freedom to use whatever model I wanted, I’d never run out… Fast-forward a few weeks, and I had re-written the app a third time, this time in Flutter. I have never written a line of Dart prior to this. And the app was is beautiful. It is working better than I could have imagined, animations were buttery smooth. But I was only about 1/2 way through my cursor billing cycle and I already blew through my $400 in spend. Opus 4.5, Gemini 3, GPT-5.2 were all amazing, I’d have them review each other, correct mistakes. They were writing better code than I ever could. To be fair, this was a greenfield project, so it really gave them a chance to shine. But that two week AI binge changed me forever.
So little code Link to heading
I’m fortunate enough to work for a company that fully embraces AI. We’re empowered to use AI tools from all the big players and we’re not (overly) token constrained. After the holidays, I have been like a hungry hippo, gobbling up anything AI I can. The past few months have been incredible. Ralph loops, gas towns, it’s mind-boggling the pace at which things are improving.
I’ve said this to multiple colleagues, but if someone would have told me in mid-December that in a month or two I’d barely be writing any code by hand and outsource everything to AI, I probably would have told them they must be on good drugs.
It’s hard for me to articulate how this feels, but I’ll try. In a way it’s amazing, really truly incredible the progress that has been made. In a focused, 30m-1h session, I can plan, revise and implement something that would have previously taken a day (or even days). I’m now getting PRs sent to review that are really good. I have been known as a bit of a stickler on code reviews, but I’m finding fewer and fewer things I need to comment on. More LGTMs than I’m used to. In a way it’s sad that a lot of the experience I’ve amassed over the past 20+ years is now almost irrelevant, as anyone can prompt their way to a solution or get a bot to review their code. I actually really like writing code. I love the problems that you can’t get out of your head till you solve them – they keep you up at night (in a good way).
I can’t remember a time in my career when I’ve written so little code (by hand). Yet, on the flip side, I also don’t think there’s been any time in my career when I’ve written so much code. Is this a bad thing? Absolutely not, I’m fully embracing where we’re at, and am excited to continue to use this new found leverage.