Cognitive Dissonance

Around this time last year I started "writing" code with AI agents for the first time. I'd been using Github Copilot for a while at this point, initially with its impressive autocomplete functionality which seemed at times to read my mind then moving to writing bigger and bigger chunks of code through chat and copying and pasting the results into my applications. When it got to the point where the IDE plugin was putting code directly into my files without the copy and paste it felt like the start of a new dawn. The speed I could now write code increased exponentially and the only thing holding it back from being truely great was it failing to get the context of my wider application.

It was around February 2025 where, while working on my Jumpers for Goalposts game, I discovered Augment Code. This was a fairly new player on the AI scene and they were testing a new coding agent that had a Webstorm plugin. At first, I'd switch between Copilot and Augment depending on what I was doing but really quickly the quality of Augments tool improved and blew everything else I'd been using out of the water. This was the first tool that actually seemed to understand my entire project. When I asked it to write me an API endpoint to do x it knew that I had a Pulumi microservice architecture and I'd need a new lambda wiring up to an API gateway to serve it. When I'd ask for a vague feature in my frontend, it would implement it and add functionality I hadn't thought of but were really good ideas. Without it, JFG would never have gotten off the ground.

This was all really new back then, but 12 months later and coding agents are not only normal they're now being superceeded by agents swarms and orchestrators to do more even smarter and even faster. The rate of change is mind boggling and it's surprisingly left me grasping at the fundamentals of who I am and what I think about all this.

I've been coding professionally since 2003 and the job has pretty much stayed the same. I've worked my way up from writing C code to C++, Java, PHP, C#, Javascript and Typescript. From working on the Playstation and XBox to Nintendo DS, feature phones, smart phones to web dev. From a junior developer to a Staff Software engineer and Technical Lead. I've changed a lot of things but the job hasn't really changed. I have a problem, I think it through, I write some code, I test some code, I fix some code and I ship cool things. In 23 years the process has never not felt magical.

The "AI Revolution" is a real paradigm shift and it feels strange. It leaves with really mixed emotions, these contradictory feelings, this cognitive dissonance where it brings me great joy and great sadness. Excitement and fear. Hope and anxiety.

Here are some examples. Coding with AI has increased my productivity by a factor of ten. I don't think thats an exaggeration. If anything, it might be under selling it. I was estimating work tasks using the same method I've always done. Something I've always been incredibly reliable at doing. I'd have a task that traditional coding would take me 3 days. Giving that task to Codex, reviewing it and testing it and I'm done in an hour. Not only is it good code, it is covered by unit tests and I have introduced some additional quality of life improvements that previously I wouldn't have bothered with because they weren't worth my time. Now I may be going even quicker than that. I may create 5-10 PRs in a day. This is generally a good thing.

A downside is that I am exhausted. There is a really good article by Steve Yeggie about the "AI Vampire" that I encourage you to read. Coding with AI is really draining. It sounds contradictory because you're not having to do all that tappy tappy code writing stuff and figure out all these stupid things like why is my test not mocking this function properly or how can I get this framework and my code to play nicely together but it's true. I didn't realise what was happening at first but coding with AI is now so fast I am probably working harder than I ever have done but in a completely different way. My day isn't spent trying to solve one problem from end to end. It's planning 10 different tasks and how they should all fit together and understanding code I've not written and I'm jumping from one task to the next to the next so quickly that my brain turns to soup. In the last few years I've found myself in tech lead roles where I'm often mentoring juniors, explaining work we need to implement, reviewing other peoples code, keeping an eye on the overall architecture, steering engineers in the right direction. Working with AI agents is all the same skills but at 100mph.

Another thing is this confession: I've not written a line of code since November 2025. A confession that I feel embarrassed writing out loud. I'm an experienced programmer who is letting an AI do it all for him. It feels like cheating and negligent. There is an understandable stigma with many people and their feelings around AI and I feel quietly judged when I explain to anyone what I do. AI is churning out so much slop images, video, articles. Every product in every store is now "AI powered". And yet, programming might be one of the most suitable and valuable uses of AI. There is so much training data and the output is deterministic so verifying whether it's good or bad is far easier. As I said above, I'm doing far more quantity of code and far more broadly than I ever have before. Whether the quality is as good is open for debate but I would argue it makes no more mistakes than I did.

I feel like I'm forgetting how to code though. I still remember the concepts, best practices, all of the experience I've gained over the years, all the mistakes I've learned from. But the act of writing an application from scratch? My head kind of resists it now. I want to get AI to do it because taking that idea out of my head and into a computer was the menial labour part of the job. I thought I enjoyed it but I'm not sure that was the fun bit. I think that was just the means to the end. I talked earlier about how programming feels magical. When "vibe-coding" (a phrase I dislike greatly because it comes with its own stigma) on JFG for the last year I realised programming wasn't the part that was giving me the buzz, it was taking an idea out of my head and making it REAL. I couldn't sleep some nights because I wanted to build more of this game and it was all coming together so quickly I never got bored. Ask any programmer who has been coding for more than a few years about their personal projects and you will get story after story of projects that started, got about 70% done and then died. Not because they weren't good idea but because the idea of spending hours or days trying to fix a stupid annoying bug isn't worth it. Especially when your day job is doing that exact same thing but for actual money. With JFG I would have hit SO MANY of those issues that I just would have given up. What if in this new reality all those personal projects actually get finished. Some might turn into businesses. Some might be little things you build to entertain your kids for 15 minutes some might just save you doing the same repetitive task over and over again. How can that be a bad thing?

Then theres the dark side of AI. The environmental aspect of AI is fairly well documented and shouldn't be dismissed. I rejected jobs that had anything to do with crypto or blockchain for that reason and the seemingly exponential growth of AI and the massive datacenters required to maintain that trajectory do't sit well with me. Unlike crypto I can see the value in AI and not just a bunch of grifters selling you pictures of apes (now you can generate your own exclusive apes with AI anyway! 😂).

There is, of course, the impact on the creative industries. AI art, despite the not unfair "slop" accusations made agaisnt it, is improving worryingly quickly. The cost and convenience of creating AI images is naturally going to take work away from artists, musicians and creators. I don't like that idea. And I know I say this with more than a dollop of hypocricy having used AI images in this site and in JFG. For what its worth (and it's not worth a lot) I do want to replace the AI art in JFG with actual art if I get enough money to pay a human to do it.

All these things are before we get onto bigger picture stuff like OpenAIs deal with the US "department of War", deep fakes and whatever the fuck Elon is doing with Grok.

The biggest personal conflict is, am I accelerating my own obselecence? This I don't know but I'd like to hope not. In an industry that AI seems to be incredibly well suited for and at the pace it is improving, will AI eventually make the human in the loop redundant? Maybe, yeah. We're already seeing the number of junior dev roles flatline while senior roles continue to be in demand. Will that change?

At the moment, it's clear to see the value in an experienced developer at the wheel of agentic AI compared to an inexperienced developer for anything beyond the smallest throw away applications. It might not always be that way though and that does worry me. Am I going to be a farmer in 2 years time?

So in summary; AI good. AI bad. I love AI. I hate AI. All at the same time.