Coding agents have robbed me of my emotional connection to software I produce
It’s 4 pm on a Wednesday, and the massive feature I started working on 2 days ago is all but done: 85% test coverage, clean, easy-to-follow architecture (doesn’t look like slop at all 😂), and all known edge cases covered. This should be cause for celebration. I’m moving way faster than I used to, still delivering at a high level, but I’m not feeling it; there’s no true sense of accomplishment. Is this normal?
I’ve been designing and building high-quality software systems for over a decade now, and each time I complete a project, especially one in which my technical boundaries were stretched, I’d have this euphoria, a dopamine hit second to none. I’d quickly rush to my ‘brag doc’, a living document where I jot down the proudest moments of my software engineering career. Gone are those days, at least for the most part. Ever since I started using coding agents (they are great 80-90% of the time by the way), I don’t feel the same way about the software I produce.
This is my workflow when I’m using coding agents:
Draft a high-level spec, a 25 thousand foot view of the feature, including diagrams in some instances.
Iterate over the spec with the agent till I’m satisfied it understands what I need to build. This normally takes hours, but some have gone for more than a day.
Ask the agent to draft bite-sized chunks that it will build. If I’m using GitHub, I then create epics and issues from those chunks.
Ask the agent to work on one chunk at a time. I review its work as it’s busy working, as well as the PR when it’s done.
After a few hours for small tasks or a few days for larger features, the work is complete.
As you can see, I don’t throw the work at the agent to go figure it out and come back with a result. I’m actively involved, watching it with such immense attention that as soon as it starts to stray from the path, I jump right in to steer it back onto the right path.
Despite all this involvement, I don’t feel as emotionally attached to the end product, for some reason. The spec is mine, the high-level plan is mine; everything except the actual lines of code. This is baffling. Why don’t I feel as connected to the features I build using agents? Is it because I didn’t personally type the code? Is it because I didn’t ‘sweat’ as much while working on the feature? Why don’t I have the same dopamine hit?
I remember back in the day when I’d spend days on end researching the best way to solve a particular problem. I have a whole directory on my computer dedicated to proof-of-concept projects and spikes to verify or debunk certain assumptions. I’d go to page 10 of the Google search results, read countless forums, GitHub issues, and StackOverflow (my wife recently asked me why I haven’t mentioned StackOverflow in a long time), all in a quest to find the best solution. And man, when I finally find the solution! The feeling was second to none.
These days, all this is now wrapped in a tool call for an AI agent. What used to take me days is now done in a couple of seconds, minutes at most. That’s progress, right? Well, I’m not too sure to be honest.
Am I moving quicker? Yes.
Am I still producing high-quality (in some cases, higher-quality) software? Yes.
Am I learning and growing as a practitioner? Absolutely note.
That may be why I’m struggling to feel connected to the work I produce with coding agents. When the work is done, feature is built, I personally didn’t gain much from the experience. I feel stagnant, stuck.
Something’s got to change. Coding agents, by the looks of things, are here to stay, and they’re going to continuously get better (again, looking at the trends). The onus is now on me to find different ways to get the dopamine hit I used to get pre-AI. I need to find a way to feel connected to my work.
I will find a way, like I always do.