How Agentic Coding Changed Developers

If you have not tried coding with AI agents, and you work in SWE, I would like to emphasize that you are not a true developer.

That being said, this isn't another article about "wow, look what I built with Claude Code," "omg Codex is so good," "why I never trust my agents anymore," or any of those thousand articles about how agentic coding changed developers. Instead, I want to talk about the true price of agentic coding. Not tech debt, but how the cost of being a developer has shifted.

The Old SWE Model

I started coding 2 years before the release of GPT-3.5. Prior to owning a Mac, I was one of those students who had a Chromebook that couldn't be used for actual development work. So instead, I booted Ubuntu onto a crappy Windows computer. It had some i5 chip and wasn't that good by any standards. But that was my go-to computer. I would start a remote desktop session from my Chromebook to my Ubuntu machine and do all my work on that. There were many times when I got frustrated with how slow it was, but still pleasantly surprised when everything worked when I needed it to.

On that computer, I learned my first JavaScript framework, React Native. I attended my first-ever hackathon using that computer. I didn't win, but I didn't believe I would be limited. My computer never limited my code.

It used to be that you would reach your skill ceiling way before you were limited by your tools. I don't remember ever paying for anything back then. If I needed a free database, MongoDB or Firebase hit me up. Hosting, Heroku. Wanted to connect my projects to the internet? ngrok. There were so many tools that I had at my disposal for free.

I never felt that I had a barrier to entry other than my own skill. Everything I needed was at my finger tips. I had to learn to use my tools to the fullest ability before I could max them out. Moreover, the old SWE model was all about mastering, building your skills project by project, and competing with others through your understanding.

Another way to put it: software developers then were more about leveraging knowledge to get ahead. It revolved around the meritocracy of understanding.

The New SWE Model

claude create subagents to first research a neural network structure for an image classification model and then implement the code step by step, an agent for each non-overlapping step

That line is what coding has shifted to. You are more of a "manager" than a developer.

These tools have completely changed how people are coding, but this also means the average developer has to spend more on these tools than ever before. If you want to get ahead, you have to pay for them. Without these, it's very hard to keep up with people who have access to agentic coding tools.

I want to give an example using Claude Code. If I want the smartest implementation, I have to use the smartest model. Smarter models cost more. Each token it generates costs more money. If I want the model to think more about its response, I need to spend tokens. Tokens cost money. Before, if someone wanted to get ahead, they would spend time learning the concepts they cared about and then building up a project slowly. But now people can just spin up billions and billions of agents at the same time and cherry-pick the one that they feel would be the best project, spending more and more money in the process.

Speed has become a purchasable tool, where before it was based on skill. If you want to be faster and smarter than the people around you, or want to make more money, etc., you need to have the best tools, which cost way more compared to the more equal playing field I had when I began coding.

During the years before ChatGPT, I knew I was the only thing stopping me from doing something crazy. But now, many times I look at people spending thousands of dollars on agentic tools and wonder how I can even compete with them if I don't pay as much as them. Compared to other industries that require expensive credentials, connections, or capital, software was unusually accessible. As a 12-year-old, I was able to teach myself coding, build projects, and compete with people who were much older than me. But now I don't think I can compete with people using a higher codex subscription than me because they just can generate better code faster than I can.

To put this shift into a sentence, now, software engineering leverages capital instead of knowledge. You pay more, tokenmax, and you will probably succeed.

The Solution

Sadly, I don't think there is an actual solution to this problem. If you ask me if you should spend money on coding tools, my answer would sadly be yes. If I said the 20 dollars I spend on Codex is not worth it, that would be BS.

But there will be a point when the software environment reaches the equilibrium it had before. It would be when the hardware allows you to run world-class models on your own device. It'll be the moment when the power to operate software is given back to each developer and taken away from huge corporations. I feel there is still a long journey for this to happen

However, I also think there might be something sooner though. It already started through the growth of AI slop. This will soon pull attention back toward human-created beauty. It will make people realize the importance of deslopping and human craft. It's like if you enter a gallery and every art piece around you is almost the same, and then you see one different piece. You are attracted to that piece just because it's different. I hope human-created projects will turn into that someday soon--put on a pedestal, and honored. I hope AI will create a pathway for human work to shine.