Generative AI and the Changing Role of Software Engineers

Generative AI is still fresh. It’s new enough that there hasn’t been enough time to formalize things and who knows if the pace of innovation will even allow for it. There are no universal best practices, no established playbooks, and honestly, no consensus on where any of this is headed. That uncertainty can be uncomfortable, but it’s also what makes this moment interesting.

Software development is one of the industries feeling the pressure, and it makes sense. AI is built on software, trained by software engineers, and deployed through software infrastructure. It’s only natural the industry closest to the tools would be the first to be reshaped by them.

Anthropic, one of the leading generative AI companies and creator of Claude, recently published research on labor market impacts that shows observed AI use is still a small percentage of its theoretical potential in math, science, law, and even the arts. Of course an AI company is going to be optimistic, but it does shed a bit of light on where they and others are going to be placing their bets.

After talking with a few folks who use AI daily, the shift in how time gets spent is striking. Where coding once occupied 80% of a developer’s day, that same time is now going toward prompting, planning, optimizing agents, and reviewing AI-generated output. The act of writing code hasn’t disappeared, but it’s no longer the center of gravity.

My brother, a Principal Software Engineer at an early-stage gaming company, described something I hadn’t considered: context switching between AI agents is a problem he didn’t see coming. AI moves faster than most of us can keep up with. When you have five different agents running in parallel, your job becomes: jumping to whichever one finishes first, reviewing its work, making corrections, pointing it in the right direction, and repeating that process all day. That’s a fundamentally different kind of work than sitting down and writing code from scratch.

Most of us aren’t good at predicting the future, and I won’t pretend otherwise. But I do think the next two years will surprise more people than not. I know some software developers don’t want to hear it, especially those who just graduated with a fresh degree in hand, but the nature of the work is going to change. I don’t think the bulk of our time as engineers will be spent in the code of the applications we’re building. It’s going to shift toward higher-level thinking: outcomes, architecture, methodology, and judgment. It’s what the senior and staff level engineers spent much of their time on and even there, AI has the potential to take over much of that work.

Initially, the challenge will be figuring out how to work with AI and how to fit it into existing workflows. Then we’ll question whether those workflows even make sense anymore. As we get better at using these tools, we’ll start asking ourselves uncomfortable questions:

• Why should I concern myself with how to write and optimize a Go routine if AI can do it and I can focus on the output?
• If AI is doing all the coding, is my experience most valuable as a reviewer and does that mean I’ll stop writing code?
• How much does code quality really matter if AI can fix bugs the moment they surface in production?
• If AI is going to keep getting better at writing code (at what seems a monthly basis), why invest heavily in writing code manually now?

I don’t think AI is going to replace most of the software jobs in the next five years, but it is a powerful tool, and it will have a meaningful impact. It will influence our daily work, what we consume across news and social media, and especially how we learn. More importantly, how our children learn.