I’ve been working with agentic coding a lot lately (I know, so unique), but it’s made me think, in a world where our primary interface with the code is an AI intermediary, how does that change the roll of programming languages and tools? What do useful language features and libraries look like when typing and reading code is only a secondary activity?
Language features and libraries basically trade an increase in vocabulary for the ability to express concepts clearly and concisely. For a programmer, they reward familiarity with efficiency and expressiveness. When a codebase is written by hand, you eat the cost of gaining that familiarity up front, basically once, then get dividends in efficiency gains for every place you exercise that language feature or library component again. Coding with AI agents though, there’s much less familiarity with the codebase, and you often don’t end up eating that familiarity cost until you have to debug something the AI did wrong. Then, that heavy lift of familiarity shows up later in the development cycle, and with less if any amortized spread, as the relevance of that familiarity might be scoped to only one or two bugs instead of entire codebase.
I wonder how much this is contributing to the problems we’re seeing with maturing AI assisted/generated code bases having higher maintenance burdens. When adopting a new library meant taking a day or two, or sometimes a week or two to learn and integrate it, there was a headwind that pushed back against adding that complexity. Now, when you can adopt 5 new libraries by clicking “yes” in an agent console, we’ve buried the familiarity cost in the long tail of sustained engineering.
It will be interesting to see where things go now that AI adoption is basically the industry standard for software development. It seems like there’s a difference in kind, not just in degree, in how AI affects how we build software vs previous innovations in coding. I don’t think we’ve even scratched the surface on the implications to the profession and the craft.