Abstract syntax tree representing a vertical slice of a legacy codebase stored in a graph database for retrieval as context for AI completions. The full set of symbols and and filtered for methods and named types (e.g. classes)
Originally published on LinkedIn on Apr 9, 2025. Enhanced for this site with expanded insights and additional resources.
Reports say Amazon is mandating human code reviews by senior engineers on AI generated code to reduce failures in production.
Several questions:
Aren’t you just offsetting the time spent in an activity senior engineers find engaging (writing good coding) with one they generally detest (reading bad code)?
Who actually learns to author better code in this cycle?
Don’t PR reviews mainly catch maintainability issues not bugs?
We need to directly engage engineers strengths at the moment code is written by retaining more of what works in our existing coding practices even when authoring code with agents.
AI simplifies complicated tasks. That ease of use masks complexity. People already fail to grasp the unintuitive and unpredictable nature of abstract tasks like building safe, effective software. The challenge is not just adapting change it’s communicating what has not changed.
I believe that using AI agents to build code can significantly enhance our speed without compromising delivery. By analyzing our codebase to minimize duplication, breaking work into manageable increments, and meticulously planning our implementations, we can streamline our processes. I emphasize the importance of testing early, frequent commits, and reflecting on our work to identify areas for improvement. Let's harness these strategies to elevate our coding practices!