Your AI coding agent is only as good as the codebase you give it
I’ve spent nearly 15 years building production systems — portfolio management, research and quant systems, enterprise financial technology across multiple geographies. The last three of those years have been with AI coding agents as a core part of the workflow. In that time I’ve seen the same pattern play out repeatedly: agents perform brilliantly on new projects and fall apart on large legacy ones. People blame the model. Usually they’re wrong.
Read More
Interview with Cursor
Executive Summary AI-driven development is here to stay, but it is not replacing skilled developers anytime soon. Using Cursor is helpful for existing codebases. It works well when you provide the AI with small pieces of information. However, it struggles to create something big from the ground up. Too much responsibility from the start can overwhelm it. This post explores the interactions with the model used by Cursor and the feedback it gave when asked to share.
Read More