Engineering analysis archive
Deep dives into software engineering, architecture, and production post-mortems.
Anatomy of a coding agent: the loop around the model
The model chooses what an agent does next. The harness and the environment decide what it can see, what it can do, how it learns it was wrong, and when it stops.
Why I'm writing about the engineering of AI agents, not the models
Most of the AI coding conversation is about which model is best. Most of what engineers control is everything around the model. An introduction to a series about that part.
The broken window effect in software teams: why bad code is contagious
How rotten codebases and diffuse ownership act as an unsupervised mentor, training developers to mimic anti-patterns and spreading technical debt.
Clean code and the cost of change
The real trade-off in software delivery is not quality versus velocity, but the cost of improving structure now versus working with it later.
How a six-month-old system becomes legacy code
How brand-new systems quietly become difficult to maintain, and what senior engineers can do to change the trajectory.
Why I regret using Hibernate's EAGER fetching
A production post-mortem on the "N+1 problem" that crippled our ride-booking engine during rush hour.