Skip to content

Cutting Through the Slop: Lessons Learned from a Year of Claude Code

Code is cheaper than it's ever been, and it will never be this expensive again. Here's what survived a year of building damn near everything with Claude Code.

Introduction

Every few decades a technology flips the table on a profession, and software engineering is having that moment right now. The noise around it makes the moment hard to read, and most developers either haven't tried these tools at all or got slop and quit.

I spent the last year learning, building, and refining damn near everything with Claude Code. This talk is what survived contact with the work, checked against the research instead of the discourse.

The most-quoted study in this space found that AI made experienced developers slower. It also says several things past that headline that nobody screenshotted, and the authors have since redesigned the experiment.

The talk is built around three lessons that outlast whichever tool is winning this week:

  1. Use it. The slop you got in week one was the learning curve, not the ceiling.
  2. Codify it. An agent with no context writes the average of every codebase it has seen, and the average is not yours.
  3. Post-mortem it. Bad output is an incident to investigate, not a verdict to deliver.

Attendees leave knowing what the evidence actually shows, three practices that survive a tool switch, and why cheap code makes them worth more rather than less.

Delivery

This talk was given at:

The talk grew out of my five-part Brain Dump Ramblings on AI series: