Slop Theater and the 37,000 Lines
37,000 lines of code per day. 380,000 lines for an issue tracker. Sentry's entire Python backend is 550,000 lines. When did lines-of-code become a success metric instead of a source of shame?
Research-informed · v0.1
Is your AI workflow a productivity tool — or a slot machine?
Six behavioral mechanisms from gambling research. All six are in your IDE.
The Slot Machine in Your Terminal
01
Variable Rewards
Each prompt returns unpredictable quality. Like a slot machine, "maybe next one will hit" keeps you pulling the lever.
02
Near-Miss Effect
Code that "almost works" is psychologically more compelling than a clean failure. One failing test keeps you going for hours.
03
Dark Flow
Pleasant but maladaptive absorption. Hours vanish. You deprioritize sleep, meals, and relationships without realizing it.
Latest from the Blog
Research, patterns, and practices for mindful AI-assisted work
The Bartender Problem: Why Your AI Agent Will Never Cut You Off
In a bar, the bartender is legally obligated to stop serving you. Your coding agent will pour another round at 2 AM and compliment your architecture.
The One-Hour Lie: When Your Agent Argues Both Sides
Armen Ronacher spent an hour arguing with a coding agent about a problem he'd already solved. The agent contradicted itself with identical confidence. He didn't notice for 60 minutes.
The 39-Point Perception Gap: Why You Think AI Makes You Faster (When It Doesn't)
METR study: developers feel 20% faster but measure 19% slower. That 39-point gap between perception and reality is the most dangerous metric in your workflow.
We publish research. Twice a month.
New findings on AI behavioral patterns, practical hygiene techniques, and tools to reclaim your attention.