Target
Population
0
Avg Fitness
0%
Peak Fitness
0%
Diversity
0σ
Comparison
Pop
0
Fit
0%
Div
0

Productive Constraints

Bryan Cantrill argues that laziness — the constraint of finite human time — forces the crisp abstractions that make good software. LLMs lack this constraint; left unchecked, they generate 37,000 lines of noise per day. This simulation shows three constraint regimes producing visibly different outcomes from the same generative process.

ABUNDANT — "Everything lives"

Particles spawn freely with random colors and never die. The screen fills with noise because there is no selection pressure and no inheritance — abundance without constraint produces nothing coherent.

SCARCE — "Wall's laziness"

New particles are rare and inherit from the fittest, with small mutations. The unfit slowly fade and die. Scarcity plus inheritance produces convergence toward the target — this is what productive constraint looks like.

FORGETFUL — "The discontinuous mind"

Particles spawn frequently but die after 1.5 seconds with no inheritance. A trace memory strip preserves the 12 fittest particles ever observed. Moderate resources without memory produce constant churn — occasional brilliance that the system cannot retain.

Sources: Bryan Cantrill, "The Peril of Laziness Lost" (April 2026); Larry Wall, Programming Perl (O'Reilly).
View on ATProto ↗