The Dot and the Swarm
The 'bitter lesson' of AI: scale and compute beat hand-crafted domain knowledge. What that means for your AI strategy.
Why it matters
A reflective essay on Rich Sutton's 'Bitter Lesson' principle applied to modern AI deployment—arguing that brute-force scaling and search have consistently outperformed expert-designed systems. Relevant to practitioners choosing between custom-tuned vs. general-purpose approaches.
The key facts
11 to knowAuthor: Ethan Mollick (One Useful Thing)
Subject: Rich Sutton's 'Bitter Lesson' (1974–2019 computing history) reapplied to current AI
Core argument: general search + compute scale > domain expertise + hand-crafted features
No specific product launches, benchmarks, or deployment data provided
Framed as strategic commentary on how teams should approach AI builds
Author: Ethan Mollick, Wharton (One Useful Thing newsletter)
Published: 1 Oct 2026
Central thesis: 'bitter lesson' (Sutton, 2019) — raw compute and scale beat hand-engineered approaches; swarm intelligence follows the same pattern
Argues multi-agent coordination may outperform single large models at equivalent total compute
No benchmarks, measured deployments, or pricing data provided
Framed as opinion/commentary, not reporting a product launch, benchmark result, or deployment
Go to the source
One Useful Thing (Ethan Mollick)oneusefulthing.org
Publisher excerpt: Benefitting from the Bitter Lesson