AI Doc
Thinking Patterns

Thinking Patterns

Mental models frontier engineers use to make AI and systems decisions

Short, dense essays and case studies from frontier practitioners — the mental models actively driving modern engineering decisions. AI-specific paradigms sit alongside general engineering philosophy that applies to AI work.

1972
David L. Parnas

Decomposing Systems into Modules

The original information-hiding paper. Module boundaries should hide design decisions that might change — not mirror execution flow. Every modern module system (OOP, Rust traits, Go interfaces, microservices) descends from this.

1980
C.A.R. Hoare (Turing Award lecture)

The Emperor's Old Clothes

Hoare's Turing lecture. The 'billion-dollar mistake' (inventing null reference) + 'make it so simple there are obviously no deficiencies, not so complex there are no obvious deficiencies'. Shapes every modern language's Option/Result types.

1986
Fred Brooks

No Silver Bullet

Essence vs accidental complexity. 'No single advance in technology or management will give a 10× improvement in a decade.' 40 years on, vindicated by LLM coding productivity data (~2× not 10×).

2011
Rich Hickey (Clojure)

Simple Made Easy

The 60-min Strange Loop talk that gave engineers a vocabulary — simple vs easy, complect vs decomplect. 15 years later, still the default framework for architecture review arguments.

2014
Chris Olah

Neural Networks, Manifolds, and Topology

The blog post that framed neural networks as geometric deformation of manifolds. Foundation of the 'embedding space / feature direction' vocabulary used in all modern LLM interpretability — and seed of Anthropic's mechanistic interpretability program.

2017
Andrej Karpathy

Software 2.0

Reframes programming: neural network weights are source code. A 2017 prediction fully vindicated by the 2025 LLM era.

2019
Rich Sutton

The Bitter Lesson

70 years of AI research in 800 words: general methods that leverage computation win. The North Star of modern AI engineering.

2024
Case study · Musk / Kaplan / Chinchilla

First Principles in Engineering

First-principles thinking as a repeatable operation — not a mindset. Three cross-domain cases: SpaceX rocket economics (30× cost gap), Tesla cell-to-pack (55% cost cut), AI scaling laws (budget → calculable result).