marco

Paul Graham's advice that early-stage founders should do things by hand that they'll later automate — and why 'unscalable' is often the right choice in the beginning.

“Counterintuitive but true. The most successful startups I've seen all had a phase where the founders were doing the work manually. This essay explains why.”

Brian Potter's detailed history of how America built 48,000 miles of highway in 35 years - the planning, politics, engineering, and what it cost.

“The best explanation I've read of how large infrastructure projects actually happen. Spoiler: it wasn't just 'political will' - there were specific institutional innovations that made it possible.”

Gergely Orosz's insider look at hiring freezes, layoffs, and the AI pivot across Big Tech - based on conversations with engineers at Google, Meta, Amazon, and Microsoft.

“Best single source for what's actually happening inside these companies vs. what the press reports. Gergely's network gives him access most journalists don't have.”

Ben Thompson's framework for understanding how the internet shifted power from suppliers to aggregators — and why this explains everything from Google's search dominance to Netflix beating studios.

“The single best framework for understanding tech platform dynamics. Once you have it, you see aggregation everywhere — and you can predict which businesses will win without knowing any company specifics.”

Alfred Twu on why China builds towers in parks while America builds squat mid-rise blocks. The difference is regulation, not culture. Half of China's population lived in high-rises by 2015.

“The side-by-side comparisons are eye-opening. Same density, completely different urban form. China's approach of towers-in-parks produces more green space per resident than American mid-rises. Regulation shapes cities more than preference.”

Eugene Wei on how the metrics-driven culture of tech companies creates a bias toward visible work over important work - and why the best engineers learn to resist this.

“Eugene names something everyone in tech feels but can't articulate. The pressure to make work 'visible' warps what work gets done. The section on infrastructure engineers is spot-on.”

Paul Graham's early essay on how Lisp gave Viaweb a competitive edge — and on the Blub Paradox: why programmers using weaker languages can't see what they're missing.

“The Blub Paradox is one of the most useful concepts for understanding technology adoption and hype cycles. More true now than when it was written.”

Paul Graham's short, important essay on the question every early-stage founder should answer before optimizing anything else: is your startup default alive or default dead on current trajectory?

“The framing is useful precisely because it's binary. Most founders don't actually know which state they're in — they assume alive without running the numbers. This forces the calculation.”

Most AI breakthroughs are thought of as isolated advances, but Karpathy argues they signal a broader shift: code is being replaced by neural networks trained on data.

“This reframe changed how I think about software. Writing code is increasingly the wrong level of abstraction — you train the behavior instead.”

Anna Salamon on why humans default to 'locally good' actions instead of actually planning toward their goals - and how to notice when you're doing it.

“Short but devastating. Once you see this pattern in yourself, you can't unsee it. I now ask 'is this strategic or just comfortable?' at least once a day.”

The iconic hydraulic Atlas is replaced by a fully electric successor

“The shift from hydraulic to electric signals a move from research demos to commercial deployment”

Understanding model quantization formats from first principles

“Best explanation of model formats and quantization - essential for anyone serving models locally”

Why you probably do not need Redux

“Changed how I think about state management - most apps are way over-engineered”

Dan Abramov on four principles that make React components more resilient to change: don't stop data flow, be ready to render any time, no singleton components, and isolate local state.

“Still the best practical guide to writing components that don't break when you look at them sideways. Applies well beyond React.”