AI Code Generation

Reflections on the hottest new programming language — English.

Training Wheels for AI Devs

Current AI coding tools are “great at files, bad at projects.” Yet much of the complexity in modern software lives outside of individual files — in sparsely documented integrations, duct tape CI workflows, and other hidden pockets of tacit knowledge.

Improvements to context length, RAG, training data, and scale all conspire to close these gaps. But in the meantime...

• How can we structure software projects to make them easier for LLMs to grok?
• What is the "Phoenix, Arizona" of self-developing codebases?

Trivial Software + Long Tail "Niche Fracking"

As software becomes easier to write and maintain, software that never would’ve made sense before now becomes interesting.

Personally, I’ve been exploring this by attempting to use AI to create “trivial software” — small, single-serving utilities that eliminate day-to-day frictions. Ex: A recurring calendar invite that’s editable by all attendees.

Zooming out, the same premise likely also applies to SaaS. Long tail niches previously too small for large incumbents to bother with, might all of a sudden become economically viable.

90 / 9 / 1 Rule Reinvented

Technology lifts all boats, but not always equally. Similar to the observation that 1% of users are creators, 9% are commenters, and 90% are passive viewers — my hunch is that only 1% of people are well placed to fully take advantage of AI.

This is not just about capital and access — put provocatively, AI probably helps certain personality types more than others.

New Years

That special time of year where self-improvement is cool.

Foundations vs Speculative Bets

I notice most New Year's plans seem to cluster around Foundations — fitness, diet, habits, etc. But past relatively low thresholds, these are subject to rapid diminishing returns.

Assuming you already have decent foundations, I find New Years is better framed as a question about deliberate risks — "What speculative bets am I making in 2024?"

Memory Dividends

The value of personal photos can reasonably be measured in Warm, Fuzzy Feelings™️ multiplied across people and time.

If I were to chart this for my own photo collection, most it would be dead weight (read: never looked at again). But the handful of unicorn outliers are telling — they were mostly episodes where I was forced to curate the best photos for sharing (wedding, travel albums, 1SE).

Some easy wins here — digital photo frames, shared collections, even printed albums sent to family. Compounded over a lifetime, this is likely the easiest way to become a Warm, Fuzzy Feelings™️ billionaire.