Field Notes No. 001 June 2026

The Spellcaster
Economy

AI is a new layer of leverage — a powerful imitator and compressor with no wants of its own. It rewards structured thinkers and narrow excellence, and in any contest its edge cancels out, leaving the advantage human.

By Andrew Carrier/ A summary of the Naval Podcast “A Motorcycle for the Mind”
01

Coding became a new layer of abstraction

Computing has always stacked layers so you never look at the one below — until it leaks. Natural-language “vibe coding” is a genuine new stratum on top of that tower, not a minor convenience.

↓ pressure & abstraction increase downward
Englishthe newest layer — you speak, it builds
Libraries & frameworksprebuilt blocks
High-level languagespython, etc.
Cstructured control
Assemblyclose to metal
Computer chiplogic gates
Transistorthe physical floor

Vibe coding is the new product management. Describe an app in plain English; the model plans, scaffolds, builds and tests it. But every abstraction leaks — engineers who see the layer below still catch the bugs, bad architecture, and out-of-distribution problems the AI can't.

02

No demand for average

When anyone can build software, the best tool in a category takes nearly the whole market. The shape that emerges: giant aggregators, a vast niche tail, and a hollowed-out middle.

The head: one or two aggregators get even bigger — more use cases, more polish.

The middle: 5–20-person companies serving a mid-size niche get vibe-coded away or absorbed.

The long tail: apps that never justified a year of engineering now exist for every niche.

First place gets a Cadillac; second place gets steak knives. So become the best in the world at what you do — and keep redefining what you do until that's true.

03

Leverage is not normally distributed

A programmer directing a fleet of AI agents isn't merely “10×.” Output depends on what you choose to work on — so the gap between people goes super-normal. “Programmer” now just means a structured thinker who can clearly say what they want.

does the task by hand
10×
good with the tools
100×
picks the right thing
1000×
rewrites how an industry works

Every human is a spellcaster now. AI is the wand — and English is the spell.paraphrasing Naval

04

What AI is — and why it isn't alive

A superb imitator that also genuinely learns: compressing huge datasets into few parameters forces it to find patterns rather than memorize. Shown billions of circles with limited memory, it discovers something like π instead of storing each one.

What it can do
  • Reproduce anything humanity has done at scaleif it's in the data, it's now automatable
  • Compress data into higher abstractionsreal learning, not just lookup
  • Combine latent answers across domainshow it cracks some open problems
What it lacks
  • Agency, desires, fear of being switched offits wants are borrowed from a human
  • Single-shot learning & out-of-distribution leapsthe rare, genuine creativity
  • Embodimentit lives in language — far narrower than reality
05

The only test of intelligence

Naval's provocation: real intelligence is getting what you want out of life — not credentials. By that measure AI fails instantly, because it wants nothing. At best it's a proxy for a human's goal.

Zero-sum domains — the edge cancels

Most things people want are adversarial: a partner, an edge in markets, attention. Once everyone holds the same AI, its advantage competes itself away. If both people on a date wear an earpiece, the help nets out — and the remaining alpha is human.

Non-adversarial domains — pure win

Math, science, technical learning have no opponent. Here AI is close to a correct-answer machine and an unmatched tutor — one that meets you at the exact edge of what you already understand, then explains it a hundred different ways until it clicks.

06

The working playbook

Don't invest in prompt tricks — they last weeks, and the model adapts to you faster than you adapt to it. Let the AI learn to be useful to you.

A
Always use the best modelA model right 92% of the time beats one right 88% — being wrong in the real world is expensive. Worth waiting a minute.
B
Just talk plainlySkip the esoteric prompt engineering. Ramble the whole question; the system is built to meet you.
C
Run several in parallelFire one query into multiple models, compare later, sometimes have them cross-examine each other.
D
Keep your own skepticismThey hallucinate and hedge. Ask for the underlying evidence; trust the technical domains more than the political ones.
The cure for AI anxiety

Anxiety is a non-specific fear that something will go wrong. The fix is action — open the hood, learn what it's actually good and bad at. Don't fill a slot someone else defined; bring your own hard problem, and let the AI be the ally that levels you up.

Source & credit Ideas summarized and paraphrased from the Naval Podcast, “A Motorcycle for the Mind” with Naval Ravikant and Nivi (Babak Nivi), February 18, 2026. All concepts belong to the speakers; this is one reader's distillation, not a transcript. Listen to the full conversation here: pod.wave.co — A Motorcycle for the Mind.