We started with nine ways builders extract leverage and went looking for all of them. Here are
twenty-three distinct patterns, ranked into tiers by how much they multiply, how freely anyone can start, and
how long they keep paying.
The ranking criterion
A pattern's tier is its leverage score on three axes multiplied together:
Multiplier — how much output one unit of your effort becomes.
Permissionless — can you start without anyone's money or sign-off? This is
Naval Ravikant's key split:
labor and capital need permission; code and media don't. The permissionless ones rank higher because the door
is open to everyone.
Durability — does it keep working while you sleep, and for how long?
So the top tier is permissionless, compounding, near-zero marginal cost. The bottom tier is slow, gated, or
defensive. The honest caveat up front: this ordering is an argument, not a measurement, and it's built to be
disagreed with per row.
S Permissionless, compounding, zero marginal cost
Build or write once; it serves a million times, for free, without anyone's approval.
Open-source / code compounding — software others depend on. Naval's "code," the most powerful
permissionless lever. One git push can run on every machine on Earth. (Linux, SQLite, Ruff.)
Agent-native development — parallel AI agents doing the work. New since Naval's framing, permissionless
(anyone with an API key), and uniquely it multiplies every other lever on this list.
Framework / abstraction — code leverage's strongest form: you don't just ship software, you shape how
everyone else builds it. (Rails, Next.js, PyTorch.)
Media / audience — Naval's "media." Write, teach or record once; it reaches millions and compounds
into trust, adoption and distribution for everything else you do. (A widely read blog, great docs.)
Proprietary data / compounding dataset — a dataset that gets better with use and that others can't
copy. The real moat in AI, and it compounds automatically as the product runs.
A Very high, but need skill, timing, or a few people
Enormous multipliers, but they ask something of you first.
Standards / protocols — define the thing everyone implements (HTTP, the S3 API, MCP). Near-S
multiplier, but it needs adoption you don't fully control.
Platform / network effects — each user makes the product more valuable to the next. Compounds
ferociously past critical mass, which is the hard part.
Taste + small elite team — judgment about what's worth building, with no coordination drag. The
multiplier is quality per head. (Linear, Figma.)
Automation-of-self — eliminate your own repeated work so you never do it twice. Scripts, CI, infra
as code; the one-person company runs on this.
Timing / riding a platform shift — being early to mobile, to cloud, to agents. The beta on every
other lever; it multiplies whatever you already have.
Composability — small tools that pipe together, recombined without rebuilding. The Unix bet; cheap
to start, compounds across every future task.
Fast feedback loops — a tight inner loop so each iteration is seconds, not minutes. Small gains that
compound thousands of times a day, for you and your agents.
B Real, but gated by capital, org, or scope
Powerful, but they need money, people, or a company behind them.
Capital / compute concentration — Naval's "capital." Enormous, but someone has to decide to fund
you, and in AI the compute is the gate. (The frontier labs.)
Talent density / one great hire — Naval's "labor." A single exceptional person can outproduce ten;
but hiring is permissioned, slow, and they must choose you.
Own-the-stack / vertical integration — control every layer so nothing external caps you. Huge
ceiling, capital-heavy floor. (Apple silicon, SpaceX.)
Culture / mechanisms — org design that makes good outcomes the default (Amazon's written memos,
the Linear method). Compounds across everyone, but takes years and a team to build.
Radical simplification — leverage by subtraction: less to build, run and get wrong. Reliable, but
the upside is bounded by what you remove. (37signals.)
Delegation / orchestration — multiplying through others, whether people or agents. The manager's and
the agent-runner's lever; bounded by how much you can review.
Open-weights ecosystem play — give the model away to win the layer above it. Strategic and powerful,
but only works at scale you've already reached. (Llama, DeepSeek.)
C Slower, gated, or defensive
They matter, but they're situational, slow to build, or protect more than they multiply.
Brand / trust / reputation — reduces friction on everything you do, but it's earned over years and
lost in a day.
Capital allocation / M&A — buying leverage instead of building it. Needs capital and rare judgment;
mostly a large-company move.
Regulatory / compliance moat — leverage that keeps competitors out rather than making you faster.
Defensive and situational.
Energy / compute ownership — owning the data centres and power behind AI. The ultimate physical
substrate, but available only to the few with billions.
What the tiers say
Three things fall out of the ranking:
The top tier is all permissionless. Naval's point holds and sharpens in 2026: the leverage
that multiplies most is the leverage anyone can start without a gatekeeper — code, media, and now agents.
Agent-native development is the one genuinely new entry, and it's special because it's not
just another lever — it's a multiplier on the other permissionless ones. An open-source maintainer,
a writer, a solo founder all get more out of their existing leverage with it. That's why it jumped to the top
so fast.
The lower tiers aren't worse, they're gated. Capital, talent and owning the stack have the
highest ceilings of all — the frontier labs run on them. They rank lower only because you can't start them
from a laptop tonight. If you have the gate key, they're S-tier for you.
The practical read, unchanged from the companion pieces: stack the permissionless levers you can start now,
on a substrate of taste and specifying-before-building, and reach for the gated ones when you've earned the key.
The full taxonomy of how builders and teams use these is in the rest of this series,
and the method behind collecting it is the research stack.
Reanalysis: how this could be wrong
Turning our own tradecraft on this ranking surfaces two things worth stating.
The key assumption underneath it is that publicly visible leverage is the leverage that
matters; if that's wrong, the gated tiers are under-ranked and quiet, classified or non-Western work is missing
entirely. And the competing hypothesis we under-weighted on agent-native development: some of
its sudden rise is a gold rush, not a durable shift. The direction is real — the frontier labs build
this way — but any magnitude read from GitHub trends is inflated, because stars are gameable and
trend-driven. Treat the top of the A-and-above tiers as the confident part, and agent-native's exact rank as the
live bet.
Provided as is, for information only, with no warranty and no liability. Not legal or professional advice. Security and radio material is for authorized, lawful use only. Full disclaimer.
Visits are logged on our own server: IP address, browser and device details, and pages viewed. We use them to understand traffic and spot attacks, and never share or sell them.