The Leverage Field Manual

Effort is bounded by hours in the day; the multiplier you apply it through is not.

The one idea

Productivity advice optimizes the wrong variable. Wake earlier, time-block, work harder — all of it assumes effort is the term to maximize, when effort is the least scalable input you have. A doctor working 80 hours sees twice the patients of one working 40; that’s linear, and linear is a trap. The people whose output seems to defy the hours aren’t working harder, they’re applying force through a different mechanism.

Leverage = Effort × Multiplier

Most careers are spent trying to grow the left term. It’s bounded by biology. The right one isn’t.

The engine

Four multipliers, of which the piece covers two properly.

Code is the most accessible, and Naval’s word for it — permissionless — carries the whole argument. No gatekeeper approves your software. The defining property is zero marginal cost of replication: the work is done once and then serves the millionth user exactly as it served the first. The manual adds a second zero-cost property that gets less attention and may matter more: code compresses learning. A physical business gets two to four real iterations a year; a software team runs hundreds of experiments a month. The advantage isn’t only that you serve the next user free — it’s that you find out what’s true faster than anyone operating in atoms.

The practical heuristics are where it earns its keep: do the automation math before automating (time saved × frequency × five years, against build cost — and treat that as a floor, since it ignores context-switching, error rates, and the fact that each automation makes the next cheaper). Automate the bottleneck, not whatever’s easiest, because automation ROI follows a power law. Build only what differentiates; buy the undifferentiated heavy lifting.

Then the AI section, which is the piece’s actual contribution. Traditional software automates procedures; AI automates judgment — ambiguous input, reasonable decision. That category used to scale linearly with headcount. And because AI collapsed the skill barrier to producing code at all, the bottleneck moved: the question stopped being “can you code?” and became “do you know what’s worth building?” Domain expertise just repriced sharply upward relative to technical skill.

Media is the second lever, and its distinguishing feature is that it works on two axes simultaneously — a podcast episode builds distribution and credibility, where a billboard does the first and a private meeting the second. Hence the compounding: 80% evergreen against 20% timely, every piece republished in five formats, rent platforms for acquisition but own the relationship for retention. The back catalogue is the asset. Three hundred solid posts beat five viral ones, because a library creates thousands of permanent entry points and a spike creates one.

The best heuristic in the section is the least obvious: audience multiplies your other multipliers. Distribution lowers the activation cost of code, capital, and hiring all at once, which is why it’s the right lever to pull first.

My take

The automation-math heuristic is the most immediately usable thing here, and the insistence that it’s a floor rather than a ceiling is the part I’d keep. The naive version of that calculation talks people out of automations that are obviously correct, because it counts only the direct minutes and ignores that a five-minute interruption can cost thirty minutes of focus.

The observation I’d stand behind longest is the bottleneck shift — “do you know what’s worth building?” That reframes a decade of career advice. If producing software is no longer scarce, then taste, domain knowledge, and problem selection are the constraint, and those are acquired differently and much more slowly than syntax.

Where it gets thin

The piece is unfinished. It ends mid-sentence in the Capital section, and Judgment — the fourth multiplier, and arguably the one that governs the other three — is never written. There’s also a \sectionsection typo where \section{Capital Leverage} should be, which will not compile. Anyone reading this summary should know they’re reading about two and a half of four levers.

Beyond the truncation: the examples are extreme survivors. MrBeast and Rogan are the top of a distribution with millions at the bottom, and the manual’s own Part-I logic elsewhere would demand the base rate. “Build an audience” describes an outcome, not a method.

The AI claims are also the most perishable thing in the library. “70 to 80 percent quality” and “AI collapsed the skill barrier” are observations with a short half-life, and the piece would be stronger if it dated them explicitly rather than writing in the present tense.

The distilled principle

Stop asking how to get more done. Ask what mechanism would make the same work keep paying after you’ve stopped doing it.