Hacker Equity

A New Asset Class

Private equity made much of its money before anyone touched the work. In the easy-money window the return could come from the position more than from fixing what was bought. A fund bought a healthcare company, financed it with other people's debt, held it while multiples drifted up, and sold it into a market that paid more for the same earnings. Bain estimates that in the low-rate era, steadily rising entry and exit multiples powered over half of all buyout returns. The gain was bought at entry, not built during the hold. Where that was not enough, the fund bought four more clinics and called the sum a larger company. The denial pattern at the billing desk, the referral a scheduler still chased by hand in a payer portal, the intake field a front-desk staffer typed three times into the EHR: in that window those could stay exactly as broken as the day the fund wired the money, and the return survived anyway.

Trace the incentive and the behavior explains itself. A fund earns a fee on capital committed and a share of the gain on exit. Both reward the size of the position and the cleverness of the financing more than the hour spent inside a payer portal. So the people who set up the structure built the skill they were paid for. They learned leverage, multiple arbitrage, and the rollup, then wrote a generic operating playbook and handed it down a deck to a portfolio operator who was three layers from the work and one bonus cycle from moving on. Paul Graham's old line is the spine of this argument: to get rich you need measurement and leverage, a position where your output can be measured and your decisions have a large effect. The holder of a financial position had the leverage and let the measurement of the actual operating work go slack, because the structure never paid to tighten it.

The Honest Chart

Here is where most arguments like this one cheat, so I will not. The decomposition you pick decides the story you get to tell, and there is more than one honest decomposition.

The charts that flatter my thesis are real. The cleanest large-cap decomposition, 395 European deals from 1991 to 2007, puts about half of a 56% gross deal IRR on financial leverage and only about a third on the firm's own value creation. StepStone's model buyout across 1,533 realized deals puts leverage at 57% of value creation, with only 6 of its 14 multiple points coming from passive market re-rating. Read those two and the position did the heavy lifting.

Now the chart that does not flatter it. In a broad sample of 701 exits from 1990 to 2013, EBITDA growth was the single largest driver of returns at about 41%, ahead of leverage at 31% and multiple expansion at 18%. And across 2,951 deals from 1984 to 2018, leverage's share of value creation fell from roughly 70% before 2000 to about 25% in the last decade, with most of the value attributed to non-market factors. So I will say it against my own argument: the operating work always mattered. In broad, long samples it ranks first. What changed is not that the work was ever worthless. What changed is that the position can no longer carry the deal without it.

The era closed the gap itself. Bain's own stylized scenario makes the shift concrete: a 2015 deal needed roughly 5% annual EBITDA growth to hit a 2.5x return; with cheaper leverage gone and no multiple tailwind, the same deal in 2025 needs 10 to 12% to clear the same bar. It is an illustration, not a deal-population average, but the direction is not in dispute. There is even a tell for which lever the playbook always skipped. Across 762 software deals over about a decade, revenue growth drove about 52% of returns and multiple expansion about 42%, while margin expansion contributed just 6%, and 94% of the deal theses, 31 of 33, had promised around 560 basis points of margin gain that the actual result badly trailed. Margin is the operating lever you only move by doing the work inside the business. The decks promised it and the holds did not deliver it, because the structure put the people who could deliver it three layers from the desk. The part of the return that the structure was built to avoid paying for is now the whole margin.

The New Leverage

So the leverage moved. For most of the buyout era, the operating improvement needed a team, a year, and a playbook, and even then it landed unevenly because the people running it did not own the result. Now one builder with focused agents does that operating work directly. The builder maps the denial codes coming back from the payer and writes the thin script that flags the recurring reason on the day it appears instead of three weeks later in a month-end report, and the billing team works it while the claim is still inside the appeal window. The work a generic playbook described on a slide is now work a single person can actually do.

It used to live on the balance sheet, in the debt and the multiple. Now it lives in the build. A single owner who reads the claims worklist and ships a reversible bridge between the RCM system and the scheduling report does the specific operating work the playbook only described, and carries the result, which the portfolio operator three layers away never did. The cost of doing that specific operating work fell toward the cost of one builder's attention, and the return follows that cost down to wherever the work is now cheap to do, which is inside the workflow, in one builder's hands.

The Scarce Thing

I will be exact about the size of the lever, because a sophisticated reader will catch an inflated one. There is no measured study showing a single builder moves operating EBITDA by an order of magnitude, and I am not going to pretend there is. In a controlled trial of 95 professional developers, those with an AI assistant finished a defined task about 56% faster, though the confidence interval ran from 21 to 89% and the task was greenfield, not a mature codebase. And in a 2025 trial, experienced developers were actually 19% slower with AI in code they knew deeply, while they believed they were 20% faster. The leverage is real, uneven, and self-reported gains run hot, so I will ground this on measured output, not on what an operator feels. Treat any ten-times or hundred-times figure as a direction, not a number. But read the dates on those trials before you read the numbers. They ran on models two and three generations back. This month Anthropic shipped Claude Opus 4.8, the current frontier, and raised on the order of sixty-five billion dollars at a valuation near a trillion to push the next one. So the measured figures are not the ceiling. They are the floor, taken before the tool got better again. The honest claim stays narrow: the direction is measured, the slope under it is steepening, and the magnitude is still unmeasured. I am not going to invent the number the studies have not run.

The result that matters is not the headline speed-up. It is the shape of it. Across 5,179 customer-support agents, an AI assistant raised issues resolved per hour by about 14%, roughly 34% for novices and close to nothing for the experienced top performers, because it spread the best operators' judgment to everyone else. That trial ran on a 2023 model, so read the 14% as a reading taken before the slope steepened. That is the lever an owner-builder exploits. The scarce thing was never the AI. It was the judgment about which denial code to chase, which referral to escalate, which intake field is the one that corrupts the chart. An owner who has that judgment can now encode it and run it across the whole company, on every claim, at once. And the encoding got cheap this month in a second way: building real operating software stopped being exotic. What got called vibe coding a year ago, a hobbyist novelty, now ships the bridge between the RCM system and the scheduling report in an afternoon. The agents are the leverage; the owner's judgment is the thing being levered. The tool to encode the judgment is no longer the constraint. The judgment is.

Some of this leverage predates AI. WhatsApp reached 450 million users with 55 people, and Instagram sold for a billion with 13, but those were network-effect software products, not operating companies. And the one-person billion-dollar company is still a bet among tech founders, not an achieved fact. Graham wrote his line about startups, and he argued the unit you can actually measure is the small group, not the lone individual. The single-builder-plus-agents version is truesilver's extension of his logic into a setting he did not have: a world where the agents are the group, the judgment doing the encoding is the owner's own, and that same leverage now reaches the administrative operating work inside a real company instead of only a consumer product. Read the timing the way you would read any spread. Cheap debt was an edge until everyone borrowed. Multiple arbitrage was an edge until everyone paid up for the same earnings. The rollup was an edge until the fourth deal looked like everyone else's fourth deal. Each was a real advantage that got crowded and bid away. Hacker equity is the same kind of spread, one builder out-executing a fund because the cost of doing this specific operating work just collapsed, and it is open now for the same reason it will close: the tool is commoditizing fast. When mapping the denial code and shipping the bridge is table stakes, the spread is gone, and the same capital now flooding the frontier, the near-trillion-dollar raise above, is the capital that arrives late to this trade, after the edge is already priced in. The spread belongs to whoever does the work first. Waiting a year is not neutral. It is choosing to enter after the edge is priced in.

Owner as Builder

Name it. Hacker equity is ownership where the person who holds the upside is the person inside the workflow. The return comes from execution, not position. There is no fund team writing a thesis, no operating partner translating the thesis into a deck, no portfolio operator translating the deck into a plan that a staff member finally tries to run against a payer portal that does not behave the way the deck assumed. The owner reads the worklist, builds the bridge, watches it catch a denial pattern, and owns both the EBITDA movement and the receipt that proves it.

It is more aligned for one structural reason. Every layer between the owner and the work is a place where the incentive can drift. The operating partner is measured across ten companies, so attention is rationed first. The consultant is paid by the engagement, not the result. The staff member is paid to clear today's worklist, not to question why the intake field gets typed three times. Remove the layers and the drift has nowhere to live. The person who decides what to build is the person who carries the loss if it breaks and the gain if it holds.

Private equity returned on a position. Hacker equity returns on the work.

What It Is Not

It is not a rollup with a model bolted to the side. The data does not say rollups fail; it says one specific kind does. Bain found buy-and-build deals run on multiple arbitrage alone returned 1.4x, barely above the cost of capital, while those with a real operating rationale, organic growth or margin work, returned 2.2x. And the arbitrage edge erodes as the program ages: the share of add-ons that were a single acquirer's fourth deal or later climbed from 21% in 2003 to close to half today. The loser is the financial-engineering rollup, not acquisition itself. EY found disciplined acquirers grew enterprise value and shareholder return nearly five times faster than firms that bought nothing. Pointing an agent at the billing of five clinics you bought to flip at the bigger multiple does not change where the return comes from. It adds software cost to the same position trade.

Healthcare made the failure mode legible. Of the 45 most distressed healthcare companies in late 2023, Moody's found 42 were private-equity-owned, and blamed roll-up strategies for the debt loads that left no room to adapt when the rate shock hit. The fault Moody's named was leverage with no operating cushion, not integration that was attempted and failed. The penetration was real. Private-equity-acquired physician practice sites grew roughly seven-fold between 2012 and 2021, with single firms passing 30% share in 108 specialty markets. But that was a scale-and-pricing play, not anyone doing the operating work inside the practices. Hacker equity is not that. It is not the fund that renamed itself around AI and kept the same fee structure, because the fee still rewards capital committed over work done. And it is not simply a larger team moving faster, because a larger team reintroduces the layers, and the layers are the thing being removed.

There is a test. Strip out the financing cleverness and the multiple, and ask what is left. If the answer is a company that got measurably better at the claim and the schedule because the owner did that work, the return was hacker equity. If the answer is only the spread between entry and exit multiples, it was the old thing wearing a new word. Most real deals have both. Hacker equity is the name for the deals where the work, not the spread, is load-bearing.

The Proof

Healthcare is where this stops being an argument and becomes an artifact, because the administrative drag is measured and large. U.S. health spending hit $5.3 trillion in 2024, 18% of GDP, and the slice of it that an owner-builder can actually attack is not a single tidy number, so I will not stack the studies into one. Peer-reviewed work puts administrative complexity at $265.6 billion a year, the single largest category of U.S. healthcare waste in 2019, and the only category for which the authors found no proven savings intervention. The officially measured administrative line is 7% of health spending, but that counts only insurer and program overhead and leaves out every dollar providers spend on billing and back-office work, so 7% is the floor, not the ceiling. Each figure is defined differently and from a different year; the honest move is to cite them apart, not sum them.

The drag is concrete enough to take apart one workflow at a time. Physicians report 39 prior authorizations a week and 13 hours of staff time chasing them, with 40% of practices keeping people working on nothing else, a bounded, repetitive task focused agents can lift off the clinical staff. Providers spent $25.7 billion fighting claims in 2023, and because about 70% of denials are eventually overturned, Premier estimates roughly $18 billion of that was wasted arguing over claims that should have been paid. The same data puts the cost to work a single denied hospital claim at about $57 in 2023, up from $44 the year before, a per-unit cost that compounds across thousands of claims and flows straight to EBITDA once it is closed. Industry estimates from the CAQH consortium, which has a commercial interest in the answer, report automation already avoided $258 billion in administrative cost in 2024 with $21 billion still on the table, and a quarter of provider organizations now using AI in administrative workflows. That is an interested party's number, and worth reading as one. But it points the same way the measured studies do. The direction is set. The work is unfinished.

So take the denial a billing team learns about only after the cash is late. The builder maps the denial codes coming back from the RCM system, writes the thin script that flags the recurring reason the day it appears, and a person who used to find out three weeks later now finds out the same morning, while the claim is still inside the appeal window. Less leakage, and you can point to the claim that triggered it. Take the referral a scheduler chases by hand through a payer portal: the builder uses computer-use to open the portal the scheduler already opens, read the status the scheduler already reads, and stop when the screen does not match the expected trace. Faster access for the patient, a cleaner handoff, and a log of every status it pulled. Take the intake field entered three times across the EHR, the scheduling screen, and the CRM, where the third entry is where the error enters the chart: the builder maps the source the first entry already created, and the field is entered once.

The discipline is the same one any honest builder keeps. Map the source that actually exists, even if it is a PDF emailed every Monday. Build a reversible bridge, so a bad change costs an afternoon and not a quarter. Leave a trail the owner can inspect: source file, exception reason, reviewer, final action. A model talking beside the work is theater. A model inside the work has a job and a receipt, and the receipt is the log that shows the claim, the source, the exception, and the reviewer, because the person who built it is the person who has to stand behind it. Clinical judgment stays with the provider, because a model classifying a referral note is doing administrative work, not deciding care. When EBITDA moves, it moves for a reason a person can point to. The return is sourced, not narrated.

The obvious objection comes here, and it is the right one. The honest concentration risk in hacker equity is the builder. If the upside depends on one person doing the specific work, the underwriting question is no longer the multiple at exit. It is whether the work the builder did leaves a trail that survives the builder. A reversible bridge with named sources, logged exceptions, and a reviewer is durable. A clever script only one person understands is a single point of failure dressed up as an edge. The discipline that makes the workflow legible to the next operator is the same discipline that makes hacker equity investable.

Aaron Swartz's point about the NUMMI plant was that it was not the workers, it was the system. That is exactly why ownership should sit with the person rebuilding the workflow, not the one blaming the operator inside a broken one. The intake field typed three times is not a lazy front-desk staffer. It is a system that never connected the registration form to the EHR, and a fund that was never paid to connect it. Trace the incentive, connect the registration form to the EHR, and the staffer who typed the field three times stops being the error and becomes the person who was absorbing a gap no fund ever paid to close.

The Playbook Loses

The playbook was a real edge once. It compressed generic operating knowledge into something a fund could apply across many companies at low marginal cost, which is exactly the advantage that holds when doing the specific work is expensive. But the cost of doing the specific work has fallen toward the cost of one builder's attention. The denial pattern a playbook could only describe in general, a builder now catches in this company, with this payer, on this claim. The generic loses to the specific the moment the specific gets cheap, and it just got cheap. The scarce thing is now the capacity to do the specific operating work. It is the lever the structure was built to skip, and the one the position can no longer carry the deal without. A fund that still treats the playbook as the moat is defending an advantage the falling cost of work already took, and the cost is still falling this month.

The Flag

This is truesilver's doctrine made literal. Small moves with large effects. One builder maps one source, ships one reversible bridge, and a line on the P&L moves for a reason an LP can read. Two client relationships per year, because the binding constraint is no longer capital to deploy but capacity to do the work, and the owner who is also the builder cannot be in two payer portals at once. An LP used to read a low company count as a scarcity line or a marketing pose. Here it is the underwriting. This does not scale the way a capital-raising fund scales, and it is not supposed to. It scales the way a craft scales, by the trail each build leaves, so the next builder starts from a mapped source and a logged exception instead of a blank portal.

The position is still for sale; anyone with capital can buy it. The work is not, because it belongs to whoever does it. Hacker equity is what you call ownership once the work is the return.