I Won’t Show You Mine

Twenty years ago, a famous pod shop sent me their standard IP form. I walked away. The contract that landed on my desk in 2005 is about to land on yours.

In 2005, a very famous pod shop wanted me to come work for them. The recruiter was friendly. The compensation would have been excellent. The infrastructure on offer — capital, technology, operations, the entire institutional machinery a systematic trader would otherwise have to build for himself — was the kind of thing you don’t normally walk past twice.

Then they sent over the IP form.

It was black and white. Every rule I wrote, every model I built, every strategy I developed while in their employ would belong to them. The recruiter, to his credit, tried hard to talk me through it. There was nothing to talk through. The contract was clear, and it was non-negotiable. I said no and walked away.

Twenty years later, every knowledge worker on earth is about to be handed a version of that contract.

The form that landed on my desk in 2005 was unusual for its time — directed at the small number of people, like systematic traders, whose work product was legible enough to be copied. Discretionary traders didn’t get that contract, because nobody could actually extract what was in their heads. Lawyers didn’t get it, because their judgment didn’t fit on a page. Doctors didn’t get it for the same reason.

That’s about to change. Everyone in knowledge work is on the verge of producing exactly the kind of legible, portable, codified artifact that did land me that contract two decades ago. The agents are coming. The skills folders are coming. And the contracts are coming after them. This post is about why.

A different deal, same year

Later that year, I made a different deal.

Joe Ritchie was a legend. At one point in the 1980s, his firm Chicago Research and Trading was doing more options business than any other firm in the world. CRT pioneered the use of computers in options pricing — Joe famously programmed Black-Scholes into a Texas Instruments calculator before most of the floor knew what Black-Scholes was. He sold CRT to NationsBank in 1993 for $225 million. He came to trading by way of a Wheaton College philosophy degree, a stint as a Chicago bus driver, and a job as a guard at the Cook County Jail; he learned the markets from a book a friend handed him. Self-educated in the only sense that matters. Genius in every sense.

He was also a straight shooter — devout, principled, and scrupulously honest in business. Before we agreed on anything, I asked if my wife could come meet him. Joe said not only that she should — she must. A few weeks later, there we were, in his boardroom in suburban Illinois, my two kids coloring on the boardroom table, aged four and seven. He was happy to have the whole family. We shook hands on the deal shortly after.

I traded Joe’s prop capital for three years. We shared ideas freely; he gave me ideas, I gave him ideas, and neither of us laid claim to the other’s IP. There was no contract that swallowed what one of us brought to the table. We simply trusted each other not to do the thing that the pod shop had asked me to sign away in writing.

He was, and remains, the only person I have ever met who saw markets in exactly the way I did. We had different approaches and different personalities — every system has to match the personality of the person running it, which is a different post — but on every question that mattered, we agreed. What a rule is. Whether a rule should be traded. Whether a rule is real or robust. The value of systematization. The definition of systematization. The value of what Joe called “trader logic” over the academic frame of mind that produces, for instance, the Barron’s columns advising readers to use options to take directional positions. (You don’t. You trade options to bet on volatility. If you want to bet on direction, you bet on the underlying. That this is not obvious to a class of professional commentators is itself a useful piece of evidence about what “trader logic” is and isn’t.)

That kind of agreement is what makes a handshake deal possible. It is not enough to trust someone’s character. You also have to share their definition of the thing you are trusting them about. Two people who disagree about what a rule even is can sign every contract in the world and still end up suing each other. Two people who agree about what a rule is can shake hands and never need a contract. Joe and I agreed about what a rule was.

The contrast matters because it isolates the variable. Joe and I were doing exactly the work the pod shop wanted me for — developing systematic rules and applying them with capital. We were sharing IP in both directions, which the pod shop’s contract would have made formally impossible. The thing that was different wasn’t the work. It was the structure of the relationship. Joe was trading his own money. He had no reason to own the rules — only to use them. Pod shops are running other people’s money on behalf of investors who are paying for institutionalized, transferable IP. The contract follows from the business model.

Joe passed away during the pandemic. I still miss him.

What systematic traders knew first

The reason I bring up the pod shops isn’t aesthetic. I have specific, technical reasons for thinking they’re not very good at the thing they claim to be selling.

In November 2025 I published a paper in The Journal of Portfolio Management called “The False Promise of Drawdown Rules: New Evidence and a Better Framework.” The paper used ETF and long-short data from 1993 to 2022 to test the standard pod-shop “risk management” rule — the one where if you’re down X%, you’re cut. I showed that those rules, even in perfect-hindsight scenarios, turn certain profits into losses more often than not. They don’t manage risk. They manufacture it. Pod shops succeed in spite of their drawdown protocol, not because of it. The whole apparatus is incoherent on its own terms, and I think anyone who has worked under it long enough knows this.

There’s a more general lesson hiding in this finding. Externalized rules in the wrong hands actively destroy value. The pod shops’ own written-down risk rules make them worse traders, not better ones. If they can’t codify their own risk management without manufacturing losses, what reason is there to think they’d handle a codified version of someone else’s IP any more competently? The technical problem with their risk protocol and the structural problem with their IP grab are the same problem manifested twice — an institution organized around the wrong theory of where the value lives.

So when I tell you I refused to sign their IP form, understand that I had also looked carefully at the rest of what they were offering and concluded it wasn’t worth giving up the rules for. The contract was the most visible problem, but the building behind it had structural issues, too.

But the IP problem was the deepest one, and it’s the one that’s now generalizing.

Here’s what systematic traders knew first, and what’s now becoming a fact about every white-collar field. Trading IP, when it lives in someone’s head as discretionary judgment, is essentially impossible to extract. The pattern recognition of an experienced market-maker, the gut-feel of a macro trader who has lived through three crises, the spatial intuition of a vol trader who has watched skew misbehave in real time — none of that fits on a page. You can’t put it in a contract because there’s nothing to put in.

But the moment you write down explicit rules, all of that changes. Now there’s a thing to take. There’s a piece of paper. There’s a file. There’s a contract that says it’s mine.

The current decade has produced a tidy little series of legal cases that show exactly how this plays out, and the trajectory is escalating, not stabilizing.

In 2019, Citadel Securities sued the British hedge fund GSA Capital over an automated trading model called the “ABC Strategy.” Citadel said the model had cost more than $100 million to develop and was so closely held that just 15 of the firm’s roughly 3,000 employees had access to its strategic logic. The dispute was triggered by GSA’s attempt to recruit a Citadel quant. Citadel argued in court that GSA “couldn’t unsee” what its hire had brought with him. The case settled confidentially in 2021. Whatever the precise terms, the precedent was that you could litigate over the contents of a trader’s head if those contents had once been written down.

In April 2024, Jane Street sued Millennium Management and two former traders, Doug Schadewald and Daniel Spottiswood, over an India-options strategy that had generated about a billion dollars in profits for Jane Street in 2023 alone. Jane Street alleged that the strategy’s profits dropped by half in March 2024, the month Millennium reportedly began using it. The two traders, in their defense, argued that the strategy wasn’t a trade secret at all — that it was based on their experience and expertise rather than any protected algorithm or signal. That defense is the entire post in one sentence. Where does the trader end and the strategy begin? When the rules are explicit, they belong to the firm. When they’re implicit, they belong to the trader. The case settled in December 2024, but only after Jane Street had been forced to disclose sensitive details in court — a small demonstration of what employment lawyers now call the “enforcement paradox”: the harder you fight to protect a trade secret, the more of it you have to reveal in court.

Earlier this year — February 2026 — Citadel went after Daniel Shatz, a former portfolio manager who took global credit at Marshall Wace. Citadel alleged that Shatz had accessed proprietary documents — recruiting lists, strategic roadmaps, candidate evaluations — while still employed and during a fifteen-month garden leave that preceded his formal departure. Fifteen months. Pause and consider that. The firm was paying him to do nothing for over a year because they were afraid of what he would do if he were free to act sooner. The IP problem has become so acute that the cost of avoiding it now includes paying senior people to sit on their hands for over a year.

The arc here is the part that should worry the rest of you. In 2009, the IP fights in finance were over code and information (Sergey Aleynikov leaving Goldman for Teza, Citadel suing Teza’s founders). In 2024, they were over strategies. In 2026, they are over the garden-leave intervals between strategies. The thing being protected is getting more abstract, the protection mechanisms more elaborate, the legal stakes higher. And until now, this has been a peculiar little drama playing out in finance, watched by the rest of the world with a mixture of curiosity and contempt.

It’s about to stop being peculiar.

The smartphone bit, briefly

The cliché version of what I’m about to argue is the smartphone analogy: by 2027 everyone will have AI agents the way everyone has a phone now. That’s true and it’s trivial. The same model will be available to everyone. The same chat interface, the same general capabilities, the same default toolkits. There will be no scarcity in the thing.

The smartphone analogy nails one thing: the social-expectation treadmill. Right now, “I’ll get back to you next week” is normal. In two years, that response will sound the way “let me check my answering machine” sounds now. Why didn’t your agents have it ready Tuesday? You’ll get used to that question, the way an entire generation got used to “why didn’t you text back.”

But the smartphone analogy fails on the part that actually matters. Phones are isomorphic across users. Yours and mine do the same things and produce the same value for both of us. Agents won’t work that way. The capability of an agent depends entirely on what you’ve taught it to do for you, in your domain, on your problems. The tool is generic. The configuration is everything.

That configuration lives in a folder.

What a skills folder is — without showing you mine

A skills folder is the file system in which an expert tells an agent how their domain actually works. It contains documents, written in plain language, that codify procedures, conventions, decision rules, common errors, edge cases, and the specific corrections that turn the agent’s enthusiastic competence into actual usefulness. It is not a collection of one-off prompts. It is a permanent, evolving body of instruction.

I have skills for backtesting trading strategies, for generating code against my own specifications, for supporting my physics research, for building and querying my reference library, for editing prose. The publishing piece — the part that helps me write Substack posts — is the smallest tail of what’s there. The rest is the real work.

I’m not going to show you any of those skills. That’s the most important sentence in this post.

I’ll tell you what category of failure each of them addresses, because that’s enough to understand what’s at stake. The trading-strategy skills exist because agents will, by default, write you a backtest with subtle look-ahead bias and report it confidently. They will use future information to predict the past. Every quant who has ever shipped a paper-trading system before discovering this knows the feeling. A skill, written carefully, catches it. The physics research skills exist because agents will, by default, get computational details in particular calculations exactly wrong in ways that look right to a non-physicist. The code generation skills exist because agents will, by default, ignore conventions specific to the codebase they’re working in unless you write down what those conventions are. The reference-library skills exist because agents will, by default, hallucinate citations.

The general principle: a skill is a codified record of where a generic agent fails in a specific expert domain, plus the correction. It is the externalization of tacit knowledge. It is the thing that takes me from being one person with an agent — which is what everyone will have — to being one person with a configured agent that knows how a particular kind of work actually has to be done.

But here’s the thing the skills also are: they are the sort of artifact that can be lifted. They are exactly the kind of thing the pod shop’s contract would have said belonged to them. They are the same form, in a different domain, of the artifact I refused to externalize in 2005.

This is why I’m not going to show you mine.

It’s also why this post is, in a small way, an enactment of the argument it’s making. I’ve told you the moat exists. I’ve told you what shape it has. I’ve explicitly refused to walk you through the contents. That refusal is the model, and the first thing anyone with a working skills folder should understand is that nobody who has one is going to show it to you.

The mechanism

The reason a skills folder is uniquely valuable — uniquely yours, structurally — has nothing to do with the cleverness of any individual skill. It has to do with who can write one in the first place.

A skill cannot be written by a non-expert, because writing one requires knowing which agent failures matter and which are noise. The trading skill that catches look-ahead bias is only writeable by someone who has shipped a system that contained look-ahead bias and learned the hard way to recognize it. The physics skill is only writeable by someone who knows that a particular calculation has a sign error in textbooks A through D and a different sign error in textbook E. The legal skill is only writeable by someone who has personally been on the receiving end of a discovery review that went sideways.

The skill is, in this precise sense, a record of accumulated scar tissue. The agent provides the muscle. The expert provides the scars. You can give an agent to anyone, but you cannot give scars to anyone, which is why the folder is the real asset.

A skill also cannot be written by the agent itself. The agent doesn’t know what it doesn’t know. The whole reason a skill is needed is that the agent’s confidence and the agent’s correctness diverge in places only an expert can see. Asking the agent to write its own skills is exactly the same epistemological category as asking a freshman pre-med to write the residency-training curriculum. There’s no path from inside.

Tacit knowledge, in the philosophy-of-science literature, was always understood as the part of expertise that couldn’t be verbalized — the irreducible human residue. What’s happening now is that, for the first time, we have a tool — the agent — that creates pressure to verbalize as much of it as possible. And the people doing the verbalizing are the experts themselves, in their own voices, in their own folders.

This is a good thing, in the abstract. It is the most efficient knowledge-transmission mechanism ever invented. It is also the conservation law that defines the new labor market.

In knowledge work, productivity gains and IP defensibility are now conjugate variables. You can have one or the other at full strength. You cannot have both. This is the systematic trader’s old dilemma, and it now belongs to everyone.

There is exactly one way out, and we’ll come to it. Everyone who doesn’t take it lives with the dilemma.

The trap

Here is what’s about to happen — and is already happening — in fields that have nothing to do with finance.

A junior transactional lawyer at a big firm — the office lawyer, not the litigator who’ll be in court — builds a 200-line skill that captures her firm’s contract-review process. The specific clauses to flag in M&A diligence, the formatting conventions, the standard redlines, the way to summarize a target company’s liabilities for a partner’s memo. The skill saves her 20 hours a week. Her billables look extraordinary. Her partners notice.

Six months later, she gets an offer from a competitor. Her current firm sends her a letter reminding her of the IP clauses in her employment agreement. The skills she wrote, the firm asserts, were written on firm time, using firm matters, under firm supervision, and constitute work product belonging to the firm. She is welcome to leave, but the folder stays.

A radiologist builds a skill that encodes her department’s reporting conventions — the phrasing, the diagnostic priors that fit the patient population, the specific failure modes of the imaging equipment that produces 80% of her cases. The skill makes her twice as fast and improves her accuracy. The hospital, which paid for her training, the equipment, and the patient relationships, has a view about who that skill belongs to.

A management consultant builds a skill that captures the way her firm sells engagements — the diagnostic frameworks, the standard slide structures, the negotiation patterns that work and don’t. She is, in a real sense, more productive than any consultant before her. The firm has thoughts about whether she gets to take the folder when she leaves.

Each of these people has just produced, in 2026 or 2027, exactly the kind of artifact I was asked to produce inside the pod shop in 2005. And each of them is about to discover that the law of intellectual property does not care about the medium. A trade secret is a trade secret whether it’s an options-pricing algorithm or a discovery-review checklist. The contract clauses that protect employer ownership of work product are not going to politely exempt skills folders from their reach. The Jane Street–Millennium case turned, partly, on whether explicit trading rules counted as a protected trade secret. The next decade’s version of that case will turn on whether explicit legal rules, medicalrules, consulting rules count.

They will. The legal infrastructure for this fight already exists. The Defend Trade Secrets Act, the body of state-level non-compete law, the standard work-product clauses in every white-collar employment agreement — none of this was designed with skills folders in mind, but all of it applies. The asset is portable, the asset is valuable, the asset was created during employment using the employer’s resources. The fight writes itself.

Not every white-collar role is equally exposed. Therapists, salespeople, executives whose work is primarily relational — and litigators, whose work is performed live in adversarial courtrooms — produce less codifiable output than a transactional lawyer or a radiologist or a quant. The trap applies to the codifiable portion of knowledge work — a large and growing share, not a total one. One implication worth naming: the discretionary work that has historically been the higher-prestige half of several professions — the trial lawyer over the office lawyer, the master clinician over the junior diagnostician, the partner running the room over the associate running the spreadsheet — is about to become more valuable, not less. The trap closes hardest on the work that can be written down. Work that can’t be written down stays safe.

And whether the firms will win those fights even within the codifiable share is itself a live question. The Jane Street defense — strategy as experience and expertise, not protected algorithm — is exactly the argument workers in every codifiable field will start making. If the courts buy it, the trap stays porous. If they don’t, it closes hard. We don’t yet know which way the law lands. What we know is that the fight is coming.

And here is what makes this particularly cruel: the better you are at your job, the worse this trap is for you. The most productive lawyer at the firm has the most valuable folder. The most productive radiologist has the most valuable folder. The most productive consultant has the most valuable folder. The structural pressure of agent-augmented work concentrates value into a portable artifact, which means the workers most worth fighting over are also the ones most exposed to the contract. The pod shops always knew this. It’s why their contracts were drafted the way they were. Now everyone gets to learn.

Three positions

When the trap closes, the labor market stratifies into three positions, each with real costs.

The first position is the W2 employee on the sanctioned firm stack. The firm provides the agents, the skills folder lives on the firm’s drives, the IP clauses are tight, and the productivity is real but rented. You are more productive than the workers of 2024, but the productivity is captured by your employer. Most large institutions will end up here for the same reasons they ended up with company-issued laptops and company-managed email: compliance, audit trails, IP protection. This is not a bad position. It is just not a position with a moat.

The second position is the 1099 contractor, the consultant, the solo practitioner with their own stack. Your skills, your folder, your IP, your risk. You eat what you kill, but you also keep what you build. This was not the systematic trader’s typical path. The typical path was the first position — at firms whose contracts captured everything the trader learned. The IP cases I described earlier are exactly what the typical path produces on exit. What’s new is that fields where the W2 career was the stable default — law, medicine, design, research, consulting — are about to face the same fork systematic traders faced, and a higher fraction will take the solo path: partly because the IP grab pushes them out, and partly because the agent makes solo work newly scalable.

The third position is the loophole — the one exit from the conjugate-variables trap I flagged earlier. It’s the one I took, and it’s still available. Externalize the rules — but only for yourself. Build the folder, refine it, depend on it, and refuse to transfer it. The rules live in writing; they just live in writing you own. This is what walking away from the pod shop’s IP form actually meant in 2005, and it’s what it has meant since. The skills exist. They simply don’t belong to anyone other than me.

This stance used to be expensive because solo work didn’t scale. The cost of keeping the folder yours was approximately the cost of staying small. Agents change that math. A solo practitioner with a mature private folder can now do work that previously required a team — without ever turning the folder into something an employer can capture. The third position used to be a niche. It’s about to be a real option.

What you still give up: exit-by-acquisition, because acquisition requires the folder you refuse to hand over. Formal institutionalization, because nothing has been formally transferred to anyone. The ability to train juniors using your own materials, except in arrangements where you can actually enforce confidentiality — which most arrangements aren’t. And the appearance of belonging anywhere obvious — there’s no firm name on the business card. Many will look at that trade and decline it. That’s part of the story, not a weakness in it.

If you want a single image for the new labor market: the LinkedIn profile of 2028 reads “Samir Varma + 14 agents. References available.” And the agents have references too. The W2 worker’s profile reads “Samir Varma at FirmName. Agents licensed.” And the third worker’s profile, the one who built the folder and kept it private, just reads “Samir Varma” — because there’s nothing on offer. The folder is fine. It’s just not for sale.

The strongest counter, killed

The strongest objection to all of this is what I think of as the frontier-eats-the-stack worry. If the next-generation model is sufficiently smarter than the current one, your carefully written skills folder becomes obsolete. The plumbing-level skills — the ones that compensate for current-generation failures — get absorbed into the model itself. The moat is a mirage that recedes as the frontier advances.

Some of this is true. Current skills folders contain a lot of brittle, model-specific scaffolding that will indeed be obsolete in two years. Anyone writing skills now should expect a real maintenance burden as models change. There is no permanent skill, the way there is no permanent quant strategy.

But the structural argument survives. The skill is not a static asset; it is a continuously updated record of where this generation’s agents fail. As models improve, the failures shift, but they don’t disappear. The agent that no longer makes look-ahead errors in backtests will start making subtler errors in regime detection. The agent that no longer hallucinates citations will start misattributing them. The agent that no longer ignores codebase conventions will start over-applying conventions where they don’t fit. The work renews. It does not finish. And the orchestration and taste layer — the part of the skill that decides which failures are worth catching, which outputs are worth using, which corrections are worth codifying — does not depreciate at all. That layer is where the expert lives. It’s irreducibly yours.

To make the distinction concrete: the skill I wrote that catches look-ahead bias in backtests will become obsolete when models stop making look-ahead errors. The judgment that look-ahead bias was the failure mode worth writing a skill for does not. The first kind of skill is plumbing. The second is expertise.

The frontier eats the plumbing. It does not eat the expertise.

I won’t show you mine

Here is what I have told you. There is a thing called a skills folder. It is the externalization of expert tacit knowledge into a portable, configurable artifact. It is what makes one human-plus-agent more valuable than another human-plus-agent. It is also exactly the kind of artifact that creates the IP problem I walked away from in 2005, and that finance has been litigating over for the last decade and a half.

Here is what I am not going to tell you. Anything specific about what’s in mine.

The instinct to refuse, in 2026, looks like a personality trait. In two years it’s going to look like a labor-market strategy, and in five it’s going to look like the only sane labor-market strategy for anyone who built their folder themselves. The systematic traders saw this first. The lawyers and the doctors and the consultants are about to see it. There are excellent essays on the internet teaching you how to write skills, and the people writing them are doing good work. But look at who isn’t among them. There are no systematic quantitative traders out there walking you through their skills folders. There are no senior options market-makers publishing their factor libraries on Medium. The people who learned twenty years ago what externalized IP costs them are not the people teaching skills today. That absence is itself the lesson.

The first thing you are going to have to learn — same as I did, two decades ago — is that nobody who has one is going to show it to you.

You will have to figure it out yourself.

That, too, is the point.

---

If you found this essay worth your time, you may enjoy my book The Science of Free Will, which asks an equally uncomfortable question about an equally cherished story.

If you want to know more about India, start with The Paradox of India, the first essay — or browse the whole series.

Related country-specific work: Albion — Britain’s institutional decline, by someone who first saw the place in 1979.