The Second
Harvest

A doorbell company came to make most of the world's ball bearings. Nobody planned that.

In the eighteen-sixties, on the grasslands of what is now Uruguay and Argentina, cattle were slaughtered by the millions almost entirely for their hides and their tallow. The meat itself was close to worthless. There was no refrigeration to carry it to Europe, so the carcasses were largely left to rot. A German chemist named Justus von Liebig had, years earlier, worked out a process for concentrating beef into a shelf-stable extract, but it was hopeless economics at home, because it takes roughly thirty kilograms of lean meat to make a single kilogram of the stuff. Then an engineer named Georg Giebert saw the one thing that made it work. He did not improve the chemistry. He moved it to where the meat was free. The plant he helped build at Fray Bentos turned discarded flesh into a brown paste that became the Oxo cube, arguably the first global processed-food brand, advertised across fourteen countries with collectible cards. The factory is now a UNESCO World Heritage site.

I open with that story because it contains the whole argument of this paper in miniature. The value was already sitting on the ground. What was scarce was not the raw material or even the idea; it was a cheap, disciplined way to see that the material had value and to act on it. For almost the entire history of business, that seeing was the bottleneck, and it was expensive enough that most of the value stayed on the ground.

Consider a company called New Departure. It began in Bristol, Connecticut, in 1888, making a doorbell that ran on a clockwork mechanism rather than a battery. It set up in Bristol because that was New England's clockmaking town, and the thing it was actually good at was stamping tiny, precisely-toleranced metal parts in volume. Bicycle bells followed. Then coaster brakes, which forced the company to master small steel spheres. By the nineteen-thirties, this former doorbell maker is reported to have produced roughly three-quarters of all the ball bearings in the United States and about half the world's, and it was folded into General Motors. A ball bearing and a clockwork gear are, if you look past the shape, the same manufacturing problem pointed at a different object. The company's real asset was never a product. It was a capability, and the capability was portable into a market that looked nothing like the one it started in.

The strange companies are not curiosities. They are evidence that the upside of innovation was always present, and that reaching it was the hard part.

I could keep going, and later in this paper I will. A firm that sold industrial carbon black to iron foundries and tire makers invented the children's crayon, because selling schoolroom chalk had put its founders physically inside classrooms where they watched kids struggle with poor imported crayons. A brewery wrote the best-selling copyrighted book in history to settle arguments in the pubs that sold its beer. A factory built to grow baker's yeast grew penicillin in the same vats. In every case the transformation looks, at first glance, like luck. Look closer and it is always the same shape: a capability, a byproduct, or a customer's own behavior turned out to be valuable in a market nobody had connected it to, and one person or firm did the expensive work of connecting it.

Here is where the science begins to matter, and where the story turns from history into something you can use. The reason all that value stayed on the ground for so long is partly economic and partly human. The economic reason is that exploration was slow and costly: you could not cheaply generate a hundred candidate ideas, test them against reality, and let most of them die. The human reason is more interesting, and it is the subject of the next chapter. Our minds are not built to explore well. We are built to protect what we have.

The behavioral economists have measured this precisely. People feel the pain of a loss roughly twice as intensely as the pleasure of an equivalent gain, a finding so robust it anchors an entire field. An owner who has spent thirty years building a company does not experience a speculative new idea as a lottery ticket with attractive odds. She experiences it as a threat to a thing she loves. That asymmetry is not a flaw to be scolded out of people; it is the correct emotional accounting for someone with a lot to lose. But it means that, left to intuition, businesses systematically under-explore. They stay with the problem they know, or they buy generic software built for a different company's version of it, and the strange, valuable adjacency goes unexamined.

The picture, then, is a quiet paradox. The upside of innovation has always been real and often enormous, and the companies in this paper prove that a discarded byproduct or a transferable skill could become a category worth more than the business that spawned it. Yet inside the established company, innovation has stayed one of the least reliable investments there is, because the two things required to capture that upside, cheap exploration and disciplined judgment against our own instincts, have historically been scarce. Both have suddenly become abundant, and inside the last few years. One of them we can now buy. The other we can build. The rest of this paper is how.

Your gut has already ranked these ideas. The job is not to trust it, or to ignore it, but to make it explicit and then argue with it.

There is a moment, when I sit with an owner and lay out the possibilities their business is holding, when I can see them decide before they have consciously thought. It happens in the face, a beat before the reasoning starts. For a long time I distrusted that moment, because it looked like exactly the kind of snap judgment we are warned about. I have come to believe something more precise, and it is grounded in the science I lean on most. The brain does much of its ranking before conscious thought begins. The neuroscientist Antonio Damasio showed that people with damage to the emotional-signaling parts of the brain, whose pure reasoning is intact, become catastrophically bad at ordinary decisions, because the fast, feeling-based system that tags options as promising or dangerous is doing essential work. Kahneman gave us the two-system language for it. The point is not that intuition is wise or that it is biased. The point is that it is a prior: a fast, compressed guess built from everything the person knows and cannot articulate, and it is sometimes remarkably good and sometimes badly anchored, and you cannot tell which from the inside.

This is the hinge of how I work. The value of intuition is highest where a person has had long, honest feedback in a stable environment. A founder who has run a distribution business for thirty years has a gut worth listening to about their own operation. That same gut is close to worthless about a market they have never touched, and worse than worthless when it is defending an idea they have already fallen in love with. So the discipline is not to trust the gut or to override it. The discipline is to force it into an explicit number, a stated prior, and then to gather evidence and watch the gap.

The gap between what your gut believes and what the evidence shows is the single most valuable signal in the room. It tells you exactly where to look.

That gap is the heart of the instrument I build for clients, and it is worth dwelling on because it inverts how most people think about data. When the evidence agrees with the gut, you have learned very little; you already believed it. When they diverge sharply, one of two things is true. Either the person's intuition is holding knowledge the data has not yet captured, in which case you have found something the market does not know, or the person is anchored, in which case you have found a bias about to cost real money. Both are the most useful thing that can happen. A large divergence is not a problem to resolve; it is a flag telling you precisely where the next hour of attention is worth the most. One real divergence is worth more than ten confirmations.

The second body of science I trust is game theory, and it reframes innovation from a creative act into an allocation problem. Every hour you spend developing one idea is an hour not spent on another. Every dollar committed to a favored project is a dollar not available to the bet the evidence actually prefers. Innovation, seen clearly, is a game you are playing against your own opportunity cost, and the hardest move in that game is not starting things. It is killing them. Loss aversion, again, makes us terrible at it. We keep funding the idea we like past the point where the evidence says stop, because stopping feels like admitting a loss. So the most valuable structure you can impose is a set of explicit gates, agreed in advance, with kill-or-advance criteria that do not bend to mood. In behavioral terms, a stage gate is a commitment device: a promise your disciplined self makes to bind your future, hopeful, anchored self. The graveyard of killed ideas is not an embarrassment. It is the proof that the system has a spine, and, as I will argue later, it is an asset in its own right.

Owning the standard

There is a third strand worth naming, because it explains something about how markets actually get made, and because it comes with a warning. Some of the most successful companies in history did not win by making a better product. They won by owning the standard that governed the decision to buy. A brewery, Guinness, built a book of world records to settle the arguments that happened over its beer, and the promotional giveaway became the best-selling copyrighted book of all time. A potter, Josiah Wedgwood, more or less invented modern marketing in the seventeen-sixties, capturing a royal endorsement and then engineering scarcity and aspiration so that the middle class would covet what royalty owned. And a publicist named Edward Bernays sold more bacon in the nineteen-twenties not by advertising bacon but by surveying thousands of physicians, publicizing their agreement that a hearty breakfast was healthier, and letting the public read a manufactured consensus as independent medical science. The bacon-and-eggs breakfast is his invention.

Bernays is the shadow version of the same move. He manufactured something that only looked like evidence. The mechanism, own the standard that governs the decision and the demand flows to you, is exactly right. The durable version is simply the harder one: you build the real measure. You state your predictions before you run the test, so the goalposts cannot move afterward. You keep the ledger of what you got wrong. You let the number that decides whether to build be one that can come back and say do not build. That is not a softer version of owning the standard; it is a stronger one, because a standard people can trust is worth more over time than one they eventually catch you rigging.

So the science gives us a working definition, and it is not the romantic one. A good bet is not a flash of creative genius. It is a disciplined act of belief under uncertainty: an explicit prior, honestly stated; evidence gathered against it; attention aimed at the divergences; capital gated by rules rather than mood; and a standard of measurement you would be comfortable showing the person on the other side of the table. Every part of that is learnable, and, as it happens, most of it is now partly automatable. Which brings us to the machine.

For two centuries the build was the bottleneck. It is not anymore. That single fact rewrites the economics of everything in this paper.

In Delft, in 1869, a company was founded to make baker's yeast and spirits. Its craft was fermentation: growing a living organism in large vats under careful, sterile control. During the German occupation of the Netherlands, a microbiologist worked quietly with that firm to grow something else in the very same vats, using the very same craft. He grew Penicillium mould, and the company became one of Europe's first producers of penicillin. Growing yeast for bread and growing mould for a miracle drug are, mechanically, the same operation. You change the organism and keep the process. I love this story because it is the least romantic and most exact illustration of a principle I will build the rest of this paper on: the same apparatus can serve a second, far larger market, and the value is unlocked not by new machinery but by pointing the existing capability somewhere new.

Hold that thought, because there is now an apparatus of exactly this kind available to everyone, and almost nobody is pointing it at their own business yet. I mean the current generation of AI models, and the claim I want to make about them is narrow and consequential: the cost of the expensive part of innovation has collapsed.

Recall the two scarce ingredients from the first chapter, cheap exploration and disciplined judgment. Exploration was expensive because a human team could only generate, research, and pressure-test so many ideas before the money and the calendar ran out. When I ran a venture studio in the years before this, the standard way to surface a portfolio of bets across a group of companies was to gather smart people in a room, argue for weeks, generate a few hundred candidate solutions, and emerge holding twenty or fifty ideas worth validating. It worked, and it was the single most expensive imaginable way to learn what a business already half-knew. That cost structure is what has changed. The generation, the research, and much of the first-pass testing can now be run by a machine, continuously, at a fraction of the cost, which means the hundred-problem workshop stops being an event you can afford once a fund cycle and becomes a process you can simply leave running.

Here is what it looks like in practice. I point agents at a business's own history, years of operating data most owners have never actually read against the market it sits in. The agents surface problems, score them, and generate candidate solutions wide. Most of those candidates are bad, and that is the design, not a failure. They get tested against a matrix of what we know, and most of them die there, cheaply, before anyone has spent real money. The few that survive get built and backtested against reality. The models I orchestrate for this are the current frontier, and the frontier keeps moving. The most capable of them, like Fable 5, can hold and reason over a whole business's worth of context in a way that was not possible eighteen months ago, and the next models will do more. The direction is what matters: generating and killing a candidate idea now costs a tiny fraction of what it did a few years ago, and the fraction keeps shrinking.

When the build stops being the bottleneck, the constraint moves to judgment. The scarce thing is no longer making the idea. It is knowing which idea to make.

This is the part most commentary on AI gets backwards. The machine does not remove the need for human judgment. It moves the bottleneck onto it. When you could only pursue three ideas a year, choosing among them was a small part of the work and building them was almost all of it. When you can generate and cheaply test hundreds, the building is no longer where the difficulty lives; the difficulty is entirely in the choosing, the pricing, the killing, and the knowing when the machine is confidently wrong. Models still hallucinate. They will produce a beautifully reasoned case for a number that is simply false, and they will do it in the same even voice they use for the truth. Left unsupervised they are dangerous exactly in proportion to how persuasive they are. So the value of the human in this new arrangement goes up, not down, but it concentrates into a narrower and more demanding role: the person who sets the priors, prices the divergences, enforces the gates, and refuses to let a confident falsehood through. That role has a name, and it is the subject of the fifth chapter.

What comes next is more of the same slope, and it is worth being clear-eyed rather than breathless about it. The models will keep getting cheaper and more capable, and the horizon of what a single operator can generate and test will keep expanding. The improbable adjacencies that the strange companies in this paper reached by accident, the byproduct nobody valued, the capability nobody thought to move, will become things you can hunt on purpose, systematically, because you can finally afford to look in a thousand places at once. The workshop becomes a standing process. The quarter becomes a continuous feed. That is not a prediction about far-off artificial minds. It is a description of the tools that exist today, getting a little better and a little cheaper every few months, which is the most reliable kind of forecast there is.

One idea is a lottery ticket. Eight ideas run by rules is a strategy. The difference is not scale. It is that the math becomes something you can plan around.

A single innovation bet is frightening in exactly the way loss aversion predicts. Its base rate of failure is high, the failure is vivid and personal, and the upside, however large, is uncertain. Told as a story about one idea, innovation will always feel like gambling, and a careful owner will always, rationally, flinch. The reframe that changes everything is to stop telling it as a story about one idea. The base rate that terrifies you at a single bet becomes a planning assumption across eight. If you know, roughly, that most candidates die, that a few reach build, and that a smaller number become genuinely valuable, then a portfolio of bets is not a gamble at all. It is an actuarial exercise, and it is precisely the language that funds and disciplined operators already think in. The frightening number and the plannable number are the same number. What changes is whether you are holding one ticket or a book of them.

This is why the collapse in the cost of exploration from the last chapter matters so much more than it first appears. Cheap exploration is not just nice. It is the thing that makes a real portfolio possible for a company that could never have staffed one. When generating and testing a candidate cost a fortune, only the largest enterprises could run enough bets for the base rates to work in their favor. Now a mid-market business can hold a portfolio that only a giant could have afforded a decade ago, which means the actuarial logic that made innovation safe for the Fortune 500 is suddenly available several tiers down.

One problem, paid twice

There is a second reason a portfolio changes the economics, and it is the idea this paper is named for. Most engagements that solve an operating problem stop after collecting the first return: the business is worth more because the problem is gone, valued at the company's own multiple, and that value holds up when you sell. That first harvest is real and it is usually enough to justify the work. But a well-chosen problem holds a second return that almost nobody collects. If the problem you just solved turns out to be common across your industry, the solution you built is no longer just an internal fix. It can become an independent company, and you can hold founding equity in it. The same insight, harvested twice: once as profit inside your business, and once as a stake in the company your fix becomes when the industry shares the problem.

Remember the Delft fermenter. One apparatus, two markets, yeast and penicillin. The second harvest is not a metaphor; it is that fact, applied to a business problem. And it belongs to the mid-market for a reason that inverts the usual order of things. A founding stake worth a couple of million dollars is a rounding error to a billion-dollar enterprise, which is exactly why the large venture builders only chase nine-figure ideas. To a company worth twenty million, the same stake can be ten percent of enterprise value, earned from a problem the owner was going to pay to fix anyway. The smaller company gets the larger relative benefit from the identical stake. That is why almost no one has offered this below the very top of the market: the wins are too small for their cost structures and exactly the right size for yours.

A portfolio does not just spread risk. It converts a solved problem into an option on a company, and it turns the record of your failures into an asset.

The capability-migration stories from the first chapter are, in the end, portfolio stories. New Departure did not bet its whole existence on doorbells; the doorbell skill became bells, brakes, and finally the bearings that ran the automobile age. The glass-jar company Ball is now Ball Aerospace, which built parts of Hubble and the James Webb telescope. Some of these leaps were less a poetic transfer of craft than the disciplined acquisition of the right people, and the patience of a company that thought in decades rather than quarters. It points at the last and least obvious asset a portfolio produces.

Every ranked problem and every killed idea carries priced reasoning: why it looked promising, what evidence changed the picture, why it lived or died. Run enough bets and that record compounds into something proprietary, a library of what works and what fails and why, across companies and across quarters. The graveyard I defended in the second chapter turns out not to be a cost of doing business but a form of capital. A firm that has run this process across an industry walks into the next situation already knowing things the people in the room do not. The failures were never waste. Priced and kept, they are the map.

Most firms hand you a deck and wish you luck. The value dies in the handoffs. So I removed the handoffs.

If the machine has moved the bottleneck onto judgment, then the question that decides everything is who holds the judgment, and how few times the work has to change hands on its way from insight to something that ships.

The traditional way to do this work is a relay. A strategy firm hands you a deck. You hand it to an agency, which hands a build to a development shop, which hands something to your operators, and at every exchange a little of the original insight leaks out, because the person receiving it never sat in the room where it was formed. By the time anything ships, the sharp idea has been rounded off by a dozen translations. The value did not fail to exist. It died in transit. When I say I removed the handoffs, I mean something specific and slightly unusual: one person reads the business until the edge shows, prices it, designs the system that captures it, and answers for the number we wrote down together on the first day. Not because one person is heroic, but because the machine now makes it possible for one well-equipped operator to do what used to require a chain of firms, and every link you remove from that chain is insight you keep.

I am able to say this because I have done the pieces of it separately for a long time before they could be combined. I have built seven companies and run a venture studio. I have generated, on a single engagement, more than two hundred concepts, narrowed them to twenty-five of real value across twelve categories, and designed twelve new companies from the winners. And because a great many valuable problems end in a physical object rather than a piece of software, I keep a fully equipped prototyping lab and a dedicated engineering partner, so that a product can be designed, prototyped, and actually made manufacturable rather than merely described on a slide. That stack of capabilities used to live in separate firms. It now sits behind one desk, which is the only reason the no-handoff model works at all.

The instrument

The strange companies taught specific lessons, and each one is built into how Modven operates rather than admired from a distance. Guinness owned the moment around its product by building the reference book that settled the pub argument. The recurring moment around my work is an owner wondering where their portfolio stands and what it is worth today, so I built a living instrument that answers exactly that, continuously: every idea shown as a point of light, moving through the stages, re-priced as new evidence lands, narrowing to the few worth building. It is not a status report you wait for. It is the standard by which you watch your own innovation happen. Wedgwood taught that a status anchor and engineered scarcity beat volume, so the practice is deliberately small, a few engagements at a time, and it opens with a paid working session that is credited in full if we build, which removes the buyer's risk the way his money-back guarantee did. And the crayon makers taught that the valuable adjacency reveals itself only when you are physically in the room for another reason. That is what the opening engagement really is: not a proposal, but the on-site act of reading the business closely enough that the second problem, the one nobody hired me to find, shows itself.

You do not have to trust me. You have to trust the measure — and the measure is built to be checked.

Which returns me to the promise I made in the second chapter, the honest inversion of Bernays. An orchestrator with a persuasive machine and no discipline is a dangerous thing, because the machine will happily generate an eloquent case for whatever its operator wants to believe. So the discipline is the product. I state what I expect a system to do before it is built, so the result cannot be quietly reinterpreted as a success afterward. I keep the ledger of what I got wrong, because a record of honest misses is worth more than a wall of testimonials. I let the number that decides whether to build be one that can genuinely tell us not to build, and when it does, that verdict saves you from spending real money on the wrong system, and it is yours to keep either way. The kill decisions are visible, not hidden, because a practice that never shows you what it killed is a practice asking you to take its judgment on faith. I would rather show you the graveyard.

That is what I mean by orchestration, and why I use the word instead of a grander one. I am not the source of the intelligence any more than a conductor is the source of the music. The models generate and test at a scale I never could alone. The client holds knowledge of their business I will never fully have. My work is to set the priors, aim the attention at the divergences, enforce the gates against everyone's hopeful instincts including my own, refuse the confident falsehoods, and keep the standard honest enough that you never have to take my word for anything. Done well, it is quiet. It looks less like genius and more like a very disciplined way of paying attention, run across a portfolio, at a scale and cost that were impossible until about a year ago.

The strange companies needed luck, or an accident, or one stubborn person. The change worth caring about is that the luck can now be manufactured.

Return, one last time, to the men who found value on the ground. The engineer who moved the beef-extract process to where the meat was free. The doorbell makers who happened to master precisely-round metal. The chemists who invented the crayon because selling chalk put them in a classroom. The brewer who noticed that his customers argued. Each of them reached an improbable adjacency, and in every case the reaching depended on something they could not summon on demand: a chance observation, a physical proximity, a stubbornness that outlasted everyone's patience. They were, in the end, lucky, and their luck was not repeatable. You cannot build a company on the hope of standing in the right classroom.

The single most important thing I can tell you is that this is the part that has changed. The reason those adjacencies stayed rare was never that they were rare in the world; the first chapter argued the opposite, that value of this kind is scattered everywhere, sitting in byproducts and unread data and capabilities pointed at the wrong market. They were rare because looking was expensive, so you could only look in one or two places, and you needed luck to be looking in the right one. When the cost of looking collapses, you can look in a thousand places at once, and the thing that used to require a lucky accident becomes a process you can run on purpose. You do not have to wait to be standing in the classroom. You can search every classroom, continuously, and price what you find.

Value has always been lying on the ground. What is new is that we can finally afford to look everywhere at once, on purpose, and pick it up.

The horizon deserves a measured claim, not a breathless one. This is not a forecast of artificial minds or a world remade overnight. It is that the tools that exist today are getting steadily cheaper and more capable every few months, that this slope is about as dependable as forecasts get, and that its consequence for innovation is specific: the standing, continuous, portfolio-scale search for valuable adjacencies stops being a thing only the largest companies can afford and becomes something a mid-market operator can hold, run by one disciplined person and a machine. The workshop that took a room and a quarter becomes a feed that never stops. The one-in-a-hundred accident becomes a base rate you plan around. And the second harvest, the solved problem that becomes a company, stops being the occasional windfall of a Fray Bentos or a Ball and becomes a repeatable line item in a portfolio.

That is the whole of the argument. The upside of innovation was always real; the strange companies prove it. Capturing it required cheap exploration and disciplined judgment, which were historically scarce; the science of the good bet describes the judgment, and the machine has made the exploration abundant. A portfolio turns the frightening base rates into a plan and the solved problem into an option. And an orchestrator who removes the handoffs and keeps the measure honest can now do, from one desk, what used to require a chain of firms and a great deal of luck.

If any of this is useful to you, take it and use it; I wrote it to be quoted. And if you want to see what your own business is holding, the way to start is small: a paid working session that ends in a written brief, yours to keep whether or not we ever build anything, which will tell you whether the value I keep insisting is on the ground is actually on yours. I can't promise you a second harvest; no one can. What I can do is look for it with you, show you exactly what I find, and stand behind the number, which, in a field that has run for two centuries on luck and salesmanship, is rare enough.

Strategy, shipped. One problem, paid twice.