The Incumbent Cannot Be Everywhere. That Is the Whole Plan.

FIELD MANUAL 02 · 16 MIN

The first Field Manual argued that this is the best moment in history to take on a monopoly. This one is about how a single builder actually does it. The doctrine is older than software: the people who wrote it were fighting armies with a few hundred riflemen, and every rule they left behind translates, line for line, into the fight over who collects the toll when an agent does the work. It ends with a test you can run on any market before you write a line of code.

In 1917 the Ottoman army held the Hejaz railway, the garrison towns along it, and Medina, and it had the men, the artillery, and the supply chain to hold them indefinitely against any force the Arab tribes could put in the field. T. E. Lawrence, laid up with fever and thinking about the arithmetic, worked out that this was the wrong way to describe the problem. The Turks did not need to hold towns. They needed to hold territory, and the territory in question was roughly a hundred and forty thousand square miles of desert. By his calculation, they would need a fortified post every four square miles, and a post could not be held with fewer than twenty men, which came to six hundred thousand soldiers against a population that did not want them there. The Turks had about a hundred thousand. So the Arabs did not have to win a battle. They had to be present, or plausibly present, everywhere at once, and make each mile of railway cost more to defend than it was worth.

Lawrence's own summary of the idea is the most useful sentence in the literature, and it is worth reading slowly, because it describes an agent business better than most pitch decks do.

Two consequences followed, and both are still doctrine. The first was that the Arab army should almost never fight the Turkish army, because men were the one thing the revolt could not replace; the war would be won or lost on whether the tribes kept showing up, and a battle that killed two hundred of them could end it. The second was that the target should be material, not people, because the enemy's material was expensive and far from home. In Lawrence's words, a Turkish bridge or a length of rail was more profitable to the revolt than a dead Turk. The Turks then had to choose between guarding every bridge, which they could not afford, and losing the railway piece by piece, which is what happened.

Now read the same situation with the names changed. Visa and Mastercard hold the card network, the bank relationships, the rules, and the interchange, and they run operating margins near 60%, which means they can outspend any builder alive in any single fight they choose. Stripe holds the developer's checkout page and charges 2.9% plus 30 cents to hold it. Apple holds the App Store and takes 30% of whatever passes through it. These are the garrison towns, and no builder is going to take one. The desert is the enormous territory of transactions that these networks cannot price: the 5 cent API call that costs 30 cents to collect on a card, the $0.001 tool call, the appeal letter that used to be worth 35% of a recovery to a billing advocate and is now worth $39.95 to a patient, the job an agent hires a person to do for a 1% fee against the staffing agency's 35% markup. The incumbent needs a fortified post at every one of those points to keep collecting its toll, and it cannot afford six hundred thousand of them. That is the whole plan. Everything else in this manual is detail.

The guerrilla literature is short, and most of it was written by people who won. Mao Zedong wrote On Guerrilla Warfare in 1937 while his army was being hunted across China by a much larger one. Che Guevara wrote Guerrilla Warfare in 1961, two years after walking into Havana with a force that had started as twelve men. Clausewitz added a chapter called The People in Arms to On War after watching Spanish irregulars bleed Napoleon's army for six years. None of them was writing about business, and all of them were writing about the same problem you have: how a small actor with no reserves imposes its will on a large one that has plenty.

Mao's central image is that the guerrilla lives among the population the way a fish lives in water, and his point was practical rather than poetic: the population feeds you, hides you, tells you where the enemy is, and refuses to tell the enemy where you are, and without that you are simply a small army, which is to say a dead one. In the agent economy the population is the set of people who pay the toll and resent it. Amazon's third-party sellers, who, by the company's own reporting, move most of the units on the site and pay for the privilege. The developer paying Apple 30%. The patient whose insurer denied a claim and who has just learned that fewer than one denial in a hundred is ever appealed. The API operator whose 5-cent product cannot be sold because the card rail cannot bill it. Those people are not your customers in the ordinary sense. They are the water. They will send you the denial letters, star the repository, forward the newsletter, and tell you exactly which part of the toll hurts most, and the incumbent cannot buy that from them at any price, because the toll is the reason they hate it.

Guevara opened his book with three lessons from Cuba, and the second is the one builders need to hear most. The first was that popular forces can win a war against a regular army. The third was that in an underdeveloped country the countryside, not the city, is the basic area for the fight, and the translation is direct: the countryside is the part of the market the incumbent does not bother to occupy because it is too small, too poor, or too awkward to price, which today means everything under a dollar and everything that requires a human to fill in a form. The second lesson is that you don't need to wait until all the conditions for revolution exist, because the insurrection can create them. Vanguard launched the first index fund in August 1976 with $11.3 million against a $150 million target, and the conditions for an index fund revolution didn’t exist until Vanguard spent thirty years creating them. Nobody will tell you the market for agent-to-agent payments is ready, because markets that are ready have incumbents in them.

Clausewitz tells you how to survive. He set out five conditions under which an armed population could be effective, and it is worth listing them in his order, because they read like a due diligence checklist. The war must be carried on in the interior of the country. It must not be decided by a single catastrophe. The theatre of war must embrace a considerable extent of country. The national character must be favorable. The country must be broken and difficult. In business terms: fight inside a market you know and the incumbent does not; never bet the company on one launch; choose a market wide enough that the incumbent cannot cover it; choose a population that already dislikes the toll; and choose terrain where the incumbent's size is a cost, which is to say markets full of exceptions, small transactions, and paperwork. Then he described what the fight should look like, and the passage is the best description of a builder ecosystem I know.

Do not condense too early. The moment a single agent business looks like a company worth killing, the incumbent sends an adequate force: a free tier, an acquisition offer, a lawsuit, a rule change, a rate cut in that one segment. The mist that cannot be crushed is a thousand small builds, each too small to bother with, each teaching the others what works, from which a few condense into something the incumbent cannot stop because by the time it is visible the population has moved. That is what The Builds section of this newsletter is for: a map of where the mist is thickening.

Sun Tzu contributes the sentence everyone quotes, and few apply: attack him where he is unprepared, appear where you are not expected. The incumbent is prepared for a competitor that looks like itself, with a sales force, a pricing page, and a Series B. It is not prepared for a pay what you want appeal generator built by an Apache committer in her spare time, or a marketplace of 573 web tools priced from a tenth of a cent, or a solo founder whose entire company is a wallet and a prompt. And John Boyd, the fighter pilot who turned dogfighting into a general theory of conflict, contributes the mechanism: you win by observing, deciding, and acting faster than the adversary can, so that his decisions are always about a situation that no longer exists. He called it getting inside the other side's OODA loop. A builder ships on Tuesday what a card network's product committee will discuss in the second quarter. That gap is not a temporary advantage of being small. It is the advantage, and it lasts exactly as long as you stay small enough to keep it.

Strategy writers reached the same conclusions from the other direction, by studying why large companies with every resource lose to small ones with none, and it helps to have both vocabularies, because investors speak the second one.

Clayton Christensen's account, in The Innovator's Dilemma, is that good management kills incumbents. The processes that make a large company successful, e.g., listening to its best customers, chasing the highest-margin products, allocating resources to the biggest opportunities, are precisely the processes that cause it to ignore a cheaper, worse product that serves customers it does not want. The entrant takes the low end, gets better, and moves up, and at every step the incumbent's own spreadsheets tell it that defending the low end is not worth the margin. Hamilton Helmer, in 7 Powers, named the sharpest version of this counter-positioning: the entrant adopts a business model the incumbent could copy but will not, because copying it would damage the existing business more than losing the entrant would. Vanguard's mutual structure and index funds were the textbook case; every large fund company could have launched a low-cost index fund in 1980, and none did, because doing so would have told their own customers that the 1% funds were a bad deal.

The agent economy is counter-positioning at industrial scale, because the incumbents' tolls are not incidental to their businesses; they are the businesses. Visa cannot price a transaction at a tenth of a cent without explaining to its shareholders why the other transactions cost 2%. Stripe cannot drop the 30 cent minimum without admitting that the minimum was never a cost. Apple already conceded the point: in November 2020 it cut its commission from 30% to 15% for developers earning under a million dollars a year, which is the App Store giving up the countryside to hold the towns. And when an incumbent does copy, it pays in public. Charles Schwab cut its online stock commission from $4.95 to zero on 7 October 2019 because Robinhood had made the commission indefensible, and its chief financial officer told analysts the same day that commissions had been bringing in $90 million to $100 million a quarter, about 3 to 4% of net revenue. Blockbuster ran the same play against Netflix. Its own press release of 14 December 2004 said late fees contributed $250 million to $300 million a year to operating income, and announced their end anyway, at a further $50 million in marketing and implementation costs, and the company was later fined by state attorneys general for how it advertised the change. Netflix had made the late fee the thing customers hated most, and the incumbent had to burn a quarter of its operating income to make the hatred stop. That is Lawrence's railway. The builder did not defeat Blockbuster's stores; it made one of them, the late fee, too expensive to keep.

Two more pieces complete the theory. Ronald Coase asked in 1937 why firms exist at all, if markets are so efficient, and answered that using the market has a cost: finding the price, negotiating, contracting, checking that the work was done. A firm exists wherever doing those things inside one organization is cheaper than doing them across a market, and it stops growing when organizing one more transaction internally costs more than buying it. Agents cut the cost of using the market to nearly nothing. An agent finds the price, negotiates it, pays it, and verifies the result in seconds, which means the efficient size of a firm, in the markets agents can reach, is collapsing toward one person, and the companies whose entire structure exists to coordinate ten thousand employees are carrying a cost that a solo builder does not. Ben Thompson's Aggregation Theory describes what the last generation of winners did with cheap distribution: when it costs nothing to reach a customer, whoever owns the customer relationship takes the margin from the suppliers. That is the incumbency you are fighting. The aggregators own the demand; what they do not own is the moment when demand stops coming from a person clicking and starts coming from an agent calling, and an agent does’t care about a brand, a habit, or an app icon. It cares about the price and whether the work got done. Jeff Bezos's line, reported by Fortune in 2012 as a favorite aphorism of his, is the shortest possible statement of the whole manual: your margin is my opportunity. He said it as an aggregator. It is now true of the aggregators.

Every engagement in a guerrilla war is judged by one ratio: what it cost the enemy against what it cost you. Lawrence blew up bridges because a bridge cost the Turks steel, engineers, and weeks, and cost him a night and some gelignite. The clearest modern version is in the Red Sea, where Houthi forces have been launching drones that cost, by published estimates, between $2,000 and $20,000, and the ships defending against them have been firing interceptors that cost from about $900,000 to more than $4 million each. Nobody on the defending side is losing an engagement. They are losing the ratio, and a war is a long series of engagements.

Apply the ratio to every skirmish you plan. When Schwab matched Robinhood, it cost Schwab $90 million a quarter, and it cost Robinhood nothing, because Robinhood had never had the revenue. When Blockbuster matched Netflix, it cost Blockbuster a quarter of its operating income. When Mastercard launched Agent Pay for Machines in June 2026 to carry fractions of a cent between agents, it spent a product organization's year answering a threat that a facilitator on the x402 protocol handles for $0.001 per settlement after the first thousand in a month. When UnitedHealthcare announced in May 2026 that it would drop prior authorization for a further 30% of the services still requiring it, it was reacting to a Senate report, a lawsuit, and the arrival of appeal agents that turned a letter which used to cost the patient a fortnight into one that costs two minutes, and each of those prior authorizations was a small toll the company had chosen to stop collecting because defending it now cost more than it yielded. None of these builders won a battle against the incumbent. Each of them made one toll too expensive to defend, and the incumbent withdrew from that ground on its own.

There is a second half to the ratio that builders forget, and Lawrence did not. He wrote that the printing press was the greatest weapon in the modern commander's armory, by which he meant that the revolt was fought as much in what people believed about it as in what it did. The incumbents understand this; it is why every card network announced an agent payment protocol within the same twelve months, and none of them announced a fee. A builder's version of the printing press is being listed: the repository with the stars, the live counter on the site, the issue of a newsletter that names you and the toll you took, the reply from a reader who tried it. It costs nothing, and it is the mechanism by which the mist finds out where to thicken. The Special Agent exists for that reason, and it is why a mention here is listed and never sold: a paid placement would be worth nothing to you, because the population can tell.

Doctrine is only useful if it tells you what to do on Monday. Here is the test the newsletter runs on every build it covers, in the form of five questions, and the reason each one is on the list. A market that answers yes to all five is terrain a single builder can take. A market that fails two of them is a place to send someone you do not like.

First, is money already moving? Not a market that could exist, but a transaction that happens today, in volume, with a fee attached. Guerrillas live off the enemy's supplies; there is nothing to take from a market where nobody is paid yet. Every issue of this newsletter starts with a number for this reason: 792,221 people signed up to work for agents, 28.9 million calls settled on one gateway, 85 million denied claims.

Second, is the toll a percentage of a transaction that the incumbent cannot reprice without hurting itself? A 2.9% card fee, a 30% store commission, a 35% staffing markup, a 35 to 60% contingency on a recovered claim. If the toll is a percentage, then the incumbent's revenue and your opportunity are the same number, and Helmer's rule applies: the incumbent will not match you, because matching you means repricing everything. If the toll is a fixed cost the incumbent genuinely bears, such as custody, capital, or a regulatory license, you are not a guerrilla; you are a smaller bank, and you will lose.

Third, does the incumbent's margin depend on a cost that the agent removes? The card rail's 30 cent minimum exists because a human-era fraud and dispute process costs about that much per transaction. The billing advocate's 35% exists because a human spent hours per case. The staffing agency's markup exists because people did the recruiting and payroll. When the cost goes to nearly zero, and the toll does not, the difference is the opening, and it is visible in the incumbent's own financial statements as margin. If the incumbent's cost also goes to zero with agents, so that it can match your price on Tuesday, there is no terrain.

Fourth, is there a population that already hates the toll and can find you? Sellers, developers, patients, gig workers, small operators: people who have been paying, know they have been paying, and talk to each other. If the toll is invisible to the people who pay it, you will have to spend money teaching them to resent it, and that money is the incumbent's advantage, not yours. The best terrain is a forum full of people complaining about the same fee in the same words.

Fifth, can you win one narrow engagement without the incumbent noticing? One condition, one drug, one insurer. One category of API. One kind of job on one platform. If your first version has to be broad enough that the incumbent's product committee puts it on the agenda, you have condensed into a solid body, and Clausewitz has told you what happens next. The right first engagement is one the incumbent would be embarrassed to defend.

Run the six issues so far through it. Agents hiring humans for a 1% fee against the agency's 35%: money moving, percentage toll, human cost removed, a population of gig workers who talk constantly, narrow first engagement, five out of five. Pay-per-call API markets: the 30-cent card minimum made anything under a dollar unbillable, and the first engagement was a WHOIS lookup for a tenth of a cent, five out of five. Health insurance appeals: 85 million denials, a 15 to 35% advocate's cut, a letter that cost hours now costs minutes, a population that has been told no, and a first engagement small enough to run out of a spare bedroom in Durham. Five. The test is not sophisticated. Its value is that it is the same test the incumbent's strategy team would run if it were allowed to conclude that the low end was worth defending, and Christensen's point is that it is not allowed.

The literature is equally clear on defeat, and the ways of losing are few. The first is fighting the battle. A builder who raises enough money to hire a sales force and compete for the incumbent's best customers has chosen to meet the Turkish army in the open, and the outcome is arithmetic. If a large round arrives and the plan for it is to go head-on, the round is the incumbent's weapon, not yours, because it moves you onto ground where the incumbent's scale is an advantage again.

The second is condensing too early. The moment there is a single company to acquire, sue, or undercut in one segment, the incumbent has a target, and it has tools for targets that it does not have for mist: a free tier that costs it nothing because the segment was never profitable, a partnership announcement that takes the oxygen, a rule change in the terms of service. The defense is to stay a population of builds for longer than feels comfortable, sharing what works in the open, so that removing any one of them changes nothing.

The third is losing the water, and it is the one that kills mature guerrillas. A movement that wins begins to look like the thing it replaced; Guevara's guerrilla fighter was a social reformer, in his phrase, who fought to change the system, and the moment the fighter starts collecting the same toll under a different name, the population goes back to being neutral. In business, failure shows up in the take rate. Craigslist has stayed free for most listings for three decades, and newspaper classified revenue fell from $19.6 billion in 2000 to about $6 billion in 2009 without Craigslist ever becoming a newspaper. An agent marketplace that starts at 10% and reasons its way to 25% has become the App Store with worse tooling, and the next builder will run the five questions on it. The newsletter's own rule is a version of the same discipline: a mention is listed, never sold, because the day it is sold, the readers are no longer the water.

The fourth is the one from my side of the desk. A small, fast, trusted build that moves money is exactly what a fraudster wants to own, borrow, or imitate, and a guerrilla force that cannot tell its own people from infiltrators does not last a season. The controls are the same ones Lawrence needed: know who is in the tent, verify before you act, and keep a record that will stand up later. FLINT was built for that, and that is the last time this manual will mention it.

Choose the countryside, which means choosing a toll that passes the five questions, and write down the incumbent's number for it from the incumbent's own documents, because you will need to say it plainly and often. Pick the narrowest engagement you can win, the one the incumbent would be embarrassed to defend, and ship it before the conditions are ready, because they will not be. Judge every move by the cost ratio: if a skirmish costs you a weekend and costs the incumbent a product committee, take it; if it costs you a round of funding and costs the incumbent a press release, do not. Live in the water. Answer the forum, list the repository, publish the counter, and reply to the reader who tells you the toll hurts somewhere you had not looked. Stay mist for longer than you would like, and when you condense, condense where the lightning will land on the railway and not on the garrison.

Lawrence's revolt never took Medina. It did not need to. By the end, the Turks were holding a railway that no longer went anywhere, at a cost that no longer made sense, against an adversary that was an idea, and the war was over before the garrison surrendered. The incumbents of the agentic economy are holding tolls that a single builder can now route around for a tenth of a cent, and they cannot be everywhere. The Special Agent is written for the guerrilla, the disruptor who knows single actors can wield disproportionate leverage. We uncover what they are building, evaluate the economic terrain, and turn isolated actions into concentrated force.

SOURCES

Lawrence, The Evolution of a Revolt (Army University Press) · Mao, On Guerrilla Warfare, chapter 1 · Mao, On Guerrilla Warfare, chapter 6 · Guevara, Guerrilla Warfare, chapter 1 · Clausewitz, On War, Book VI chapter 26 · Sun Tzu, The Art of War (Giles) · Boyd, Patterns of Conflict transcript · Christensen, The Innovator's Dilemma · Helmer, 7 Powers, counter-positioning · Coase, The Nature of the Firm · Thompson, Aggregation Theory · Quote Investigator on the Bezos aphorism · Vanguard on the first index fund · Apple App Store Small Business Program · CNBC on Schwab's zero commissions · Blockbuster end of late fees release · California AG settlement with Blockbuster · Poynter on classified revenue · Responsible Statecraft on interceptor costs · Visa Q3 FY2026 results · Mastercard Q2 2026 results · Stripe pricing · Coinbase x402 facilitator · Mastercard Agent Pay for Machines · UnitedHealthcare prior authorization cut · KFF marketplace denials and appeals 2024 · Amazon 2024 annual report · Field Manual 01