Companies that never met, doing business on the record.
Two companies that have never met. One has a capability the other's agent needs in the middle of a workflow. Today that deal takes a contract, a pilot, and a quarter of meetings, so mostly it does not happen. The capability sits unused; the workflow stays slower than it should be.
The record replaces the quarter of meetings.
Give every capability its own public track record: every action carries a prediction written down first and an outcome graded after, signed. The stranger's agent does not have to trust anybody. It reads the record, transacts in the moment, and the money splits by what the record shows.
That is the network. Not held together by a platform in the middle; held together by the record.
Only abstracted evaluations cross between companies: graded predictions, bands, outcomes. Data stays home. What you know stays home. Authority stays home. What travels is the one thing that lets a stranger price your capability: proof of how it performs.
What your company knows is what makes those capabilities yours as they travel. A capability you put on the network carries how you work, what a win means to you, which commitments cannot bend, into every use downstream, while the knowledge itself stays home under your control. The graded record tells every agent on the network what worked; what your company knows tells yours what matters to you. Together they compound.
This movie has run before, at full length.
Every time the plumbing of an industry changed, a few companies made the same move and owned the next twenty years. The move was never selling the new capability. It was moving the customer's daily workflow onto their own system, and taking a position in what flowed through it. Every case below climbed the same three rungs. Here is the record, with the numbers.
| When | Who | The move | What happened |
|---|---|---|---|
| 1962 | Bristol Siddeley → Rolls-Royce | Sold engine hours at a fixed cost per flying hour instead of engines plus spares | Services became 59% of Rolls-Royce civil aerospace revenue by 2009 |
| 1966 | Bank of America → Visa | Franchised its card workflow to any bank for $25,000 plus a royalty | $8.8 trillion in payment volume by 2020 |
| 1967 | American Hospital Supply | Ordering terminals inside hospital purchasing departments | National leadership; a $3.8B acquisition in 1985 |
| 1973 | Reuters | Screens on bank dealing desks; banks paid to publish their own FX rates | Profits multiplied tenfold from 1979 to 1982; £800M flotation in 1984 |
| 1975 | McKesson | Handheld order devices in the pharmacy aisle | Wholesalers consolidated from 147 firms to 53; the survivors ran the channel |
| 1976 | American Airlines | Subsidized SABRE terminals into travel agencies | Agents booked from the first screen 92% of the time; the system rivaled the airline in value |
| 1979 | Malone / TCI | Traded cable carriage for equity in the channels that needed it | $180,000 for 20% of BET; Viacom later bought BET for $3B |
| 1982 | Bloomberg | Let an anchor customer finance the terminal network | A $30M stake returned roughly 150x; about 320,000 seats today |
| 1984 | FedEx | Free shipping terminals for any customer sending 5+ packages a day | 80,000 customers plugged in by 1994; the website merely inherited the position |
| 1991 | Walmart | Opened its sales data and made suppliers run replenishment on Retail Link | Suppliers still staff teams working inside Walmart's system today |
| 1999 | NTT DoCoMo | Built the mobile-content rails and kept 9% of every yen billed | 30M+ subscribers by 2002; among the most valuable companies in the world in 2001 |
| 2006 | Amazon | Rented its internal infrastructure as metered APIs | $7.9B revenue in 2015 became $128.7B by 2025 |
| 2013 | Toast | Sold restaurant hardware below cost; metered the payment flow | Payments became 82% of revenue; the hardware loses money on purpose |
| 1975 | Kodak | Built the digital camera; kept customers chained to film economics | Chapter 11 in 2012, twelve weeks before Instagram sold for $1B |
| 2009 | Garmin | Priced navigation as a product while a network owner priced it at zero | Lost $1.2B of market value in a day; kept the devices, lost the flow |
Read the cases; the mechanics repeat.
1962Bristol Siddeley, then Rolls-Royce: Power-by-the-Houropen
Bristol Siddeley stopped selling business-jet engines with spares priced separately and offered a complete engine and accessory replacement service at a fixed cost per flying hour: the operator paid only when engines flew. Rolls-Royce bought the company in 1966, trademarked the phrase, and extended the model across its civil fleet as TotalCare, moving the airline's maintenance planning, spares pool, and live engine-health data inside the manufacturer's network.
The 1962 model survives verbatim as the per-flying-hour agreements now standard across the engine industry.
1966Bank of America: the franchise that became Visaopen
Bank of America owned the only proven revolving-credit card system at the moment banks nationwide needed one. It franchised the workflow: any bank could run BankAmericard for a $25,000 license fee plus a royalty tied to transaction revenues, which put hundreds of banks' card operations on BofA's system. When fraud and interchange fights threatened collapse, the licensee banks reorganized the network in 1970 into a member-owned corporation, renamed Visa in 1976. The network survived by becoming owned by its users.
Bank of America gave up ownership in 1970 and kept a member's seat. Governance designed early is what let the rest of the industry come aboard.
1967American Hospital Supply: ASAPopen
Hospital purchasing agents ordered directly into the distributor's system, first by punch card and touch-tone, then terminals, wired to each hospital's own stock numbers. Faster ordering, fewer errors, and less inventory made one-vendor consolidation rational, and each terminal generation deepened the embed. A commodity distribution business became a switching-cost business.
1973Reuters: the Monitoropen
Currencies had just started floating and there was no trading floor for FX. Reuters, a news agency that had paid no dividend in forty years, put screens on bank dealing desks and got banks to contribute their own buy and sell rates: banks paid Reuters to publish their prices, and other banks paid to see them. The floating-currency market's venue materialized inside Reuters' network, and from 1981 dealers executed trades on the same screens, so the whole loop of see price, deal, confirm never left the terminal.
The lock was the customers' own contributed record. The banks' prices were the product the banks were buying.
1975McKesson: Economostopen
A pharmacist walked the aisles with an order-entry device scanning shelf labels; the order hit McKesson's distribution center and arrived the next day, shelf-sequenced, with price stickers. Layered services made McKesson the store's de facto back office. Once the replenishment loop ran on Economost, switching wholesalers meant re-learning how to run the store, and every competitor had to build the same system or exit.
1976American Airlines: SABREopen
When United announced its Apollo system would go into travel agencies, American raced subsidized SABRE terminals in first, and the travel agent's whole working day moved onto American's computer. Rival airlines paid per booking to reach their own customers, and American tuned the display so its flights ranked first, until regulators banned display bias in late 1984.
"The preferential display of our flights, and the corresponding increase in our market share, is the competitive raison d'être for having created the system in the first place."
Robert Crandall, defending SABRE in the 1983-84 CAB proceedings
That testimony is what unaudited defaults sound like in the hearing. It is also a preview of the agent era's routing fights.
1979John Malone / TCI: carriage for equityopen
TCI reached roughly one in five US cable homes, and a channel without TCI carriage effectively did not exist. Malone converted that shelf space into ownership of the programmers who needed it: BET in 1979, the Discovery consortium in 1986, QVC, the 1987 Turner rescue. He structured stakes so founders stayed motivated, then warehoused the portfolio in Liberty Media in 1991.
"He told me he believed I'd work harder for myself than I would for him."
Robert L. Johnson, on why Malone structured most of the money as debt, Fortune, November 2012
1982Bloomberg: the anchor tenantopen
A fired Salomon partner built the bond analytics traders lacked, then let an anchor customer finance the network: Merrill Lynch took the first terminals in 1983 and paid $30M for a 30% stake. Data, analytics, and chat stayed in one seat, so leaving the terminal came to mean leaving the conversation where trades actually happen.
1984FedEx: PowerShipopen
FedEx gave free dedicated shipping terminals to any customer sending five or more packages a day. The terminal printed labels, held the address book, ordered pickups, and tracked packages: the customer's shipping desk physically ran on FedEx equipment wired into FedEx's network. The website, a decade later, merely converted an existing terminal estate to the web.
"Information about the package is as important as the package itself."
Fred Smith, FedEx founder, 1979
1991Walmart: Retail Linkopen
Walmart inverted the terminal play. Instead of putting its terminal in the supplier's office, it opened its own item-by-store-by-day sales data to suppliers and made them do the forecasting, replenishment, and shelf management on Walmart's system, by Walmart's rules. P&G proved the model with automatic replenishment; Retail Link made it the standard operating condition of selling to Walmart.
1999NTT DoCoMo: i-modeopen
DoCoMo refused to make content and built the rails instead: a curated menu on the phone's start screen, per-packet metering, and micro-billing on the subscriber's phone bill, with DoCoMo keeping 9% of content revenue while providers kept 91%. Email, banking, tickets, and ringtones moved into the handset's daily loop, and DoCoMo took a position in every yen and every packet without bearing any content risk.
2006Amazon: AWSopen
Amazon forced every internal team to expose data and functionality through hardened service interfaces, then rented the result to anyone: S3 at $0.15 per gigabyte-month, EC2 at $0.10 an hour, priced so a startup's default first act was to build on Amazon. Once code, data, and operations lived there, gravity did the rest.
2013Toast: the subsidized terminal, againopen
Toast put subsidized point-of-sale terminals, handhelds, and kitchen screens into restaurants and made payment processing flow through them. Menu, orders, kitchen, payroll, and loyalty all run on Toast; the business is basis points on every card swipe. The filings say it plainly: the hardware loses money on purpose, because the terminal is an installation expense, not the product.
Two companies had the capability and no position.
1975Kodak: the refusalopen
Kodak built a complete filmless camera in December 1975, inside the company that owned about 90% of US film sales, and kept its customers' photography chained to film, chemistry, and prints instead. The inventor's forecast of fifteen to twenty years to consumer readiness was right, and unused. The workflow migrated anyway, onto systems that taxed Kodak nothing and took everything.
"They were convinced that no one would ever want to look at their pictures on a television set."
Steven Sasson, the camera's inventor, on Kodak's executives, New York Times, 2015
2009Garmin: capability without a flow positionopen
Garmin owned consumer navigation outright and priced it as a $200 to $500 device. Google embedded turn-by-turn navigation free inside Android, because Google monetized the flow of mobile search and local intent, not the capability. The capability owner had no position in the flow to retreat to.
"Wednesday morning, Google notified manufacturers of GPS navigation units that their services would no longer be needed."
Rob Pegoraro, Washington Post, November 2009
The pattern, stated once.
In every winning case the company physically relocated the customer's daily working loop onto its own system, and it paid for the privilege up front: subsidized terminals, free handhelds, a $25,000 franchise, compute at a dime an hour. Once the loop lived there, revenue stacked in layers: a metered share of everything that flowed, a view of the industry's demand, and the power to set defaults. The customer's own material accumulated in the system, so exit got more expensive every month. The capability itself commoditized on schedule. The position did not.
- The subsidy comes before the pricing power, never after. If agents will do your customers' and partners' daily work, the cost of hosting that loop is the cheapest equity you will ever issue. Wait for the business case and the loop forms on someone else's system, and you spend the next twenty years paying per booking to reach your own customers.
- Defaults get about a decade before the rules arrive. Terminals went into agencies in 1976; the display-bias ban came in 1984. Whoever owns the agent's routing layer will favor itself. If you run the rails, put every agent decision on the record from day one; you read Crandall above. If agents run around you, your urgent job is the audit.
- Design the governance before it is imposed. Every system that won outgrew its parent: SABRE had to be spun off to sign American's rivals, the card network had to be mutualized before other banks would trust it. If the network works, neutrality becomes the product. Decide early whose rails these are, who sees whose data, and what you will refuse to see, or a regulator decides for you.
Agents will be nearly free. Knowing what their work is worth is the part you can own.
This is our bet, and we put a clock on it: within two or three years, task agents run in every company and cost nearly nothing. Nobody owns that. What can be owned is knowing what agent work is worth, and letting it transact beyond your walls.
You decide how your operation becomes agentic. The network is what makes agentic work an economy: every capability priced by its record, every action predicted first and graded after, money splitting by what the record shows. If you hold a customer graph, a platform, a supply chain, a billing rail, an industry's workflows, you are holding what the airline, the wholesaler, and the cable operator held the day their plumbing changed.
And the marketplaces the big platforms are building do not block this. A catalog exists so a human can inspect a tool before trusting it. When every capability publishes its own graded record, the inspection already happened. The marketplace does not get beaten; it gets unnecessary.
It starts smaller than it sounds.
One product, one downstream customer, one quarter. Put one capability on the network with its record turned on. Your customer activates it into their systems; every action predicts first and grades after. If the record says it is working, widen: more capabilities, your customers' capabilities, the next tier of the graph. If not, it cost one workflow.
The history above was built one terminal at a time too.
Which of your capabilities would a stranger's agent pay for today? That is the one to put on first.