The History of E-Commerce

First version 23 October 2026, last changed 24 October 2026 · Andreas Stockter

E-commerce, electronic commerce, is the buying and selling of goods and services in which the order, and usually the payment too, is handled over a data network, and no person on the seller’s side has to take it down. The buyer sits in front of a screen, the seller is a program, and the goods come by post. In 2025, goods worth around 6.4 trillion dollars were sold to consumers this way worldwide, about a fifth of all retail; in China the share was a quarter, in Britain a good quarter, in the United States a sixth, and in Germany, depending on how you count, 13.5 per cent or 83 billion euros.^{1, 2, 3, 4} Behind each of these numbers is a transaction that first succeeded in 1994 and that has since changed the high streets, the post, paying and, most recently, searching. This text tells how it came to that.

Like the history of AI, the history of the mathematics behind AI, the history of software development, the history of the Internet, the history of open source, the history of venture capital and the history of vibe coding, this text follows a chain: each era resolves a bottleneck the previous one left open, and creates the next one in doing so. Here the bottleneck has had the same shape since the first mail-order catalogue: between a buyer’s wish and the goods in their hand lie distance, time, money and trust, and someone has to bridge them. First the medium was missing, then the open network, then trust, then the business model, then the customer in the trouser pocket, then the rules, and lastly the answer to the question of who is actually buying when a program does it (Figure 1). Every technical term is set in bold at the point where it is explained, and is only used after that. A glossary lists all terms alphabetically.

Contents: 1 Catalogue and post · 2 Videotex, Minitel, Btx · 3 The web and the first purchase · 4 Rules, trust, bubble · 5 The platform · 6 The smartphone · 7 Pandemic and rules · 8 Agents and tariffs · Chain · Timeline · Further reading · Sources

1. The catalogue and the post: trade over distance (1861–1979)

The problem. Trade means that a buyer and a seller come together, the goods are examined, the price is named, the money is handed over and the goods are taken away. For thousands of years this happened in one place, at the market, in the shop, at the fair, and whoever lived far from the nearest market bought little and paid dearly. Trade over distance needs four things that the market supplies for free: a medium in which the buyer sees what is on offer; a way for the order to reach the seller; a way for the money to flow without the two standing face to face; and a way for the goods to come back. All four became cheap at once in the nineteenth century: printing, the railway, the post and the postal and cash-on-delivery services. Mail order, selling from a printed catalogue with delivery by post, was the first form of trade in which buyer and seller never saw each other, and it is the ancestor of everything that follows.

Pryce-Jones. A draper from Newtown in Wales is regarded as the inventor. In 1861 Pryce Pryce-Jones began selling flannel and blankets from his Royal Welsh Warehouse to customers he never saw: he sent them sample cards and price lists, took orders by letter and handed the goods to the railway post, which had recently begun running into the valleys. By 1880 he had more than a hundred thousand customers, the Queen among them, and he sold his Euklisia rug, a forerunner of the sleeping bag, as far as Russia and Australia. In 1887 he was knighted.⁵ His invention was not the catalogue, merchants had been sending out price lists for a long time, but the business that consisted only of catalogue, post and warehouse and no longer needed a shop.

Ward and Sears. In the United States the distances were greater and the need more pressing. The farmers of the Midwest bought from general stores that dictated prices to them, because there was no competition within a hundred miles. Aaron Montgomery Ward, a travelling salesman, rented a storeroom in Chicago in 1872 and printed a single sheet with 163 articles and prices, which he sent to the farmers’ organisation, the Grange, with the promise that goods would be taken back if they did not please. Out of it grew a catalogue of hundreds of pages that millions of households received every year. Richard Sears, a station agent in Minnesota, sold watches by post from 1886, printed his first catalogue in 1888, containing only watches and jewellery, and in 1894 a general catalogue of 322 pages; Sears, Roebuck and Co. became the country’s largest retailer. The catalogue, “the wish book”, lay in every farmhouse next to the Bible, and it brought people in the countryside the same goods at the same price as city dwellers. Both companies later grew into department-store chains and discontinued their big catalogues, Ward in 1985 after 113 years, Sears in 1993, shortly before the network arrived that reinvented their business.^{6, 7}

Germany. In Germany, Mey & Edlich in Leipzig, founded in 1870, sent out illustrated catalogues for collars, underwear and clothing from 1886, and August Stukenbrok sold bicycles by catalogue across the whole Reich from Einbeck from 1888. Gustav Schickedanz founded the mail-order house Quelle in Fürth in October 1927, which after the war became the largest in Europe. The second founding era came after 1945, when the shops were destroyed and people scattered: Werner Otto founded a mail-order business in Hamburg on 17 August 1949 and had his first catalogue printed in 1950, 300 copies, 14 pages, bound by hand, with pasted-in photographs of 28 pairs of shoes; Josef Neckermann resumed mail order in Frankfurt in March 1950. Otto was among the first to allow his customers to buy on invoice, that is to examine the goods before paying for them; this trust, advanced by the merchant, has remained the German peculiarity of distance selling to this day.^{8, 9, 10} In the 1970s every household received the thick catalogues of Quelle, Otto and Neckermann twice a year, more than a thousand pages, and ordered by postcard or, increasingly, by telephone.

The money. Paying over distance long remained cumbersome: payment in advance by postal order, cash on delivery, where the postman collects the money at the door, invoice with a payment term. What mail order needed, a means of payment that the buyer could present to the seller without knowing him, arose in a different business. Frank McNamara had forgotten his wallet in a New York restaurant in 1949 and founded the Diners Club in February 1950: a cardboard card that 27 restaurants and hotels in the city accepted, settled at the end of the month. In September 1958 the Bank of America sent 60,000 unsolicited credit cards to the households of Fresno, California, the “Fresno Drop”, and lost a lot of money through fraud and defaults, but within ten months more than a million BankAmericards were in circulation; in 1976 it became Visa. The credit card, a number for which a bank vouches on the holder’s behalf and which can be read out over the telephone, was the means of payment that made ordering from a distance possible without prepayment and without trust between the parties.¹¹

The number on the goods. A catalogue of a thousand pages needs a warehouse in which every article can be found, and bookkeeping that knows what is still there. The answer was to give every item a number that a machine could read. On 26 June 1974 at 8:01 a.m. the cashier Sharon Buchanan at the Marsh supermarket in Troy, Ohio, passed a pack of Juicy Fruit chewing gum over a scanner that read the stripes on the pack and displayed 67 cents; it was the first sale with the barcode, the Universal Product Code that the grocery industry had agreed on the year before. By the end of the 1970s only one per cent of stores had a scanner; twenty years later there were no goods left without a barcode.¹² At the same time companies began to exchange their orders among themselves without paper. Ed Guilbert had learned during the Berlin airlift of 1948 that the message about a cargo has to arrive before the cargo, and in a standard format; in 1968 railways, hauliers and airlines founded the Transportation Data Coordinating Committee, which published the first common message formats in 1975, and in 1979 it became the American standards committee ANSI X12. Electronic Data Interchange, EDI, the exchange of order, delivery note and invoice as standardised records between the computers of two companies, was the first electronic commerce, long before the web, except that here no people bought, but corporations.¹³

The screen. In the 1980s the catalogue got a competitor that could show moving pictures. In 1982 a local cable channel in Florida began praising goods in front of the camera and taking orders by telephone; in 1985 it went national as the Home Shopping Network, and on 24 November 1986 QVC went on air, soon selling round the clock. Teleshopping, selling from the television programme with ordering by telephone, came to Germany in 1995 with H.O.T., a channel owned by Pro Sieben and Quelle, and in 1996 with QVC.¹⁴ It was mail order with a better catalogue, but it had the same bottleneck as the catalogue.

The bottleneck. By 1979 a household in any industrial country could choose from a thousand pages, pay by card and receive the goods within a week. But the medium was paper or a programme that showed everyone the same thing, and the order was a letter or a call that a person on the other end took down, typed in and passed to the warehouse. The catalogue appeared twice a year and could neither change the price nor say whether the goods were still in stock. What was missing was a medium that answers the buyer: a screen that shows what is available right now, and a wire over which the order goes into the seller’s computer without passing through human hands.

2. Screens on the telephone network: videotex, Minitel and Btx (1979–1994)

The television as a terminal. The idea of bringing the catalogue to the screen came from the telephone companies, which wanted to make better use of their lines. In the 1970s the British Post Office developed Prestel: a television with an add-on box and a modem dialled a central computer over the telephone line, which sent pages of text and crude graphics that the user selected with numbers on a keypad; the service went into operation in 1979. Videotex, the generic name, was a closed system: one operator, one network, one display, and whoever wanted to offer something rented pages from the operator. Michael Aldrich, an English businessman, connected a television to a retailer’s computer over the telephone line in 1979 and demonstrated that one could order through it; he called it teleshopping and in the following years sold the systems to companies such as Ford and Nissan for their dealers.^{15, 16} The first purchase by a consumer was made in May 1984 by a 72-year-old woman in Gateshead in the north-east of England: Jane Snowball, who had broken her hip, used a remote control to choose eggs, margarine and cornflakes from a list on her television; the order went over the telephone line to the local Tesco store, where it was packed and driven out, and she paid in cash at the door. The council’s trial was meant as help for the old and the sick, not as a business.¹⁷

Minitel. The only videotex system that covered a whole country was the French one. The state telephone company tried it out in Saint-Malo from 1980 and introduced it in 1982, with a trick: it distributed the small terminals with screen and keyboard free of charge to telephone customers, as a replacement for the printed telephone directory. Whoever had a Minitel could reach thousands of services via the number 3615 and a code, timetables, account balances, classified ads, games, and could order from the mail-order houses: 3615 La Redoute, 3615 3 Suisses. Billing ran through the telephone bill, by the minute, and the provider received a share without the customer having to give it an account number or a card. At its peak, in the early 1990s, 6.5 million terminals stood in households, a fifth of all telephone lines, and 20 million people used 25,000 services.¹⁸ For twelve years Minitel was the largest network in the world in which consumers bought electronically, and it was at the same time the reason that France came late to the Internet: what already existed did not have to be reinvented. The service was switched off on 30 June 2012, when 800,000 terminals were still connected.

Bildschirmtext. The German Bundespost announced its videotex in 1977 and launched it on 1 September 1983 at the radio exhibition in Berlin, when the postal minister Christian Schwarz-Schilling pressed a button. Bildschirmtext, Btx, needed a television with a decoder and a modem that the post office rented out, and it cost a basic fee, telephone charges and often a fee per page. Banks offered transfers, Lufthansa flights, Otto and Quelle their catalogues, and the post office counted on millions of subscribers. In 1992 there were 320,000. The devices were expensive, the pages slow, the display crude, and whoever ordered got the same goods as from the catalogue, only more laboriously. Only when the service merged into T-Online in 1995 and brought the Internet with it did it pass the million; Btx itself was switched off at the end of 2001.¹⁹ That Btx failed and Minitel succeeded was not a matter of technology, which was the same, but of the decision whether the operator gives the terminals away or sells them, and of whether there is a service among them that everyone needs.

The online services. In the United States private providers took the role that the post office played in Europe, and their customers sat not in front of the television but in front of the home computer. CompuServe, a computing centre from Ohio that rented out its machines to hobbyists in the evenings, opened the Electronic Mall in April 1984, in which around a hundred merchants offered their goods on text pages, Sears and American Express among them, ordered by keyboard, paid with the credit card whose number the customer typed in. IBM, Sears and CBS founded a joint venture in 1984 that launched in 1988 under the name Prodigy and was available across the country by 1990: an online service with graphics, news, weather, mail and a department store, financed by advertising and a flat monthly fee, with Sears as the landlord of the store. At its best Prodigy had over two million users, but it lost money, and America Online, so named in 1991, overtook it with simpler software and floppy disks in the post.^{20, 21} All three were closed worlds: whoever was on CompuServe did not see what Prodigy offered, and a merchant had to join each service separately, on its terms.

EDI goes global. In the same years trade between companies got its world standard. The Europeans had their own EDI formats, the Americans X12, and the United Nations Economic Commission for Europe in Geneva brought them together: in 1987 the draft was ready, in 1988 it was published as ISO 9735, UN/EDIFACT, the rules by which orders between a supermarket and its supplier, customs declarations and freight documents are exchanged as data records to this day.²² Electronic trade between companies was everyday business by the end of the 1980s; it ran over expensive leased lines and value-added networks operated by service providers, and it was invisible to consumers.

Try it yourself. What Bildschirmtext felt like can still be seen: the Museum of Communication in Nuremberg shows Btx pages and devices, and on the web there are reconstructions in which you dial a provider’s page with *123#, 40 characters wide, 24 lines high, in six colours. Whoever calls up a page understands why the business did not take off: every step took seconds and cost money, and there was no picture of the goods.

The bottleneck. By 1994 every industrial country had a network in which one could order from a screen, and none reached beyond the country or the service. Minitel did not talk to Btx, Prodigy not to CompuServe, and whoever wanted to sell had to rent pages from each operator, in its format and at its fees. The merchants’ computers, which had mastered EDI for years, had no access to the consumer, and the consumer none to them. What was missing was an open network that belongs to no one, in which every provider has an address and every customer reaches every provider with the same program.

3. The web and the first purchase (1991–1997)

The open network. The network that was missing had existed since 1991, only it had no shop yet. Tim Berners-Lee had built the World Wide Web at CERN, pages with links that anyone could offer on any computer and anyone could read with the same program, and CERN released the technology to everyone on 30 April 1993; the browser Mosaic brought pictures onto the pages in the same year, and its developers became Netscape in 1994 (the history of the Internet tells these years). For trade the web meant three things at once: every merchant could put his catalogue on the net without asking an operator; every customer reached him with the same browser, from any computer with Internet access; and the page could come from a database, so it could show today’s price and stock instead of those of the print date. What was missing was the way back, from the customer to the merchant, and in such a way that no one could listen in.²³

The first purchase. On 11 August 1994 Dan Kohn, 21 years old, sold a CD over his site NetMarket in Nashua, New Hampshire, “Ten Summoner’s Tales” by Sting, for 12.48 dollars plus shipping to a friend in Philadelphia, who paid with his credit card; the card number was encrypted with the program PGP, which Kohn had built into a version of the browser Mosaic. The New York Times reported the next day under the headline “Attention Shoppers: Internet Is Open”, and Kohn told the paper that even if the NSA had been listening in, it could not have read the number. There were sales on the net before that day, but this was the first in which the payment went encrypted over the open network, and that is why it counts as the first online purchase.²⁴ Eleven days later Pizza Hut announced PizzaNet: whoever lived in Santa Cruz, California, could put together a pizza in the browser, the order went to a computer in Wichita and from there to the branch, which called back to be sure; the first pizza had pepperoni, mushrooms and extra cheese.²⁵

The shopping cart and the padlock. Two inventions from 1994 turned the demonstration into a business, and both come from Netscape. The web forgets everything between two pages; a server does not know that the customer now looking at the third page is the same one who selected something on the first. Lou Montulli, 24, built the cookie in the summer of 1994 for a store that Netscape was developing for MCI: a small note that the server deposits in the browser and that the browser sends along with every further request, so that the server recognises the customer again. With it came the shopping cart, the list of selected goods that persists across pages until the customer goes to the checkout; the browser Netscape 0.9 shipped with cookies in October 1994.²⁶ The second invention was the padlock. In 1994 Netscape developed SSL, Secure Sockets Layer, a procedure in which browser and server negotiate a key before the transfer and encrypt everything they send each other, and in which the server proves with a certificate that it is who it claims to be. The first version was flawed and was never released, the second came in 1995 with the Navigator, SSL 3.0 followed in 1996, and since 1999 the procedure has been called TLS; the padlock in the address bar that indicates it is to this day the sign to which buyers entrust their card number.²⁷ With cookie and SSL the online shop had its form: catalogue pages from a database, a shopping cart, a checkout in which address and card number go encrypted to the merchant, and a confirmation by email (Figure 2).

Amazon. Jeff Bezos, 30, a vice president at a New York hedge fund, had read in 1994 that the net was growing by 2,300 per cent a year, drawn up a list of twenty product categories that could be sold by post, and chosen books: there were millions of titles, more than any store could hold, and two wholesalers that carried them all. He founded the company in July 1994 in a garage in Bellevue near Seattle, first under the name Cadabra, then Amazon, after the largest river. The first order from a stranger came on 3 April 1995, while the site was still being tested: John Wainwright, a programmer in California, bought “Fluid Concepts and Creative Analogies” by Douglas Hofstadter. On 16 July 1995 amazon.com opened to everyone, as “Earth’s Biggest Bookstore” with a million titles that it did not have in stock but ordered from the wholesaler as soon as a customer had ordered them. In the first month orders came from all fifty states and 45 countries; in 1996 the company had sales of 15.7 million dollars, in 1997 147.8 million.²⁸ From the start Amazon let customers write reviews, bad ones too, which publishers thought was crazy, and it remembered what each customer had bought in order to suggest the next thing. The store was not just a catalogue on the net; it knew who was standing in front of it.

eBay. On the Labor Day weekend at the beginning of September 1995, Pierre Omidyar, a programmer in San Jose, put software on his personal website with which anyone could auction anything to anyone, AuctionWeb. The first object to change hands there was a broken laser pointer, which a collector from Canada won for 14.83 dollars; Omidyar asked him whether he knew it was broken, and the buyer did. The online marketplace, a site on which not the operator sells, but outside sellers and buyers find each other and the operator takes a fee, was thereby invented, and it grew without a warehouse, without goods and almost without employees: in February 1996 the fees were enough to replace Omidyar’s salary, and he introduced the forum in which buyers and sellers rate each other, because he saw no other way of getting strangers to trust one another. In 1997 the company was called eBay.²⁹

Dell and the business customers. That the net was also fit for large sums was shown by Michael Dell, who had sold computers built to order by telephone since 1984. From July 1996 customers could configure their computer themselves on dell.com, see the price and order, and six months later the site was taking a million dollars a day.³⁰ Cisco, the maker of routers, soon sold most of its equipment over the net to business customers who had previously sent their orders by fax. Trade between companies, which had begun with EDI, got a cheaper form in the browser, and the distinction between B2C, selling to consumers, and B2B, selling between businesses, became the vocabulary of the industry.

Germany goes online. Otto had pressed its catalogue onto CD-ROM in 1994 and opened otto.de in 1995, at first as a supplement to the paper; in Regensburg the ABC Bücherdienst had been selling books on the net under telebuch.de since 1995. The software with which a merchant puts a store on the net without programmers of his own came from Jena: Stephan Schambach had founded NetConsult there in 1992 and from 1995 built a shop system, catalogue, shopping cart, checkout and administration as a finished program that the merchant fills with his goods; the company was called Intershop from 1997, moved to San Francisco and went public on 16 July 1998 as one of the first on the Neuer Markt.^{31, 32}

Try it yourself. The padlock of 1995 is still there. Whoever clicks on the padlock next to the address in a browser sees the certificate with which the server proves its identity, who issued it and how long it is valid; that is the same chain Netscape invented, except that certificates are free today. In the browser’s developer tools (F12, tab “Application” or “Storage”) are the cookies a store has set, with name, value and expiry date; the shopping cart is usually just a number among them. And the Internet Archive shows at web.archive.org what amazon.com looked like in 1999: a text page with a search box, a list and a button “Add to Shopping Cart”.

The bottleneck. By 1997 there was the open network, the browser, the shopping cart and the padlock, and there were stores with sales in the millions. But most people had no Internet access, and those who did hesitated to give their card number to a site that a student could run from a garage. No one knew whether a contract of sale by mouse click was valid, whether one could return the goods, who was liable if the card was misused, and which state got the tax. The net had bridged the distance; what was missing was trust, and trust arises from rules and from experience, and both take time.

4. Rules, trust and the bubble (1997–2002)

The laws. Trust came first from Brussels. On 20 May 1997 the European Union had adopted the Distance Selling Directive, which applies to every purchase in which buyer and seller do not face each other, whether by catalogue, telephone or net: the seller must say before the purchase who he is and what the goods cost, and the buyer may send the goods back without giving reasons. Germany transposed it with the Distance Selling Act, which came into force on 30 June 2000 and introduced a right of withdrawal of two weeks; since 2014 it has been fourteen days everywhere in the Union. The buyer on the net thereby got what the buyer in the shop never had: the right to look at the goods and give them back, and this right is the reason why in Europe a part of every order comes back. The E-Commerce Directive of 8 June 2000 settled the rest: that a contract by mouse click is valid, that the provider needs a legal notice, that an order must be confirmed without delay, and that the operator of a platform is liable for the content of outside providers only once he knows of a violation. This liability rule, intended as protection for young net companies, became the foundation of the marketplaces.^{33, 34} In the United States it was the other way round, about what the state should not do: the Internet Tax Freedom Act of 21 October 1998 prohibited new taxes on Internet access, and a 1992 ruling of the Supreme Court on mail order meant that a merchant had to collect a state’s sales tax only if he had a warehouse or an office there. Whoever shipped from Seattle to Texas shipped tax-free, an advantage of six to nine per cent over every local shop that lasted twenty years.³⁵

The trust of strangers. The second source of trust was the merchants themselves. eBay’s feedback forum turned strangers into people with a reputation, and whoever had a thousand positive ratings got more for the same goods; Amazon let customers write reviews and showed them next to the price. In Cologne, Trusted Shops was founded in 1999, a trust mark that an auditor awards when a store handles legal notice, data protection, withdrawal and payment correctly, combined with an insurance that refunds the customer’s money if the goods do not arrive. The card organisations treated purchases on the net like purchases by telephone: whoever disputes having ordered gets the money back, and the merchant bears the loss unless he can prove delivery. The card companies’ grand plan, a procedure called SET, in which every customer would have had to install a certificate on his computer, failed on precisely that effort, and the attempts to create a money of the net’s own failed on the same problem: David Chaum’s DigiCash, which had offered anonymous electronic cash since 1994, filed for bankruptcy in November 1998 because no bank and no merchant wanted to carry it unless the others already did. The winner was the credit card that everyone already had, and in Germany the invoice and the direct debit that everyone already knew.^{36, 37}

PayPal. The means of payment that the net did produce came by way of eBay. Confinity, founded in Palo Alto in 1998, offered in 1999 to send money by email: the sender gave his card or account to PayPal once, the recipient got a message and a credit, and no one had to show the other his bank details. The sellers on eBay, private individuals who could not accept credit cards, took it up, PayPal paid every new customer ten dollars for signing up, and in 2000 the company merged with Elon Musk’s X.com. eBay, which was getting nowhere with a service of its own, bought PayPal in October 2002 for 1.5 billion dollars.³⁸ The payment service, an intermediary that holds the buyer’s card or account data and passes on to the merchant only the money and a name, was the solution to the problem of 1997: the customer no longer had to give his number to every store, but to one that he trusted.

Amazon on the stock market. On 15 May 1997 Amazon went public, at 18 dollars a share, valued at 438 million dollars, after two years of business and without ever having made a profit. In his first letter to shareholders Bezos wrote that it was “Day 1” for the Internet and for Amazon, that the company would put market leadership above short-term profit and judge investments by whether they raised long-term value. His slogan to the staff was “Get big fast”: whoever is big first in a new market gets the customers, the suppliers and the data, and whoever makes a profit too early leaves the market to the one who keeps investing. In April 1998 Amazon bought the ABC Bücherdienst in Regensburg and Bookpages in London, opened amazon.de and amazon.co.uk on 15 October 1998, built its first German warehouse in Bad Hersfeld in September 1999 and added music, video, toys and electronics to the range; sales rose from 148 million dollars in 1997 to 1.64 billion in 1999, the loss to 720 million.^{39, 40} On 28 September 1999 Amazon was granted a patent on 1-Click, purchasing with a single click for customers whose address and card are already stored, and 23 days later sued Barnes & Noble, whose store did the same; the patent, never granted in Europe, expired in 2017 and until then remained the example of what counted as an invention in the United States.⁴¹

The imitators. In Germany the Neuer Markt arose in 1998, and with it a generation that copied American stores before the Americans arrived. The brothers Marc, Oliver and Alexander Samwer founded Alando in Berlin at the beginning of 1999, an auction site modelled on eBay, and sold it after a hundred days to eBay, which was not yet in Germany, for a sum that neither side ever confirmed, around 50 million dollars according to press reports; Alando became eBay.de.⁴² The Samwers repeated the pattern for a decade and built a factory for it with Rocket Internet. On 28 June 1999 Jack Ma founded Alibaba in his flat in Hangzhou with seventeen friends, at first a marketplace on which Chinese factories found buyers abroad, and in October received five million dollars from a group around Goldman Sachs and in January 2000 twenty million from SoftBank.⁴³ In Latin America, MercadoLibre started in Buenos Aires in August 1999; in Japan, Hiroshi Mikitani had opened Rakuten Ichiba in May 1997, a marketplace with thirteen merchants that became the largest in the country.⁴⁴

The bubble. What followed is told by the history of venture capital from the investors’ point of view; what counts here is what the stores learned from it. Boo.com, a fashion store from London that wanted to launch in eighteen countries at once, with three-dimensional views of every jacket that took minutes to load on the modems of the time, went online in November 1999, had spent 135 million dollars by then and was wound up on 18 May 2000. Pets.com went public in February 2000 and closed in November, 268 days later; the sock-puppet dog it had advertised with outlived the company. Webvan, the grocer that built warehouses for a billion dollars in ten cities before it knew whether the customers would come, filed for bankruptcy in July 2001, eToys in the spring of 2001. On 10 March 2000 the NASDAQ had reached its peak at 5,048 points; two and a half years later it stood 78 per cent lower.^{45, 46} The lesson was not that the net was no good, but that a store on the net has to do the same sums as a store in town: what does it cost to win a customer, what does one earn from him, and what does it cost to get the goods to him? A ten-dollar bag of dog food that costs seven dollars to ship is not a business, no matter how many people order it.

Amazon survives. In 2000 Amazon lost 1.41 billion dollars on sales of 2.76 billion, had borrowed another 690 million euros in Europe in February 2000 and was considered by analysts to be the next victim. Bezos closed warehouses, laid off 1,300 people and promised a positive operating result for the end of 2001. In the fourth quarter of 2001 the company reported the first profit in its history, five million dollars, one cent a share, on sales of over a billion in the quarter.⁴⁷ The difference from Webvan was the warehouse: Amazon had built its own distribution centres since 1997, in Delaware, Nevada, Kansas, Kentucky, and learned to store, pick and pack millions of different articles in such a way that a parcel cost a dollar less than the year before (Figure 3). The store on the net, it now turned out, was not a program but a logistics company with a website in front of it.

Try it yourself. The prospectuses of the bubble lie open in EDGAR, the database of the American securities regulator (sec.gov/edgar). Whoever searches there for “Pets.com” and opens the Form S-1 of February 2000 finds under “Risk Factors”, on twenty pages, what happened nine months later: that shipping costs exceed the value of the goods and that the company has never made a profit. Amazon’s letter of 1997 is still attached to every annual report of the company; it is four pages long.

The bottleneck. By 2002 it was settled that a purchase on the net is valid, that one can return the goods, whom one gives the card number to, and that a store on the net needs a warehouse. But the store that could do all of that was one, and it was expensive: Amazon had spent billions on warehouses and software that a bookseller from Regensburg did not have. For the many small merchants left after the bubble, the question was how to find customers without becoming Amazon themselves, and for Amazon the question was what to do with warehouses and computers that were bigger than its own business.

5. The platform: marketplace, search and logistics as a service (2000–2010)

The marketplace inside the store. Amazon’s answer to the question of what to do with its warehouses and computers was to rent them out. In November 2000 Amazon opened Marketplace: outside merchants were allowed to offer their goods on the same pages as Amazon itself, next to the new book the used one from a second-hand dealer, and Amazon took a commission, a share of the selling price, usually between eight and fifteen per cent. The booksellers protested that Amazon was cannibalising its own sales, and Bezos replied that if someone could make customers a better offer, he would rather they did it on Amazon’s page than elsewhere. The platform model had thereby arrived in retail: the operator not only sells itself, but rents customers, software and trust to the third-party seller, a merchant who sells on its site, and earns from each of his sales without ever owning the goods. In 2015 more than half of the items sold on Amazon came from third-party sellers, in 2025 it was 61 per cent.⁴⁸

Prime and fulfilment. Two decisions of 2005 and 2006 turned the marketplace into a machine. In February 2005 Amazon introduced Prime: whoever paid 79 dollars a year got every order in two days without shipping costs, as often as they liked. The programme cost more in the first year than it brought in, but members ordered twice as often as before and stopped comparing prices; the membership, a flat fee that makes shipping costs invisible and binds the customer to one store, became the model for the whole industry. On 19 September 2006 Fulfillment by Amazon started: a third-party seller sends his goods to Amazon’s warehouse, and Amazon stores, packs, ships, takes returns and answers the questions, for a fee; the merchant’s goods thereby become eligible for Prime, and the merchant no longer needs a warehouse. Fulfilment, the handling of storage, packing, shipping and returns as a service, was the step with which Amazon turned from a store into infrastructure.^{49, 50} In the same year Amazon rented out its computers too: in March 2006 came the storage service S3, in August the computing machines EC2, and out of the store’s by-product grew Amazon Web Services, the cloud that has since earned the larger part of the company’s profit and that plays its own role in the history of software development and the history of open source.⁵¹

Search as the shop window. Whoever did not sell on Amazon had to find customers differently, and the way led through Google. In October 2000 AdWords started: merchants bid money to appear next to the search results for a word, and paid only when someone clicked. Search advertising, buying visibility per click in an auction, replaced for the store on the net the location on the high street: whoever wanted to be at the top for “running shoes” paid for it, and the more competitors bid, the more. The cost of winning a customer became measurable, and it rose year after year. Alongside arose price comparison: idealo, founded in Berlin in 2000, and Google’s Froogle of 2002 listed the same product at all merchants by price, and the buyer who in town would have walked to three shops now saw thirty at a glance. For the merchants this meant that the price became transparent and the margin thin, and that the cheapest won, until he went bust.⁵² In 2004 Chris Anderson described in Wired what search meant for the range: a store not limited by shelves sells not only the hundred bestsellers but millions of articles that are each bought only a few times, and this niche, the long tail, together makes up a large part of sales; Amazon sold books that no store had ever carried, and made money on them.⁵³

China: Taobao and Alipay. The market that ran away from the American one was the Chinese, and it solved the problem of trust differently. eBay had bought the Chinese market leader EachNet in 2003 and charged fees; in May 2003 Jack Ma opened Taobao, a marketplace with no fees for sellers, financed by advertising, and let buyers and sellers haggle with each other over a built-in chat, as at a market. The real problem was that in a country without credit cards and without a rating history nobody wanted to pay in advance. Alipay, founded in 2004, introduced the escrow payment: the buyer pays Alipay, the seller ships, the buyer confirms receipt, and only then does the seller get the money. With that the risk was gone on both sides, and Taobao overtook eBay in China within two years; eBay gave up the market in 2006. In April 2008 Taobao opened an area for brand retailers, Tmall, and on 11 November 2009, “Singles’ Day”, which Chinese students celebrated as a counter-festival to Valentine’s Day, Tmall held its first discount day: 27 merchants took part and sold 52 million yuan. Six years later it was 91 billion yuan in a day, more than Black Friday and Cyber Monday in the United States combined.^{54, 55}

The store for rent. For the small merchant who wanted to be neither on Amazon nor on Taobao, the store of his own became cheap. Tobias Lütke, a programmer from Koblenz who had moved to Ottawa, wanted to sell snowboards on the net in 2004, found no usable software and wrote it himself, in Ruby on Rails; in 2006 he turned the program into Shopify, a shop system that one does not buy and install but rents monthly and that runs on Shopify’s computers, with payment, templates and later warehousing and shipping as add-ons. Magento, released in 2008 as an open-source shop system, and WooCommerce, in 2011 as an extension of WordPress, covered the rest: whoever wanted to open a store in 2010 needed an afternoon and a credit card.⁵⁶ In Germany, Robert Gentz and David Schneider copied in Berlin in 2008, with money from Rocket Internet, the American shoe retailer Zappos, which had shown with free shipping and free returns that one can sell shoes without trying them on if one makes sending them back easy for customers; with the slogan “Schrei vor Glück” (scream with joy) and a return rate of about half of sales, Zalando became Europe’s largest fashion retailer.⁵⁷

Trade between companies. What this text, because of its subject, only touches on ran at full scale in the same years: trade between companies, which by volume exceeds trade with consumers. Ariba, founded in California in 1996, sold software with which corporations handled their purchasing over the net, requisition, approval, order, invoice; the carmakers General Motors, Ford and DaimlerChrysler announced the joint purchasing marketplace Covisint in February 2000, which fell apart again after a few years because suppliers did not want to bid against each other on prices on a platform owned by their customers. Alibaba had begun in 1999 as a marketplace between factories and traders, Mercateo in Germany in 2000 as a procurement platform for business customers, and Amazon opened Amazon Business in April 2015 in the United States and in December 2016 in Germany, the same marketplace with net prices, purchase on invoice and approval workflows for companies. In the background EDI remained: the orders of the supermarket chains to their suppliers still run as EDIFACT messages today, not through a shopping cart.⁵⁸

The calendar. Trade on the net got its own holidays. On 28 November 2005 the American retailers’ association named the Monday after Thanksgiving Cyber Monday, because on that day people ordered at the office, where the line was faster, what they had seen in the stores at the weekend; the name stuck, and in 2025 14.25 billion dollars were spent online in the United States on that day, more than on any other. Amazon invented Prime Day in 2015, Alibaba had Singles’ Day, and Black Friday migrated from the car parks of the shopping malls into the apps and to Europe, where there is no Thanksgiving.⁵⁹

Try it yourself. What a third-party seller keeps on a marketplace can be calculated. An article for 30 euros costs on Amazon, depending on the category, around 15 per cent commission, that is 4.50 euros, plus with fulfilment by Amazon a shipping fee of a few euros per item and a storage fee per cubic metre and month, plus advertising if the article is to be found. Whoever works that through in a spreadsheet for an article with a purchase price of 12 euros sees why many merchants earn nothing despite their sales and why the platform earns from each of them. The fee tables are public on the platforms’ seller pages.

The bottleneck. By 2010 any merchant could open a store in an afternoon, reach millions of customers on a marketplace and rent a warehouse; every customer found every price. But the customer sat at a desk while doing so, in front of a computer with a keyboard, and bought when he had time, mostly in the evening or at the office. The store on the net was a place one went to, like the shop in town, only that it was at home. What was missing was the customer himself: at the bus stop, in someone else’s shop, on the sofa, with the store in his hand.

6. The store in your hand: smartphone, paying, delivery (2007–2016)

The telephone. On 29 June 2007 the iPhone went on sale, and on 10 July 2008 the App Store opened, through which programs came onto the phone that started with a tap of the finger. For trade this changed the place of buying: the store was no longer on the desk but in the trouser pocket, opened on the bus, in the waiting room, in front of the television and in a competitor’s shop, where one looked up the price of the article on the shelf. Amazon brought its app in 2008, eBay in the same year, and within a few years most visits to merchants’ sites came from the phone. Mobile commerce, shopping via the mobile phone, was not a new kind of trade, but it demanded new stores: pages for a screen of ten centimetres, checkouts without a keyboard, login by fingerprint. Whoever had a website in 2007 needed an app in 2012, and whoever had an app needed a reason why anyone kept it.⁶⁰

Paying with the phone. The biggest change came where there were the fewest credit cards. In July 2011 Alipay introduced payment by barcode: the phone shows a code that the till reads, or the phone’s camera reads the code stuck to the counter, and the money goes from the Alipay balance to the merchant, without a card, without a reader, without a bank in between. In August 2013 Tencent built WeChat Pay into its messaging service, which almost every Chinese person used, and for the New Year festival in January 2014 distributed digital red envelopes with lucky money that could only be opened by creating an account; Jack Ma called it an attack like Pearl Harbor. The QR code, a square barcode that any phone camera can read, became within three years the means of payment for everything in China, from the street vendor to the rent, and made the boundary between the store on the net and the store on the street porous: both took the same money from the same app. In October 2014 Apple introduced Apple Pay, which stores the card in the phone and gives the merchant a one-time number instead of the card number, and in Europe Klarna, founded in Stockholm in 2005, grew big with the invoice purchase that Otto had invented in 1950: the customer orders, examines, keeps or sends back and only then pays, and Klarna bears the risk and the dunning.^{61, 62, 63}

Goods without a parcel. Part of trade on the net needs neither warehouse nor parcel courier, and it follows other rules. On 28 April 2003 Apple opened the iTunes Music Store, in which a song cost 99 cents and was there at once; Valve launched Steam for games in the same year, Netflix began streaming films in 2007 instead of mailing DVDs, Spotify came in 2008, and the App Store took thirty per cent of every program and every purchase inside a program. Digital goods, music, films, games, software, subscriptions, have no last mile and cost nothing to copy; their bottleneck is not delivery but the right to sell them, and the share kept by the platform that stands between provider and phone. The German association counts these sales separately, 14.4 billion euros in 2025 alongside the 83 billion for goods, and this text stays with goods, because the chain of distance, time, money and trust is their history; that of digital goods is the history of copyright and platforms, and it would be an article of its own.⁶⁴

China overtakes. On 19 September 2014 Alibaba went public in New York and raised 25 billion dollars, more than any company before; on Singles’ Day 2015 Taobao and Tmall sold 91.2 billion yuan in one day, 14.3 billion dollars, most of it via the phone. JD.com, founded in Beijing in 1998 as an electronics retailer and on the net since 2004, built its own warehouses and its own couriers as the counter-model to Alibaba, delivering the same day in the big cities. Colin Huang founded Pinduoduo in 2015, which organised group purchases inside WeChat: whoever persuaded friends to buy the same article got it cheaper, and so trade on the net reached the small towns and villages where there was neither department store nor credit card.⁶⁵ In India two former Amazon employees had founded Flipkart in Bangalore in 2007, which delivered books against cash at the door because hardly anyone had a card; in Africa Jumia started in Lagos in 2012. Where there were no shops, trade on the net was not a replacement but the first equipment.

The robots in the warehouse. Behind the promise that everything will be there tomorrow stood an arms race in the warehouses. In March 2012 Amazon bought Kiva Systems for 775 million dollars, whose flat orange robots drive whole shelves to the packers instead of the packers walking to the shelves; by 2016 there were 45,000 of them in Amazon’s warehouses, by 2025 a million. Delivery became faster and the floor space larger: warehouses the size of dozens of football pitches arose at the motorway junctions, and Amazon began delivering parcels with its own drivers because the post and the parcel services could not keep up.⁶⁶ In Germany the post had already set up the first Packstations in 2001, lockers in which the parcel waits until the customer has time; Instacart in San Francisco from 2012 had freelancers shop in the supermarket and deliver, and Ocado, founded near London in 2000, ran grocery warehouses in which thousands of robots on a grid packed the crates. The last mile, the way from the last warehouse to the front door, was the most expensive part of the chain and became more so with every promise.

Zalando and the return. Zalando went public in Frankfurt on 1 October 2014, valued at more than five billion euros, as the largest German IPO of an online retailer. Its business rested on the right of withdrawal: customers ordered three sizes and sent two back, and the company accepted the returns free of charge because in tests customers bought less as soon as sending back cost money. The return, goods sent back that are checked, refurbished and resold or destroyed, costs the merchant a few euros per item for transport, inspection and refurbishment, and in fashion about half of sales comes back. The right of withdrawal, meant in 1997 to create trust, had created a habit that costs trade billions and the environment lorry miles.⁶⁷

The button. In Germany the purchase by click was given its prescribed form in 2012: since 1 August 2012 the button that triggers an order must say “zahlungspflichtig bestellen” (order with obligation to pay) or something equally unambiguous, and immediately above it must stand price, shipping costs and contract term; the button rule was the answer to subscription traps in which users had concluded a contract with a click on “Next”. The Union’s Consumer Rights Directive of 2011, in force in all member states since 13 June 2014, standardised the right of withdrawal at fourteen days and banned pre-ticked boxes for extra services.⁶⁸

Try it yourself. How much of trade now takes place on the net has been published by the American statistics office every quarter since 1999, and the series can be downloaded with one line: curl -o ecom.csv "https://fred.stlouisfed.org/graph/fredgraph.csv?id=ECOMPCTSA" delivers the share of e-commerce in retail sales per quarter, from 0.6 per cent at the end of 1999 to over 17 per cent in 2026, with the jump in the spring of 2020 that Figure 5 shows. Whoever loads the file into a spreadsheet sees that the curve before the pandemic was almost a straight line, about half a percentage point a year, and that after the jump it continues on that line as if the jump had never happened.

The bottleneck. By 2016 the store was everywhere, paying was a tap and delivery a matter of hours. But the chain that achieved this belonged to a few. Amazon sold more in the United States than any other retailer, was at the same time the marketplace, the warehouse, the parcel service and the advertising space of its competitors, and knew about their business what they themselves did not; Alibaba and Tencent shared China between them. Google decided which merchant was found, and charged for it. The question that arose was the one about rules: who gets the tax, who is liable for the goods that a stranger offers on the marketplace, and whether a platform may be referee and player at once.

7. Pandemic, platform power and rules (2017–2023)

The tax and the search. The rules came from the courts. On 27 June 2017 the European Commission fined Google 2.42 billion euros because the search engine displayed its own price comparison at the top and sorted those of competitors to the back; the European Court of Justice upheld the decision finally on 10 September 2024. On 21 June 2018 the Supreme Court of the United States ruled in South Dakota v. Wayfair that a state may demand sales tax even from merchants without a shop or warehouse in the state as soon as they sell there in significant amounts; the tax advantage of mail order, which Amazon had used for twenty years and no longer needed, was thereby over for everyone.^{69, 70} And on 22 January 2018 the first Amazon Go opened in Seattle, a store without a checkout in which cameras and scales record what the customer takes from the shelf and the app debits it on leaving; it was the attempt to carry the technology of online retail back into the city, and it remained an attempt.⁷¹

The pandemic. In March 2020 the shops closed, and trade on the net experienced in weeks what would otherwise have taken years. In the United States the share of e-commerce in retail jumped from the first to the second quarter of 2020 from 11.9 to 16.3 per cent, an increase like that of the seven previous years combined; in Germany sales of goods rose by 15 per cent in 2020 to 83.3 billion euros and by another 19 per cent in 2021 to 99.1 billion, and people who had never bought on the net, the old, country dwellers, grocery shoppers, now did.^{72, 73} Amazon enlarged its workforce by half a million people in 2020, Shopify doubled the volume of its stores because shops that had to close their doors needed a website overnight, and in Berlin in 2020 Gorillas and Flink started, bringing groceries from small warehouses in residential districts by bicycle in ten minutes, financed with billions of venture capital betting on the permanence of the state of emergency. Quick commerce, delivery within minutes from warehouses in the neighbourhood, was the Webvan of these years: Getir from Istanbul took over the struggling Gorillas at the end of 2022 and withdrew from Germany entirely in April 2024; Flink, in which Rewe had taken a stake, was left.⁷⁴

The correction. What the pandemic had brought, it partly took back. In 2022 German online sales of goods fell by nine per cent, in 2023 by another twelve to 79.7 billion euros, and in 2025, at 83.1 billion, they were back at roughly the level of 2020, just below it; the next section tells of that. In the United States the share fell from 16.3 to 14.2 per cent and then climbed on, as if the pandemic had been a fold in the curve (Figure 5). The merchants who in 2021 had bought warehouses, people and advertising for growth that did not come laid people off, and in 2022 Amazon closed or cancelled dozens of sites it had just opened or planned.⁷⁵ What remained is the share: in Germany around 13 to 15 per cent of retail has come from the net since 2021, in Britain a good quarter, in China a quarter, and the rest of the curve follows the old straight line again.

The rules for platforms. In 2022 the European Union passed two laws that imposed duties on the marketplaces that they had not had since the E-Commerce Directive of 2000. The Digital Markets Act applies to the gatekeeper, a company that stands between businesses and consumers and without which no merchant can manage; Amazon was designated in September 2023 and since then may not favour its own offers on its marketplace or use the data of its third-party sellers against them. The Digital Services Act has applied to all platforms since 17 February 2024 and requires the large ones, Amazon, AliExpress, Zalando, Temu, Shein, to check who sells on them, remove illegal goods, explain their recommendations and examine the risks of their service annually, on penalty of up to six per cent of worldwide turnover.^{76, 77} At the same time, on 1 July 2021, the Union closed a gap in taxation: until then consignments from third countries up to 22 euros were exempt from import VAT, and millions of parcels from China were worth 21 euros on the label; since then VAT is due from the first euro, and the marketplace that brokers the purchase owes it.⁷⁸

Temu, Shein, TikTok. The merchants at whom these rules were aimed came from China and sold differently from Amazon. Shein, founded in Nanjing in 2008, had been selling fashion since 2012 straight from the factories of Guangzhou to young women all over the world: thousands of new articles a day, in small runs, reordered when they sold, and at prices no European retailer could match, because every item came into the country individually as a postal consignment under the customs threshold. Temu, launched by Pinduoduo in the United States on 1 September 2022 and in Europe in 2023, did the same with everything, tools, toys, household goods, advertised with billions at Google, Meta and the Super Bowl, and within two years was the most downloaded shopping app in the world; in Germany about one in three people ordered there in 2025. In September 2023 TikTok opened its shop in the United States, in which merchants sell goods in videos and livestreams and the purchase happens in the app without the user leaving it; social commerce, selling in the stream of entertainment, had been everyday life in China for years and now came to the West.^{79, 80, 81} What the three had in common was that they left out the middleman and the warehouse in the destination country: the goods went from the manufacturer into an aeroplane and from the aeroplane to the customer, and what was missing in between, inspection, safety, tax, customs, went unnoticed by anyone until the number of parcels made it impossible to overlook.

Straight to the customer. Alongside the platforms, a counter-model arose in the 2010s that wanted to do without them. Warby Parker sold glasses from 2010 only through its own site, Dollar Shave Club from 2011 razor blades by subscription, made famous by an advertising video, and Unilever bought the company in 2016 for a reported one billion dollars. Direct-to-consumer, a brand selling to its customers without retailers and without a marketplace, followed a pattern: a Shopify store, advertising on Facebook and Instagram that could be costed to the cent, and a product not available in shops. The pattern held as long as advertising was cheap. When Apple in April 2021 forbade apps to track users across sites without permission, the ads lost their precision, the cost per customer won rose, and many of the brands went where they had never wanted to go, onto the shelves of Target and Walmart. Alongside grew the marketplaces of the niche that Amazon did not cover: Etsy, founded in Brooklyn in 2005, for handmade goods, Vinted, in Vilnius in 2008, for worn clothing, Discogs for records; trade from private person to private person, with which eBay had begun, was in 2025 the only part of the German parcel market that grew markedly.⁸²

The shops. What trade on the net cost the cities could be counted. At the end of 2015 there were 372,000 retail shops in Germany, at the end of 2025 around 301,500, and the retail association expected a further 4,900 closures for 2026; Quelle had gone under in 2009, Neckermann in 2012, the department stores of Karstadt and Kaufhof became one chain and that chain went insolvent three times. The number of parcels rose from 2.3 billion in 2010 to 4.37 billion in 2025, and on some streets four delivery vans stopped one behind the other in the morning.⁸³ Trade had not disappeared; it had changed its place, from the parade of shops to the warehouse by the motorway and the van in front of it.

Try it yourself. The figures of German retail are public: the retail association (einzelhandel.de) publishes the Online-Monitor every year with share, product groups and forecast, the e-commerce and mail-order association (bevh.org) the annual report with sales by product group and channel. Whoever lays the series from 2018 to 2025 side by side (Figure 6) sees the pandemic as a hump and the marketplaces as the part that keeps growing throughout: in 2025, at 46 billion euros, they made up more than half of online trade in goods.

The bottleneck. By 2023 trade on the net was everyday life, regulated, taxed and with duties for the platforms, and for the first time it had found a limit: the share was growing as slowly again as before the pandemic, and the costs of winning a customer, bringing him the goods and taking them back were rising faster than sales. At the same time twelve million parcels a day came from China into the Union, past customs, inspection and merchants, and no one was responsible for them. And a new program began to search, compare and order for the customers, and no one knew to whom it belonged while doing so.

8. Agents, tariffs and the question of who is buying (since 2024)

The customs threshold falls. The answer to the flood of parcels came in 2025 from both sides of the Atlantic. The European Commission counted 4.6 billion consignments under 150 euros entering the Union in 2024, twelve million a day, 91 per cent of them from China, twice as many as the year before; up to 150 euros the goods were duty-free, and spot checks found in a large part of the parcels goods that did not meet European safety rules.⁸⁴ The United States abolished the duty exemption for small consignments, which there stood at 800 dollars, on 2 May 2025 for China and Hong Kong and on 29 August 2025 for all countries; the postal services of several states temporarily stopped parcel shipments to the United States because they did not know how to collect the duty, and Temu and Shein lost a large part of their American customers within months and shifted their advertising to Europe.⁸⁵ The Union abolished its threshold as of 1 July 2026: since then every consignment under 150 euros carries a flat duty of three euros per type of goods, as a transition until 2028, when a new customs database is to record every consignment individually; according to figures from Dutch and Belgian customs, the number of cheap parcels halved in the first weeks.⁸⁶ Added to this were the penalties of the Digital Services Act: on 28 May 2026 the Commission fined Temu 200 million euros because the platform had not examined the risks of illegal goods on its pages, and in February 2026 it opened proceedings against Shein.⁸⁷ Trade over distance, which had begun in 1861 with the railway post, thereby got limits for the first time that lay not in the technology but in what a country is willing to let in.

The agent goes shopping. The search for the goods, which since 2000 had begun on Google and since 2010 more and more often directly on Amazon, got a third route in 2024: you ask a language model. In February 2024 Amazon built Rufus into its app, an assistant that answers questions about products and makes suggestions, and from April 2025 tested “Buy for Me”, in which Amazon orders on the customer’s behalf on another merchant’s site. Within two days in April 2025 the card organisations presented their schemes, Visa Intelligent Commerce and Mastercard Agent Pay, with which a program may pay in the name of a cardholder, within limits the holder sets. On 16 September 2025 Google published the Agent Payments Protocol, in which the customer gives his agent three signed mandates, for the intent, the cart and the payment, and on 11 January 2026 at the retail trade fair NRF the Universal Commerce Protocol, with Shopify, Etsy, Wayfair, Target and Walmart as co-authors. And on 29 September 2025 OpenAI presented Instant Checkout: whoever asked ChatGPT about a product could buy it there without seeing the merchant’s site, first at Etsy, then, so the announcement went, at a million Shopify merchants and, as Walmart announced in October, soon there too; the underlying Agentic Commerce Protocol, developed together with Stripe, was published as an open standard. An AI agent, a program consisting of a language model and tools that carries out multi-step tasks on a person’s behalf, searching, comparing, ordering, paying, was to become the new buyer, and agentic commerce was the industry’s word of the year.^{88, 89, 90} The history of vibe coding tells where these agents come from.

What came of it. The first year went differently from the announcements. Of the Shopify merchants who were to sell in ChatGPT, only a few dozen had gone live by the spring of 2026 according to trade reports, and in March 2026 OpenAI withdrew from its own checkout: ChatGPT was to find and compare products, but the purchase would again take place with the merchant, in the merchant’s app inside ChatGPT. The protocol remained, maintained by OpenAI and Stripe, with PayPal as a payment partner; Perplexity had offered purchase in the chat with PayPal since November 2025, and Visa and OpenAI announced in June 2026 that they would build Visa’s scheme into ChatGPT.^{91, 92} The merchants hesitated for a reason that has come up several times in the history of this text: whoever hands his checkout to someone else loses the customer, his data and the chance to sell him the next thing, and pays a commission for it. For the marketplace of 2000 they had done it because the marketplace brought customers; the agent of 2026 did not bring them yet.

Amazon against the agent. The dispute over who the agent belongs to was fought out in court by Amazon and Perplexity. Perplexity’s browser Comet could, logged in with the user’s Amazon account, search and order on amazon.com; Amazon demanded that the company stop, and on 4 November 2025 sued it in San Francisco for unauthorised access to its computers: the agent passed itself off as a human, ignored Amazon’s prohibitions and bypassed the advertising and recommendations on which Amazon lives. Perplexity called the suit a threat to all users who wanted a program to help them. A federal judge barred Comet from access on 10 March 2026; the court of appeals set the injunction aside on 4 August 2026, because it is not Perplexity but the customer who accesses Amazon’s servers with his account, even if he uses a program to do so, and because a ban would restrict consumer choice and a young technology without need; the case itself continues.⁹³ The question behind it is the old question of trade in a new form: a shop lives on the customer coming in, looking around and taking more than he wanted. An agent does not come in; it fetches what was ordered and leaves. Whoever sells through it sells to a buyer who cannot be distracted, sees no advertising and knows every price. For customers that is the market the textbooks promise; for the platforms that live on distraction it is a threat to their business (Figure 7).

The figures of 2025. The state from which the agent sets out is this. Amazon had sales of 716.9 billion dollars in 2025, 128.7 billion of it from the cloud, and earned 77.7 billion; 61 per cent of the items sold came from third-party sellers, and in the United States, by estimates, around four in ten dollars of online retail went through Amazon.⁹⁴ Through the stores on Shopify went 378 billion dollars, through eBay 79.6 billion, through Zalando 17.6 billion euros; the Otto Group, successor of the catalogue with 28 pairs of shoes, had sales of just under 15 billion euros.^{95, 96, 97, 98} In China goods worth 13.1 trillion yuan were sold on the net in 2025, 26.1 per cent of retail. In Germany customers paid 29 per cent with PayPal in 2025, 26 per cent on invoice, 14 per cent with debit card and 8 per cent with credit card; the banks’ scheme, Giropay, had been discontinued at the end of 2024, the European successor Wero had just started. 4.37 billion parcels were delivered, and since 19 June 2026 every store in the Union has had to offer, next to the button to buy, a button to withdraw.^{99, 100, 101, 102}

Try it yourself. What an agent does when shopping can be watched. Whoever tells an assistant with net access, ChatGPT, Gemini, Claude, Perplexity, “Find me a coffee grinder under 60 euros with a conical burr, deliverable by Friday, and show me three offers with sources” gets a list with links and can check which stores the agent visited, which it left out and whether the price it names is right. Usually it is right, sometimes it is out of date, and almost never is the store among them that pays the most for advertising on the net. Whoever then buys on the merchant’s site has experienced the trade of 2026: the agent searches, the human clicks, and the platform that wants to connect the two is not yet built.

The bottleneck. The chain of this text ends with an open link. Every stage since 1861 has shifted a piece of the way between wish and goods from the customer to a machine: browsing to the catalogue, ordering to the wire, searching to the search engine, comparing to the marketplace, carrying to the parcel courier. The last stage shifts the purchase itself, and with it the decision, to a program that someone has built and that belongs to someone. Open is to whom: the customer who instructs it, the merchant who pays for it, or the platform that runs it. Open is what becomes of the shops when no one comes in any more, and of the cities when the shops go. Open is who inspects the goods when they fly from the manufacturer straight to the customer, and who pays the duty. And open is whether trade, which for two thousand years was a conversation between people, becomes one between programs that haggle for us. The answer will not be found in a laboratory, but where it has always been found: at the checkout, on the question of who is serving at it.

The chain in brief

1. Mail order (1861–1979) bridges distance with catalogue, post, credit card and barcode; the medium is paper, and a person takes the order.
2. Videotex, Minitel and Btx (1979–1994) bring the catalogue to the screen, but in closed networks that do not reach beyond the country and the operator.
3. The web (1991–1997) opens the network to every merchant; cookie and SSL create shopping cart and checkout, Amazon and eBay the store and the marketplace; trust is missing.
4. Laws, trust marks and PayPal (1997–2002) create trust; the bubble shows that the store on the net is a warehouse, and Amazon survives as a logistics company.
5. The platform (2000–2010) rents customers, warehouses and computers to third-party sellers; Google sells visibility, Alipay solves the trust problem in China with escrow; the customer still sits at the desk.
6. The smartphone (2007–2016) puts the store in the pocket, the QR code turns the phone into a till, robots and couriers deliver the same day; the chain belongs to a few.
7. Courts, pandemic and the Union (2017–2023) set rules: tax, Digital Markets Act, Digital Services Act; the pandemic brings a jump and a correction, China brings the flood of parcels.
8. Tariffs and agents (since 2024) draw limits and shift the purchase to programs; open is who owns the agent that shops for us.

Timeline

Further reading

• Brad Stone: The Everything Store. Jeff Bezos and the Age of Amazon. Little, Brown, New York 2013. The history of Amazon up to the Kindle, based on conversations with hundreds of participants.
• Adam Cohen: The Perfect Store. Inside eBay. Little, Brown, Boston 2002. The marketplace’s first seven years, with access to Omidyar and the early users.
• Duncan Clark: Alibaba. The House That Jack Ma Built. Ecco, New York 2016. Alibaba, Taobao and Alipay, by a former adviser to the company.
• Julien Mailland, Kevin Driscoll: Minitel. Welcome to the Internet. MIT Press, Cambridge, MA 2017. The French network as the prehistory of online retail.
• Brian McCullough: How the Internet Happened. From Netscape to the iPhone. Liveright, New York 2018. The years 1993 to 2008 with chapters of their own on Amazon, eBay and the bubble.
• Robert Spector: Amazon.com. Get Big Fast. HarperBusiness, New York 2000. The early years, written at the height of the bubble.
• Alec MacGillis: Fulfillment. Winning and Losing in One-Click America. Farrar, Straus and Giroux, New York 2021. What the warehouses and the deliveries do to cities and work.
• Boris Emmet, John E. Jeuck: Catalogues and Counters. A History of Sears, Roebuck and Company. University of Chicago Press, Chicago 1950. The classic account of American mail order.
• Gerrit Heinemann: Der neue Online-Handel. Geschäftsmodelle, Geschäftssysteme und Benchmarks im E-Commerce. Springer Gabler, Wiesbaden, regularly updated. The German textbook on the business (in German).

Sources

The chronicle rests on the original papers, laws, IPO prospectuses, press releases, statistics and contemporary reports named in the footnotes.