Software You Never Own

On 6 May 2013, at its annual conference for creative professionals, Adobe announced that the products used by most of the world’s designers, photographers and video editors would no longer be sold in boxes. Creative Suite 6 would be the last version anyone could buy outright. New versions of Photoshop, Illustrator, InDesign and Premiere would be available only through Creative Cloud, a subscription. The top edition of the Creative Suite had cost $2,599. Creative Cloud for all applications cost $49.99 a month.¹

The reaction among many customers was furious. A photographer and student in Pennsylvania started a petition asking Adobe to keep selling perpetual licences; within days it had thousands of signatures, and within weeks tens of thousands.² The objections were practical and also something more. Freelancers worried about paying every month whether or not they had work. Users in countries with unreliable internet access worried about software that checked in with Adobe’s servers. And many people disliked the idea that the tools of their trade, the programs they had learned over years and used to make their living, would no longer be theirs. If they stopped paying, they could not open their own files in the program that made them.

Adobe went ahead. Its revenue fell in the following year, as it had told investors it would, and then rose for more than a decade. By fiscal 2025 Adobe’s subscription revenue was about $22.8 billion, almost all of its business.³ Adobe’s transition is the most famous example of a change that swept through the software industry over the first quarter of this century. Programs that had been products became services. Customers who had bought copies became subscribers. Companies that had sold licences began to report annual recurring revenue and churn. The change is usually called software as a service, or SaaS.

This essay follows how that happened and why. It is a story about business models, but its deeper subject is ownership: what it means to own a tool, what changes when you rent it instead, and who holds power in a relationship that never ends. To see why software became a subscription, we need to begin in the period before it was a product at all.

Before software was a product

In the 1950s and 1960s computers were too expensive for most organisations to own. Many companies rented time on them instead. Service bureaus ran payroll and accounting for clients on their own machines. From the early 1960s, time-sharing systems such as MIT’s Compatible Time-Sharing System let many users work at once on a single computer through terminals, each feeling as if the machine were theirs. Companies such as Tymshare and General Electric sold time-sharing commercially, charging by the hour of connection and the second of processing. Users paid for what they used, ran programs the provider maintained and kept their data on the provider’s machine. In many respects this was software as a service, forty years early. The essay on where programs run traces the long swing between central and personal computing that this began.

When companies did own or lease computers, software was not a separate thing to buy. IBM, which dominated the industry, supplied its operating systems, programming languages and many application programs with its machines, at no separate charge, along with the engineers who installed them and the training that taught customers to use them. Software was a cost of selling hardware.

That changed through law. In January 1969 the US Justice Department filed an antitrust suit against IBM, and competitors argued that bundling software with hardware made it impossible for anyone else to sell programs for IBM computers. IBM had already decided, in December 1968, to separate its prices, and in June 1969 it announced that from the start of 1970 it would charge separately for some software, training and engineering services, while cutting the price of its hardware.⁴ Historians often date the independent software industry from that announcement. Once IBM’s programs had a price, others could compete to supply better ones. Estimates put the revenues of independent software companies at $20 million to $50 million in 1969 and about $400 million by 1975.

The age of the box

The personal computer created a mass market for software, and with it a model that lasted for decades: a program was sold in copies. In February 1976 Bill Gates, then a young co-founder of Microsoft, wrote an “Open Letter to Hobbyists” complaining that most users of Microsoft’s BASIC had copied it rather than paid for it, and arguing that no one could afford to write good software if it was treated as free.⁵ The letter set out the economic premise of the software industry. Writing software is expensive; copying it costs almost nothing; so the industry must sell copies and protect them, through law and through technical locks.

For the buyer, a box of software looked like a book: a physical object bought once and kept. Legally it was something else. Publishers sold licences, permissions to use the program under terms the publisher set, which typically forbade copying, reverse engineering and often resale. The terms were printed inside the box or displayed on screen when the program was installed, so that the buyer learned them only after paying. In 1996 a US appeals court, in ProCD v. Zeidenberg, held that such “shrink-wrap” licences were enforceable contracts: by keeping and using the software, the buyer accepted them.⁶ In 2010 another appeals court, in Vernor v. Autodesk, held that a customer who received software under a licence that reserved title to the publisher and restricted transfer was a licensee, not the owner of a copy, and could not resell it.⁷ The sense of owning boxed software was, in law, mostly an illusion. Subscription would later make the arrangement visible.

The box also shaped how software was made. Because revenue came from sales of new versions, publishers released major upgrades every year or two, each with new features that would persuade existing customers to pay again. Features were planned for the box, not for the user’s daily work, and bugs found after release waited for the next version. Customers often skipped versions, which made the publishers’ revenue lumpy and dependent on whether each release was persuasive.

Enterprise software and its discontents

For businesses, the box became something larger. From the 1970s, the German company SAP built integrated systems that ran a company’s accounting, purchasing, manufacturing and human resources from one database. In the 1980s and 1990s, enterprise resource planning (ERP) systems from SAP, Oracle, PeopleSoft and others, and customer relationship management (CRM) systems such as Siebel’s, became the operating core of large organisations. The companies that sold them grew rich.

The model had a distinctive structure. A customer paid a large upfront licence fee, often millions of dollars, and then an annual maintenance fee, commonly around a fifth of the licence price, for updates and support. Installing the software required buying servers, hiring consultants and adapting the software, or the organisation, to each other, a process that could take years and often cost several times the licence fee. Once installed, the system was hard to replace, so the maintenance fees continued for decades. Enterprise software was already, in effect, a subscription. But the customer bore the cost and risk of running it, and upgrades were so disruptive that many companies stayed on old versions for years.

The people who used the software, salespeople entering data about their customers, for example, often disliked it. It was slow, complicated and designed for the managers who bought it rather than the employees who used it. And the department that controlled it, central IT, stood between every business unit and every new tool.

The first attempt: application service providers

In the late 1990s, as internet connections improved, a group of companies offered an alternative. Application service providers, or ASPs, such as USinternetworking and Corio, would run enterprise software for customers in their own data centres and charge a monthly fee. Customers would no longer need their own servers or specialist staff.

Most of them failed. They changed where the software ran without changing how it was built. Each customer still had its own copy of the conventional software, installed and customised separately, so the ASP’s costs grew with every customer it added. Many had borrowed heavily to build data centres. When the dot-com boom ended in 2000 and 2001, funding disappeared, and USinternetworking filed for bankruptcy protection in January 2002. Analysts at the time described the failures as the result of debt-funded infrastructure, too little standardisation and a model that resembled old-fashioned time-sharing more than a new kind of business.⁸ The idea of renting software over the internet survived. What had to change was the software itself.

“The end of software”: Salesforce

In March 1999 Marc Benioff, a senior executive at Oracle, founded Salesforce in a rented apartment in San Francisco with three engineers, Parker Harris, Dave Moellenhoff and Frank Dominguez.⁹ Their product was software for sales teams to track customers and deals, the same function Siebel sold to large companies. But Salesforce delivered it entirely through a web browser. There was nothing to install. A sales manager could sign up with a credit card and pay per user per month.

The technical decision that made the business work was multi-tenancy. All of Salesforce’s customers used a single shared system, one code base and one database, with each customer’s data and customisations kept separate by software rather than by separate installations. The company’s engineers built the system around metadata: each customer’s fields, layouts and rules were stored as data describing the application, which a common engine interpreted at runtime. Harris later said that multi-tenancy pushed the company to rely entirely on this metadata model, because in a shared system there were no shortcuts.¹⁰ The result was that one upgrade improved the product for every customer at once, with no installation projects. The cost of serving an additional customer was small, and fell as the system grew.

Benioff’s marketing was as unconventional as the architecture. The company’s logo was the word “software” in a red circle with a line through it, and its slogan was “the end of software.” In 2000 it hired actors to stage a mock protest outside a conference run by Siebel, the market leader, carrying signs against software, and invited journalists to watch.⁹ The message was that installed enterprise software, with its licences, consultants and failed projects, was obsolete.

Salesforce sold first to small companies and to departments within large ones, often without the involvement of the central IT department, because a sales manager could buy it on a corporate card. Users liked it because it was simpler than the systems their companies had installed. Over time the departments became whole companies. Salesforce went public in June 2004, raising $110 million. In 2005 it opened AppExchange, a marketplace for applications built by others on its platform, and later offered its system as a platform on which customers could build their own applications. A company that had promised “the end of software” became one of the largest software companies in the world. Its revenue for the fiscal year ending January 2026 was $41.5 billion.¹¹

Salesforce was not alone. NetSuite, founded in 1998 by Evan Goldberg with financial backing from Oracle’s Larry Ellison, offered accounting and business management online to mid-sized firms. Many others followed for email marketing, collaboration and human resources. By 2001 the industry had begun calling the model “software as a service.”

The infrastructure underneath

Building a shared online service required data centres, and in the early 2000s each SaaS company built its own. In 2006 Amazon launched Simple Storage Service (S3) and Elastic Compute Cloud (EC2), which rented storage and computing capacity by the hour and the gigabyte. Any start-up could now run a global service on Amazon’s machines, paying for what it used, without buying a server. Microsoft and Google followed with their own cloud platforms. The rental model that SaaS offered its customers was now available, one layer down, to SaaS companies themselves.

Inside Microsoft, the threat had been recognised early. In October 2005 Ray Ozzie, the company’s chief technical officer, wrote a memo to its executives titled “The Internet Services Disruption,” arguing that services delivered over the internet, supported by advertising or subscriptions, would disrupt the business of selling installed software. It was echoing a similar memo that Bill Gates had written a decade earlier about the internet itself, annotated on this site as The Internet Tidal Wave.

Consolidation and a new generation

The incumbent vendors of installed software responded at first by consolidating. In 2003 Oracle launched a hostile bid for PeopleSoft, the leading vendor of human resources software, and after an eighteen-month battle completed the acquisition in early 2005 for about $10.3 billion. It agreed to buy Siebel, the CRM leader, later in 2005 for about $5.85 billion. These acquisitions gave Oracle a vast installed base paying annual maintenance, a highly profitable business.

They also had a consequence Oracle may not have intended. Dave Duffield, PeopleSoft’s founder, and Aneel Bhusri, one of its senior executives, left after the takeover and in March 2005 founded Workday, which built human resources and later finance software from scratch as a shared online service for large enterprises.¹² Workday’s bet was that even the largest and most cautious companies would put their most sensitive data, about employees and money, in a system shared with other customers. Many did. The founders’ reputation among PeopleSoft’s former customers, some of whom resented the takeover, helped make the case.

By the late 2000s SaaS companies were selling not only to small firms but to the largest enterprises in the world, and the remaining incumbents began to build or buy their own cloud products. SAP bought SuccessFactors in 2011 and Ariba in 2012. Oracle bought NetSuite in 2016.

The incumbents convert: Office and Creative Cloud

The most consequential conversions were by the makers of software that individuals used every day.

Microsoft launched Office 365 for businesses in June 2011, bundling the familiar desktop programs with online email, file storage and collaboration, sold per user per month. With Office 2013 it added subscriptions for consumers. Unlike Adobe, Microsoft continued to sell perpetual versions of Office alongside the subscription. The transition was slower, and less resented. In the fourth quarter of Microsoft’s fiscal 2017, revenue from Office 365 exceeded revenue from conventional Office licences for the first time.¹³ In 2020 the consumer subscription was renamed Microsoft 365.

Adobe chose a faster path. Its chief financial officer, Mark Garrett, later explained the reasoning in an interview. Adobe sold about three million perpetual units a year, and that number had stopped growing; the company was increasing revenue by raising prices, which he described as unsustainable. Major releases came only every 18 or 24 months, while customers’ needs changed faster. Only about 5 percent of Adobe’s revenue was recurring, so in the recession of 2008 and 2009 its revenue and share price had fallen further than those of companies with subscription income. And trials suggested that a lower monthly price would bring in many users who had never bought the full suite.¹⁴

Adobe announced the plan to investors in November 2011, launched Creative Cloud in May 2012 alongside the boxed suite, and in May 2013 ended development of the boxed version. Garrett said the company had expected its revenue, earnings and share price to fall during the transition, and told investors so. To give them something to measure, Adobe began reporting the number of subscribers, the average revenue per user and a new figure, annualised recurring revenue, and set public targets, such as four million subscribers in 2015. Revenue fell from about $4.40 billion in fiscal 2012 to $4.06 billion in fiscal 2013. Adobe had more than 3.4 million paid Creative Cloud subscribers by the end of 2014, and by 2016 its annual revenue had reached a record $5.85 billion.¹⁵ Photographers, one of the most vocal groups of critics, were offered a cheaper plan with Photoshop and Lightroom, which eased much of the protest.

The economics of recurring revenue

The subscription model changed the economics of software in ways that went beyond the price tag.

For the vendor, revenue arrived gradually. A customer who would once have paid $2,599 at once now paid $50 a month; the vendor recovered the cost of acquiring the customer only after many months, and profited only if the customer stayed. That made retention the central measure of the business. Subscription companies came to be judged by a set of numbers: annual recurring revenue (ARR); churn, the share of customers or revenue lost each period; net revenue retention, which measures whether existing customers spend more or less over time; customer acquisition cost; and the lifetime value of a customer. A popular rule of thumb, attributed to the venture investor Brad Feld in 2015, held that a healthy software company’s growth rate and profit margin should add up to at least 40 percent.

Investors valued recurring revenue highly, because it was predictable. A company whose customers renewed automatically could forecast years ahead with confidence, and a dollar of subscription revenue was often valued at many times a dollar of one-time licence revenue. The founder of the billing company Zuora, Tien Tzuo, popularised the phrase “the subscription economy” to describe the spread of the model beyond software, to cars, media, food and almost anything else that could be rented rather than bought.

For the customer, the change was mixed. The upfront cost fell sharply, which let small businesses and individuals use tools they could not previously afford. Updates arrived continually, with no upgrade projects. The vendor ran the servers, handled security and kept the software working. Spending moved from capital budgets to operating budgets, which made purchases easier to approve and to cancel. But the total cost over many years was often higher. Data lived on the vendor’s systems, in the vendor’s formats. And the customer never stopped paying.

The discontents of subscription

As subscriptions multiplied, so did complaints. Consumers found themselves paying for many services they rarely used. Businesses discovered that departments had signed up for overlapping tools without central oversight, a problem that created a market for software to manage software subscriptions.

Cancellation became contested ground. Because churn was the number investors watched, some companies made leaving difficult: cancellation pages hidden in menus, requirements to phone during business hours, offers that had to be declined repeatedly. In June 2024 the FTC sued Adobe, alleging that it had steered customers into an “annual, paid monthly” plan without clearly disclosing that cancelling in the first year could cost hundreds of dollars, and had made cancellation difficult. In March 2026 the case was settled; according to Adobe, it agreed to pay $75 million and to provide $75 million in free services to some customers.¹⁶ In October 2024 the FTC finalised a “click-to-cancel” rule requiring that cancelling a subscription be as easy as signing up. In July 2025 a federal appeals court vacated the rule on procedural grounds before it took effect, though states such as California have adopted their own requirements.¹⁷

Control over the product also changed hands. With boxed software, a user could keep using an old version indefinitely. With a subscription, the vendor could change features, interfaces and terms at will. In June 2024 a change to Adobe’s terms of use, which customers read as giving the company access to their work, perhaps to train AI, provoked a wave of anger; Adobe clarified that it did not train its AI on customers’ content and revised the language. The episode showed how much trust the relationship now required.

In Europe, regulators addressed the problem of leaving. The EU Data Act, which began to apply in September 2025, requires providers of cloud and data-processing services to help customers switch to competitors, to let them export their data and, over a transition period, to stop charging fees for switching.

AI and the seat

For two decades SaaS was priced mostly per seat: a monthly fee for each employee who used the software. That made revenue grow with customers’ headcount. The arrival of capable AI agents in 2024 and 2025 threatened that link. If an agent could handle customer service enquiries, write reports or process invoices, a customer might need fewer human users, and therefore fewer seats.

In August 2024 the chief executive of the payments company Klarna said it had stopped using Salesforce and would soon stop using Workday, replacing them with its own systems and AI, remarks widely read as a sign that AI could replace SaaS altogether. In 2025 he clarified that Klarna had consolidated its data and built its own internal tools, and had not replaced its CRM with a language model, saying he was “tremendously embarrassed” by how his comments had been interpreted.¹⁸ The episode was more significant as a signal of investor anxiety than as evidence of what AI could do.

Vendors responded by changing what they charged for. In 2024 Salesforce launched Agentforce, its AI agents, at a list price of $2 per conversation, and later moved to credits charged per action.¹⁹ Others experimented with charging per task completed or per outcome achieved. In early February 2026, after Anthropic released plugins that let its Claude agent carry out work in specialised fields, investors sold software stocks heavily on fears that agents would reduce demand for seats; commentators called it the “SaaSpocalypse,” and estimates of the value lost across the sector ran from several hundred billion to more than a trillion dollars within weeks.²⁰ Some of the losses were later recovered. Whatever the pricing unit of the future turns out to be, the episode showed how much of the subscription model’s value rested on the assumption that work is done by people sitting at software.

What SaaS is today

Today most new business software is sold as a service, and most large software companies earn most of their revenue from subscriptions. SaaS is a technical architecture, shared systems run by the vendor and reached through a browser or app. It is a business model, recurring payments in exchange for continued access and continuous improvement. And it is a relationship, in which vendor and customer remain bound together, with the vendor responsible for keeping the software running and the customer for continuing to pay.

How subscription software changed the way organisations work

Several shifts in how companies buy and use software follow from the move to subscription.

From a purchase to a renewal

A licence was a decision made once. A subscription is a decision remade every month or year, at least in principle. Vendors had to keep earning the renewal, which pushed them to care about whether customers used and valued their products after the sale. A new profession, customer success, grew up to make sure they did.

From capital to operating expense

Buying software outright was a capital investment, approved through budgets and depreciated over years. Subscriptions are operating expenses, paid as they go. That made software easier to adopt and to drop, and it shifted power from central finance and IT departments to the managers who could approve a monthly charge.

From the IT department to the user

Because SaaS could be bought on a credit card and used in a browser, individual teams adopted tools without asking permission. Usability became a competitive weapon, because the person choosing the tool was now often the person who would use it. The costs were fragmentation, security risks and a proliferation of overlapping subscriptions.

From versions to a stream

Boxed software arrived in versions. SaaS changes continuously, sometimes several times a day. Users gained fixes and features without upgrade projects, and lost the ability to choose when, or whether, their tools changed.

From owning data to visiting it

The data a company creates in a SaaS application lives in the vendor’s system. Getting it out, in usable form, depends on export tools the vendor provides. The company’s records, customers and processes are increasingly held in other people’s databases.

What the history suggests

The history of subscription software repeats patterns found elsewhere on this site. The first attempt at a new model, the ASPs, failed because it moved an old product into a new channel without rebuilding it. The winners rebuilt the product around the new economics, shared systems that grew cheaper with scale. Incumbents first consolidated the old market and then converted, with Microsoft’s gradual path drawing less anger than Adobe’s abrupt one, and both succeeding. As in computing generally, the model swung back towards the centralised, rented arrangement of the time-sharing era, now at a scale the bureaus could not have imagined.

Subscription software also shows something more particular. It aligned the vendor’s income with the customer’s continued use, which was a real improvement on a model that rewarded vendors for selling upgrades whether or not they helped. But it also gave vendors power over customers who could no longer keep an old version, take their data easily or stop paying without losing access to their own work. The tension between those two effects, a relationship that rewards serving the customer and one that rewards making it hard to leave, runs through the whole history, and it is where the philosophy begins.

Understanding the philosophy

The rest of this essay examines the ideas beneath subscription software, one at a time. Each is explained in plain terms, connected to the tradition of thought that has examined it, traced in the history above, tested against its strongest objection and related to how organisations decide today. The connections to philosophers, economists and lawyers are interpretations offered to make the ideas clearer. The founders of SaaS companies were solving business and engineering problems, and did not, as a rule, set out to answer these thinkers.

1. Owning a thing and using it

The idea in plain terms

What do people want from a tool: to own it, or to use it? Most of the time the answer seems to be use. A designer wants Photoshop to work, not a title deed to Photoshop. But ownership brings things that use alone does not: the right to keep using the tool as it is, to sell it, to modify it, and not to be at the mercy of whoever supplies it. The subscription separates use from ownership, and makes us notice what ownership was for.

The tradition

Aristotle observed in the Rhetoric that wealth as a whole consists in using things rather than in owning them, since it is the use of property that brings its benefits. Roman law, as later systematised by civil lawyers, divided full ownership into three rights: usus, the right to use a thing; fructus, the right to its fruits or income; and abusus, the right to dispose of it, by selling, changing or destroying it. A tenant of land had usus and perhaps fructus; only the owner had abusus. Ownership, on this view, is not one thing but a bundle of rights that can be separated.

The English philosopher John Locke, in his Second Treatise of Government (1689), grounded property in labour: a person who mixes their labour with something unowned makes it their own. For Locke, property was a protection of the individual against arbitrary power, including the power of the state; a person secure in their property could not easily be coerced. That idea, that ownership is a form of independence, sits behind much of the resistance to subscription.

In 2000 the American social theorist Jeremy Rifkin argued in The Age of Access that capitalism was moving from markets, in which property is exchanged, to networks, in which suppliers grant access to services for a fee, and that people would increasingly pay for access to experiences rather than own goods. He saw both convenience and danger in the shift: a world in which almost everything is rented is one in which a few suppliers control the terms of daily life.

In law, the most important doctrine here is the first-sale doctrine, recognised by the US Supreme Court in Bobbs-Merrill v. Straus (1908). Once a copyright owner sells a copy of a work, the buyer may resell or give away that copy without the owner’s permission. That is why second-hand bookshops are legal. Software publishers avoided the doctrine by licensing rather than selling copies, and courts, in cases such as Vernor v. Autodesk, largely accepted the distinction.

Where it shows up

The boxed program sat in an ambiguous position: the buyer had the physical disc and a permanent ability to use the version they had bought, usus without end, but not the full bundle. Subscription separated the rights completely. The subscriber has usus for as long as they pay, and nothing more. They cannot keep an old version, sell their access or use the tool after they stop paying. Adobe’s customers who objected in 2013 were defending the part of ownership that boxed software had allowed them: the right to keep using a tool they had paid for, unchanged, without asking anyone.

The strongest objection

The objection is that the ownership of software was always largely an illusion, as the licensing cases showed, and that most users never wanted the other rights in the bundle. Few people resold software or kept using versions a decade old. The subscription gives them what they actually used, continuous access to the current version, at a lower upfront price, and relieves them of the burdens of ownership: installation, maintenance, security. For businesses, renting tools is as old as leasing offices. This is largely right for most users most of the time. It is less right at the edges: for people in places with poor connectivity, for those whose income varies, for archives and institutions that need to open old files decades later, and for anyone whose relationship with the supplier goes wrong.

What it means today

For any tool your organisation depends on, it is worth listing which rights you actually hold. Can you keep using it if the price doubles? Can you take your data, in a usable format, if you leave? Can you keep a version that works if the vendor changes the product in ways you dislike? Where the answers are no, the organisation has a dependency, not an asset, and its continuity depends on the vendor’s continued goodwill and solvency. That may be an acceptable trade. It should be a conscious one.

2. Why companies rent what they could build

The idea in plain terms

Every business uses software, and in principle every business could write its own. Most do not, and increasingly they do not even install and run the software they buy. They rent it, run by someone else. Why? Because doing something inside the firm has costs, of hiring, managing and maintaining, and buying it from outside has different costs, of finding a supplier, agreeing terms and depending on them. Organisations draw their boundaries where one set of costs outweighs the other.

The tradition

In 1937 the British economist Ronald Coase asked a question economists had overlooked: if markets allocate resources efficiently, why do firms exist at all? Why not hire every worker and buy every input through the market, task by task? His answer was that using the market has costs, which later economists called transaction costs: finding out prices, negotiating and enforcing contracts. Firms exist when organising an activity internally is cheaper than buying it. The site’s idea Why Firms Exist explains the argument in full.

The American economist Oliver Williamson developed Coase’s idea from the 1970s onward. He emphasised asset specificity: when one party makes an investment that is valuable only within a particular relationship, such as customising its processes around a supplier’s system, it becomes vulnerable to that supplier. The supplier can raise prices or reduce service, knowing that the customer cannot easily leave. Williamson called this the hold-up problem, and argued that firms tend to bring such relationships inside, or protect them with long contracts, to avoid being held up. Coase and Williamson both received the Nobel Prize in economics, in 1991 and 2009.

In 2003 the writer Nicholas Carr argued in the Harvard Business Review, in an essay titled “IT Doesn’t Matter,” that information technology was becoming a commodity infrastructure, like electricity or railways, that every firm needed and none could use for lasting advantage. In The Big Switch (2008) he predicted that computing would be supplied by utilities, as electricity was once generated in individual factories and then bought from the grid.

Where it shows up

SaaS reduced the costs of buying software from outside. There was nothing to install, no servers to buy, no upgrade project; a manager could sign up in minutes. At the same time, it reduced the advantage of building or running software inside, because a shared system serving thousands of customers could be improved faster and run more cheaply than any single company’s installation. Coase’s logic predicted what happened: activities that firms had performed internally, running servers, maintaining applications, moved outside. Carr’s utility prediction was largely borne out for infrastructure.

But Williamson’s warning applied too. A company that builds its sales process around Salesforce, or its finance processes around a particular ERP system, has made a highly specific investment. Its data, its trained staff, its integrations and its customisations all depend on the vendor. The subscription that was easy to start is hard to end.

The strongest objection

The objection is that competition protects customers from hold-up. SaaS markets have many vendors, contracts come up for renewal, and a vendor that exploits its customers will lose new ones to rivals. Software vendors depend on their reputation. Moreover, the specific investments are smaller than with installed software, since there are no servers or years-long implementation projects. This is partly right, and the history of SaaS shows real competition in many categories. But the switching costs are concentrated in data and processes, which are hard to measure and easy to underestimate, and in the core systems of large organisations, competition operates over very long cycles.

What it means today

The decision to rent rather than build is a decision about where to accept dependency. For commodity functions, such as email, payroll or document storage, renting is almost always right. For functions that define how a company competes, it is worth asking whether a rented system will make the company more like every other company using it, and whether the company’s distinctive way of working can survive inside a vendor’s product. Contracts should be written with leaving in mind: data export, transition support and price protections at renewal are easier to negotiate before the dependency exists.

3. The relationship as the product

The idea in plain terms

A shop that sells you a hammer has finished its work when you leave. A company that rents you a service begins its work when you sign. In a subscription, the product is not the software at the moment of purchase; it is the whole continuing relationship, in which the vendor has to keep earning your payment. That changes what the vendor cares about, for better and for worse.

The tradition

In 1960 the Harvard marketing professor Theodore Levitt published “Marketing Myopia,” arguing that companies fail when they define themselves by the products they make rather than the needs they serve. Railways declined, he wrote, because they thought they were in the railway business rather than the transportation business. In 1983, in “After the Sale Is Over,” he argued that in many industries the sale is only the beginning of a long relationship, whose quality determines whether the customer returns, and that companies should manage relationships rather than transactions.

The American legal scholar Ian Macneil developed, from the 1960s to the 1980s, a theory of relational contracts. Classical contract law treats agreements as discrete exchanges, fully specified in advance. Macneil argued that most important economic relationships, such as employment, franchises and long-term supply, cannot be fully specified, because the future is uncertain. They are governed instead by norms that develop over time: solidarity, reciprocity, flexibility, the expectation that each party will behave reasonably when circumstances change. A subscription to business software is such a relationship.

In 2004 the marketing scholars Stephen Vargo and Robert Lusch proposed what they called service-dominant logic: the idea that all economies are service economies, that goods are only vehicles for delivering service, and that value is created jointly by supplier and customer in use rather than handed over at the point of sale.

Where it shows up

The subscription model made Levitt’s and Vargo and Lusch’s ideas into accounting. A SaaS vendor that spends heavily to acquire a customer loses money unless the customer stays for years. So vendors measure usage, intervene when customers seem to be struggling, and employ customer success teams whose job is to ensure that customers get value from what they bought. Salesforce’s architecture, which delivered improvements to all customers at once, and Adobe’s promise of continuous updates were ways of keeping the relationship worth renewing. Garrett’s case for Adobe’s move was in part that boxed software had rewarded the company for persuading people to upgrade rather than for serving them continuously.

The strongest objection

The objection is that a relationship one party cannot easily leave is not a relationship of equals. Macneil’s relational norms depend on each side needing the other. In practice, once a customer is locked in, the vendor’s incentive is to extract value rather than to create it, by raising prices at renewal, charging for features that were once included or making cancellation difficult. The FTC’s case against Adobe alleged exactly this. The relationship can become a hold-up dressed in the language of partnership. This objection is serious, and the history shows both sides of the relationship: vendors who earned renewals by improving their products, and vendors who secured them through friction.

What it means today

A subscription is a long-term relationship, and it should be evaluated as one. Before signing, it is worth asking how the vendor makes money after the sale: through the customer’s success, or through the customer’s inability to leave? Vendors themselves should notice which kind of retention they are building. Customers who stay because they are getting value are an asset. Customers who stay because leaving is painful are a liability that will eventually appear as resentment, churn or regulation.

4. The product that never stays the same

The idea in plain terms

A book does not change after you buy it. Boxed software did not change either, unless you bought an upgrade. Subscription software changes all the time. Features appear and disappear, interfaces are redesigned, terms are revised. The program you use today is not the one you signed up for. In what sense is it the same product, and who decides how it changes?

The tradition

The puzzle of identity through change is ancient. The Greek philosopher Heraclitus, around 500 BCE, is reported by Plato to have said that one cannot step twice into the same river, because new waters are always flowing. The Greek writer Plutarch recorded the story of the ship of Theseus, preserved in Athens for centuries, whose planks were replaced one by one as they rotted, until philosophers argued about whether it was still the same ship. Thomas Hobbes, in De Corpore (1655), added a twist: if someone collected the old planks and reassembled them, which ship would be the original?

Philosophers have answered in different ways. Some say identity lies in continuity of form or function; others in an unbroken history of gradual change; others that the question has no deep answer, and that “sameness” is a matter of convention and purpose.

Software engineering met the puzzle in practical form. From the 1970s the computer scientist Meir Lehman, working with László Belady, studied how large programs evolve, and formulated what became known as Lehman’s laws of software evolution. The first, which he called continuing change, holds that a program used in the real world must be continually adapted, or it becomes progressively less satisfactory. Software that serves a changing world cannot stay the same and remain useful.²¹

Where it shows up

SaaS took Lehman’s first law to its conclusion. Instead of accumulating changes into periodic versions, vendors deliver them continuously, sometimes many times a day. Salesforce in its early years claimed dozens of major releases, each with many new features, delivered to all customers at once. Users of Creative Cloud and Microsoft 365 receive updates they did not ask for and cannot decline. The product’s identity lies not in any fixed set of features but in the brand, the account and the data, which persist while everything else changes, like the ship whose name endures as its planks are replaced.

The strongest objection

The objection is that continuous change is what users want. Security fixes arrive immediately; bugs are corrected without waiting for a new version; improvements reach everyone. Users who stayed on old boxed versions were exposed to security risks and missed improvements. Lehman’s law shows that frozen software decays; the subscription simply makes the necessary change automatic. This is mostly true. But change imposed from outside also has costs: retraining, broken workflows, features removed that some users relied on. And the vendor’s interests in change, adding features that sell, pushing users towards new paid products, are not always the user’s.

What it means today

Organisations that depend on rented software should treat change as a constant and plan for it: test important workflows when vendors announce changes, keep documentation of how their processes depend on particular features, and negotiate notice periods for significant changes. Vendors, for their part, owe users a way to understand what is changing and why. A product that changes continuously needs to explain itself continuously.

5. The freedom to leave

The idea in plain terms

The best protection a customer has is the ability to walk away. If leaving is easy, vendors must keep earning the customer’s business. If leaving is hard, because data is trapped, processes depend on the system or contracts impose penalties, the vendor’s incentives change. Much of the economics and ethics of subscription comes down to how hard it is to leave.

The tradition

In 1970 the economist Albert O. Hirschman published Exit, Voice, and Loyalty, which asked how people respond when an organisation they depend on declines in quality. They can exit, taking their business elsewhere, or use voice, complaining and trying to change it from within. Loyalty makes people more likely to use voice than exit. Hirschman argued that both mechanisms matter, and that an organisation that makes exit impossible removes an essential signal and loses the discipline that comes with it. The site’s idea Exit, Voice and Loyalty explores this in depth.

In 1998 the economists Carl Shapiro and Hal Varian, in Information Rules, analysed switching costs and lock-in in information industries. They argued that much of the value of an information business comes from the costs its customers would bear to switch, and advised companies to understand the lock-in cycle: attract customers with low prices, let switching costs accumulate, and then capture value. The economist W. Brian Arthur showed how small early advantages can lock a market into a technology through increasing returns, an idea explored in Path Dependence.

Lawmakers have increasingly treated the freedom to leave as a right. The EU’s General Data Protection Regulation (2018) gave individuals a right to data portability. The EU Data Act extended the principle to businesses using cloud services, requiring providers to support switching.

Where it shows up

Subscription software is lock-in in its purest form. A customer’s data, workflows, integrations and trained staff all accumulate inside the vendor’s system, and each year of use raises the cost of leaving. Vendors seldom need to block exit explicitly; the accumulated specificity does it for them. Where they have made exit harder deliberately, through early termination fees or obstacles to cancellation, regulators have stepped in, as in the FTC’s case against Adobe and the click-to-cancel rule. In Hirschman’s terms, customers of a dominant SaaS product often have little exit and limited voice, and loyalty is replaced by dependence.

The strongest objection

The objection is that switching costs are mostly the natural result of the value customers get. A company whose processes are deeply integrated with a system has invested in it because the system made those processes better. Lock-in, on this view, is the shadow of usefulness, not a scheme. And markets for SaaS remain competitive at the point of initial choice. This is fair, and not every high switching cost reflects a vendor’s strategy. But Shapiro and Varian’s advice to vendors shows that lock-in can be designed, and the history of cancellation practices shows that some companies have done so.

What it means today

For buyers, the moment of greatest bargaining power is before signing. Exit rights, data export, transition assistance and limits on price increases should be negotiated then. Organisations should periodically test whether they could leave a critical vendor, by exporting their data and checking that it is usable, before they need to. For vendors, the most durable retention comes from customers who could leave and choose not to. A business that depends on customers who cannot leave is fragile in a way its metrics do not show.

6. Paying for what we do not use

The idea in plain terms

Subscriptions are easy to start and easy to forget. People sign up for a service they mean to use, use it less than they expected, and keep paying, because cancelling takes effort, because they plan to use it next month, or because the small monthly charge does not feel like a decision. Companies earn a meaningful part of their revenue from customers who are not using what they pay for.

The tradition

Classical economics assumes that people choose what they prefer and act on their choices. Behavioural economics, developed from the 1970s by psychologists Daniel Kahneman and Amos Tversky and economists such as Richard Thaler, documented systematic departures from that assumption. Thaler’s work on mental accounting showed that people treat money differently depending on how it is framed and when it is paid. In 1988 William Samuelson and Richard Zeckhauser documented status quo bias: people stick with existing arrangements far more than their stated preferences would predict. Loss aversion, the tendency to weigh losses more heavily than equivalent gains, makes giving something up feel costly, a theme explored in Loss Aversion.

In 2006 the economists Stefano DellaVigna and Ulrike Malmendier published a study of health club members titled “Paying Not to Go to the Gym.” Members who chose monthly contracts paid on average more than $17 per visit, when they could have paid $10 per visit with a pay-as-you-go pass, because they overestimated how often they would attend. They also took months, on average, to cancel after they stopped going. The authors interpreted this as overconfidence about future self-control. In 2002 John Gourville and Dilip Soman showed in the Harvard Business Review that the way payments are timed affects how much people use what they buy: people are less likely to use something whose cost they have stopped feeling.

The designer Harry Brignull coined the term “dark patterns” in 2010 for interfaces designed to trick users into choices against their interest, such as subscriptions that are easy to start and hard to stop.

Where it shows up

Subscription businesses benefit from all of these tendencies. A small monthly charge is less salient than a one-time purchase. Automatic renewal makes continuing the default. Annual commitments paid monthly, like the Adobe plans at issue in the FTC case, combine the low felt price of monthly payment with the low churn of an annual contract. In businesses, the same effects appear as unused licences: seats bought for employees who left, subscriptions to tools adopted by one team and forgotten. The click-to-cancel rule, the state laws that followed its vacatur and the EU’s rules on switching are all attempts to correct for predictable human tendencies that a business model can exploit.

The strongest objection

The objection is that adults are responsible for their own choices, and that subscriptions also benefit users with these same biases: someone who pays monthly for a gym may go more often than they would otherwise, and a business that subscribes to a tool may use it more because it is available. Regulating choice architecture is paternalistic. Moreover, most SaaS revenue comes from businesses that actively use the products, not from forgotten consumer subscriptions. This is right for much of enterprise SaaS. The concern applies most to consumer subscriptions and to practices designed to exploit inattention, which is where regulators have focused.

What it means today

Organisations should audit their subscriptions regularly: which are in use, by whom and for what. Unused seats are often a larger cost than any single price increase. For vendors, revenue from customers who do not use the product is fragile, because it disappears when they notice, and corrosive, because it teaches the organisation to value inattention. The honest measure of a subscription business is how much of its revenue comes from customers who would renew if asked.

7. The future made into a number

The idea in plain terms

A subscription business is valued for its future. Investors look at recurring revenue, assume that most customers will keep paying and calculate what that stream of payments is worth today. Recurring revenue made software companies among the most highly valued businesses in the world. It also created a set of numbers that management could optimise directly, and a vulnerability: if the assumptions about the future change, the value can disappear quickly.

The tradition

The idea that an asset is worth the present value of the income it will produce was formalised by the American economist Irving Fisher in The Theory of Interest (1930), and applied to shares by John Burr Williams in The Theory of Investment Value (1938). A future payment is worth less than a present one, because money now can be invested and because the future is uncertain; the value of an asset is the sum of its future payments, discounted to today. The more predictable the future payments, the lower the discount and the higher the value.

John Maynard Keynes, in The General Theory (1936), warned that our knowledge of the future on which such calculations depend is often extremely slight, and that valuations rest heavily on convention, the assumption that the present state of affairs will continue, and on the shifting confidence he called animal spirits. When the convention changes, valuations can move sharply even if nothing in the present has changed.

Once a number is used to judge a business, it is subject to the pressure described by Goodhart’s law, that a measure which becomes a target ceases to be a good measure. The site’s idea When a Measure Becomes a Target traces this across many fields.

Where it shows up

The SaaS industry’s metrics, ARR, net revenue retention, churn and the Rule of 40, are attempts to measure the predictability and growth of future payments. Adobe’s transition depended on persuading investors to value it by these numbers while its conventional revenue fell. They worked well, and they also became targets. Annual contracts with early termination fees reduced measured churn. Price increases on locked-in customers raised net revenue retention. The February 2026 sell-off showed Keynes’s point: the convention that seats would keep growing with customers’ headcount was suddenly in doubt, and valuations fell before any revenue had.

The strongest objection

The objection is that recurring revenue really is more valuable, because it really is more predictable, and that the metrics capture something true about the quality of a business: customers who renew and expand are evidence that the product works. Gaming exists, but investors and auditors learn to look through it. This is fair. The metrics are useful. The point is not that they are false but that they describe the future by extrapolating the present, and that management can improve them in ways that do not improve the business.

What it means today

Leaders of subscription businesses should watch the behaviour behind their metrics: whether retention comes from value or friction, whether expansion comes from usage or price increases. Investors and boards should ask what would have to remain true for the numbers to hold, and how confident anyone can be that it will. And any organisation valued on its future should remember that the future can be repriced suddenly, when the convention changes.

8. Rent and the toll on the road

The idea in plain terms

Critics of subscription describe it as rent: a payment extracted continually for access to something the payer cannot do without, by a company that controls it. Defenders describe it as payment for a continuing service. The difference matters, because economists have long distinguished income earned by creating value from income earned by controlling something scarce.

The tradition

Adam Smith, in The Wealth of Nations (1776), wrote that once all the land of a country has become private property, landlords, like all other men, love to reap where they never sowed, and demand a rent even for its natural produce. David Ricardo, in 1817, defined economic rent precisely: the payment made for the use of land, which is limited in supply, over and above what is needed to bring it into use. Rent rises not because the landlord does more but because demand for the land grows. Henry George, in Progress and Poverty (1879), argued that rising land rents captured the gains of economic progress for landowners, and proposed taxing them.

Recent writers have applied these ideas to digital platforms. The economist Mariana Mazzucato, in The Value of Everything (2018), distinguished value creation from value extraction and argued that much of the modern economy rewards the latter. The economist and former Greek finance minister Yanis Varoufakis argued in Technofeudalism (2023) that cloud platforms extract “cloud rent” from the businesses and users who depend on them, in a relationship he compared to that of lords and vassals. These are contested, polemical arguments, and many economists reject the feudal analogy, but they name a real concern.

Where it shows up

Part of every SaaS subscription pays for real, continuing work: running servers, fixing bugs, securing data, adding features, supporting users. That is not rent in Ricardo’s sense; it is payment for service. But part of it, in a market where a vendor is dominant and switching is costly, is a payment for control: what the customer pays because they cannot easily leave. Price increases at renewal for customers whose data and processes are locked in, charges for exporting data or for features that were once included, and early termination fees are where the rent shows most clearly. In the history above, the rent was visible in the cases that reached regulators.

The strongest objection

The objection is that the rent analogy fails, because software, unlike land, can be reproduced and competed with. New vendors enter every market; customers do switch; prices for many SaaS categories have fallen. And the continuing payment funds continuing improvement, which a one-time sale did not. Smith’s landlord did nothing to the land; the SaaS vendor rebuilds its product constantly. These are strong points, and they apply to most of the SaaS market. The rent concern applies where competition is weakest: in core systems that are very costly to replace, and in categories dominated by a few vendors.

What it means today

For buyers, it helps to separate what a subscription pays for into service and control. Pay willingly for the first; negotiate hard against the second, by preserving the ability to leave. For vendors, the choice between creating value and extracting rent is made in many small decisions about pricing, packaging and cancellation, and it shapes what kind of company they become. Regulators, on both sides of the Atlantic, have begun to draw the line for them.

Bringing the threads together

Subscription software separated using a tool from owning it, and showed what ownership had protected: independence, permanence and the freedom to leave. It made Coase’s logic work in favour of renting, and Williamson’s warning about hold-up more relevant than ever. It turned software into a relationship, which aligned vendors with customers’ continued success and also gave them power over customers who could not leave. It made products into streams that never stay the same. It exploited, sometimes, the human tendency to keep paying for what we do not use. It priced companies by a future extrapolated from the present, and showed how quickly that price can change. And it raised the old question of rent: how much of what we pay is for service, and how much is for the supplier’s control.

Software you never own

When Adobe ended the boxed Creative Suite, it was making a bet about what its customers really wanted, and the bet was largely right. Most of them wanted to use the tools, not own them, and the subscription gave them continuous improvement at a lower upfront price. The same bet, across the industry, made software cheaper to start using, more reliable and faster to improve, and it let small companies use tools only large ones could once afford.

But the history also shows what was given up, and by whom. A customer who rents their tools depends on the supplier’s continued goodwill, solvency and judgement, in a relationship that the supplier has every incentive to make hard to leave. The working principles that follow are not complicated. Know which of your tools you own and which you rent. Negotiate the right to leave before you need it. Measure your subscriptions as carefully as your vendors measure you. If you sell subscriptions, earn the renewal rather than engineering it. And when the unit of value changes, as AI is now changing it, price what customers actually receive. The subscription is a promise to keep serving. The test of any subscription business is whether its customers stay because the promise is kept.

Selected bibliography

• Aristotle. Rhetoric. Fourth century BCE.
• Plutarch. Life of Theseus. First–second century CE.
• Hobbes, Thomas. De Corpore. 1655.
• Locke, John. Second Treatise of Government. 1689.
• Smith, Adam. An Inquiry into the Nature and Causes of the Wealth of Nations. 1776.
• Ricardo, David. On the Principles of Political Economy and Taxation. 1817.
• George, Henry. Progress and Poverty. 1879.
• Bobbs-Merrill Co. v. Straus, 210 U.S. 339 (1908).
• Fisher, Irving. The Theory of Interest. Macmillan, 1930.
• Keynes, John Maynard. The General Theory of Employment, Interest and Money. Macmillan, 1936.
• Coase, R. H. “The Nature of the Firm.” Economica 4, no. 16 (1937): 386–405.
• Williams, John Burr. The Theory of Investment Value. Harvard University Press, 1938.
• Levitt, Theodore. “Marketing Myopia.” Harvard Business Review, July–August 1960.
• Hirschman, Albert O. Exit, Voice, and Loyalty. Harvard University Press, 1970.
• Gates, Bill. “An Open Letter to Hobbyists.” Homebrew Computer Club Newsletter, February 1976.
• Macneil, Ian R. “Contracts: Adjustment of Long-Term Economic Relations under Classical, Neoclassical, and Relational Contract Law.” Northwestern University Law Review 72 (1978): 854–905.
• Lehman, M. M. “Programs, Life Cycles, and Laws of Software Evolution.” Proceedings of the IEEE 68, no. 9 (1980): 1060–1076.
• Levitt, Theodore. “After the Sale Is Over.” Harvard Business Review, September–October 1983.
• Williamson, Oliver E. The Economic Institutions of Capitalism. Free Press, 1985.
• Samuelson, William, and Richard Zeckhauser. “Status Quo Bias in Decision Making.” Journal of Risk and Uncertainty 1 (1988): 7–59.
• ProCD, Inc. v. Zeidenberg, 86 F.3d 1447 (7th Cir. 1996).
• Shapiro, Carl, and Hal R. Varian. Information Rules: A Strategic Guide to the Network Economy. Harvard Business School Press, 1998.
• Thaler, Richard H. “Mental Accounting Matters.” Journal of Behavioral Decision Making 12 (1999): 183–206.
• Rifkin, Jeremy. The Age of Access. Tarcher/Putnam, 2000.
• Gourville, John, and Dilip Soman. “Pricing and the Psychology of Consumption.” Harvard Business Review, September 2002.
• Carr, Nicholas G. “IT Doesn’t Matter.” Harvard Business Review, May 2003.
• Vargo, Stephen L., and Robert F. Lusch. “Evolving to a New Dominant Logic for Marketing.” Journal of Marketing 68, no. 1 (2004): 1–17.
• DellaVigna, Stefano, and Ulrike Malmendier. “Paying Not to Go to the Gym.” American Economic Review 96, no. 3 (2006): 694–719.
• Carr, Nicholas. The Big Switch: Rewiring the World, from Edison to Google. Norton, 2008.
• Vernor v. Autodesk, Inc., 621 F.3d 1102 (9th Cir. 2010).
• Mazzucato, Mariana. The Value of Everything. Allen Lane, 2018.
• Tzuo, Tien, with Gabe Weisert. Subscribed: Why the Subscription Model Will Be Your Company’s Future. Portfolio, 2018.
• Varoufakis, Yanis. Technofeudalism: What Killed Capitalism. Bodley Head, 2023.
• Garrett, Mark. “Reborn in the Cloud.” Interview, McKinsey & Company, July 2015.

Research limitations

Mark Garrett’s account of Adobe’s reasons is his own, given in an interview two years after the decision; it is a participant’s explanation, not an independent analysis. Adobe’s revenue figures for 2012 and 2013 come from its annual reports as summarised in secondary sources; the 2016 figure and the 3.4 million subscribers are reported figures. Petition signature counts changed daily and are given approximately. The dates and amounts of IBM’s unbundling, and estimates of the early software industry’s size, come from historical accounts that differ in detail. The description of Salesforce’s architecture relies on the company’s technical documentation and on interviews with Parker Harris; the account of the mock protest comes from Salesforce’s own history, which dates it to 2000. The causes of the ASPs’ failure are drawn from contemporary trade press. The Oracle acquisition prices are those widely reported at the time. The Klarna episode is described from contemporary reports of the chief executive’s remarks. The FTC case against Adobe ended in a settlement; the amounts are as described by Adobe, and the settlement involved no finding on the allegations. The size of the February 2026 sell-off varies widely by source and period measured, and is given as a range. Agentforce prices are list prices and have changed. The Rule of 40’s attribution to Brad Feld follows common usage in the industry. The philosophical connections to Aristotle, Roman law, Locke, Rifkin, Coase, Williamson, Carr, Levitt, Macneil, Vargo and Lusch, Heraclitus, Plutarch, Hobbes, Lehman, Hirschman, Shapiro and Varian, Arthur, Kahneman and Tversky, Thaler, Samuelson and Zeckhauser, DellaVigna and Malmendier, Gourville and Soman, Brignull, Fisher, Williams, Keynes, Goodhart, Smith, Ricardo, George, Mazzucato and Varoufakis are the author’s interpretation; their works are paraphrased apart from short, attributed phrases. The decisions in the decisions box are judged by the author, separating what could be known at the time (process) from what happened (call).

Footnotes

1. “Photoshop’s future: Adobe ditches $2,599 price tag for $50 subscriptions,” Christian Science Monitor, 6 May 2013; Wikipedia, “Adobe Creative Suite”. ↩

2. “Backlash begins against Adobe’s subscription-only plan,” Computerworld, May 2013; CG Channel, 16 May 2013. ↩

3. Adobe Inc., fourth-quarter and fiscal 2025 results, Exhibit 99.1, December 2025; Form 10-K for fiscal 2025. ↩

4. Computer History Museum, oral history of IBM’s unbundling task force; ACS Heritage Project, chapter 14, on industry revenue estimates. ↩

5. Bill Gates, “An Open Letter to Hobbyists,” 3 February 1976; see Wikipedia. ↩

6. ProCD, Inc. v. Zeidenberg, 86 F.3d 1447 (7th Cir. 1996); see Wikipedia. ↩

7. Vernor v. Autodesk, Inc., 621 F.3d 1102 (9th Cir. 2010); summary by Electronic Frontier Foundation and Fenwick & West. ↩

8. “ASP winners and losers,” Computerworld; “Round 1 Survivors Face An Uphill Battle,” eWeek. ↩

9. Salesforce, “The History of Salesforce”; Wikipedia, “Salesforce”. ↩ ↩²

10. Salesforce Developers, “The Force.com Multitenant Architecture”; Parker Harris quoted in The Register, 19 September 2007. ↩

11. Salesforce, Inc., Form 10-K for the fiscal year ended 31 January 2026. ↩

12. Wikipedia, “Workday, Inc.”; “Why Workday has Oracle and SAP worried,” Fortune, March 2012. ↩

13. Wikipedia, “Microsoft 365”, on the 2011 launch, the 2017 revenue milestone and the 2020 renaming. ↩

14. “Reborn in the cloud,” interview with Mark Garrett, McKinsey & Company, July 2015. ↩

15. Adobe Systems, Form 10-K for fiscal 2013; “Adobe set to pass major cloud milestone,” diginomica; PetaPixel, December 2016. ↩

16. “Adobe settlement includes $75M in ‘free services’ for some users,” FOX 13, March 2026. ↩

17. “Eighth Circuit vacates FTC’s click-to-cancel rule,” Greenberg Traurig, July 2025; Morgan Lewis. ↩

18. “Klarna shuts down Salesforce as service provider,” Seeking Alpha, 2024; “Klarna CEO ‘tremendously embarrassed’ by Salesforce fallout,” CX Today, March 2025; diginomica. ↩

19. “Salesforce’s bold new pricing strategy,” Salesforce Ben; Redress Compliance, on later credit-based pricing. ↩

20. “SaaS as we know it is dead,” Forrester, 2026; “How AI is changing software pricing,” InfoWorld. ↩

21. M. M. Lehman, “Programs, Life Cycles, and Laws of Software Evolution,” Proceedings of the IEEE 68 (1980); see Wikipedia, “Lehman’s laws of software evolution”. ↩