The Subscription Trap

He told me this in a parking lot off First Avenue, leaning against the door of a Tacoma he’d bought used in 2019. Twenty-two years at a company that makes accounting software you’ve heard of. Laid off in March. Severance was sixteen weeks.

He’s fifty-one. He said the part that surprised him was not the layoff itself but how quiet the building had gotten in the year before it, the way managers stopped asking engineers what they thought and started asking what the dashboard said. I want to write about what changed in his industry, because I think the answer is more specific than the usual story about AI eating the work. The answer has to do with how software started getting paid for around 2012, what that did to the people who built it, and what I think is coming next.

I am not a software engineer. I covered factories. But the pattern is one I have watched before in places like Beloit and Janesville, where a change in how a product was sold reorganized who got hired, what they did each day, and how long they lasted. The shift is this.

Until somewhere around 2010, most business software was sold in boxes or in licenses that worked like boxes. You paid Oracle or SAP or Autodesk a large sum once, then a smaller sum every year for maintenance. Microsoft sold Office that way. Adobe sold Creative Suite that way; CS6 came out in May 2012 and cost between $1,300 and $2,600 depending on the edition.

A year later, in May 2013, Adobe announced that CS6 would be the last boxed version. From then on, you rented Photoshop and Illustrator monthly. The Creative Cloud subscription started at $49.99 a month for the full suite, $19.99 for a single app. Adobe’s stock was around $45 the week of that announcement.

It is around $480 now. The reason every software company on earth followed Adobe was not mysterious. A one-time license is a single transaction recorded as revenue and then gone. A subscription is recurring revenue, which the public markets reward at a multiple sometimes ten times higher.

Salesforce had been preaching this since 1999. By 2015 it had become received wisdom. By 2018 there was almost no commercial software left that you could buy outright. Even Autodesk, whose customers were small architecture firms and machine shops that had paid five thousand dollars for AutoCAD and used it for a decade, forced the transition.

AutoCAD now costs $1,955 a year. If you used the same seat for ten years under the old model you paid roughly what you pay in two and a half years now. This is the structural shift. Everything downstream of it, including the career my acquaintance in the parking lot just lost, follows from it.

When a company sells you a box, the work the company has to do is build the box, ship it, and then build a better one before the next purchase cycle. The engineers who build the box are the asset. They are paid well, they are listened to, and the company plans around their judgment because their judgment determines whether the next box sells. The cycle is long.

Photoshop CS5 came out in April 2010. CS6 came out in May 2012. Two years to make something people would pay thirteen hundred dollars for. When a company sells you a subscription, the work changes.

The product no longer has to be good enough to justify a thirteen-hundred-dollar decision once every two years. It has to be sticky enough that you do not cancel this month. The metric stops being how many copies shipped and becomes net revenue retention, which is a way of measuring whether your existing customers spent more or less this quarter than they did last quarter. The engineering organization gets reorganized around this metric.

Features that drive expansion get built. Features that do not get cut, or never get started. The cycle compresses. You ship every two weeks, then every week, then continuously.

Two things happen to the engineers inside this reorganization. The first is that the unit of work becomes smaller and more legible to management. Instead of a senior person owning Photoshop’s brush engine for four years, you have a team of six owning a feature called Generative Fill for a quarter, with a product manager translating customer telemetry into a backlog. The senior person’s judgment, which used to be the asset, becomes a constraint on velocity.

The second thing is that the company learns to measure each engineer against the dashboard. Pull requests merged. Tickets closed. Story points completed.

Lines covered by tests. None of these are bad numbers on their own. The thing is, once you have them, you also have a way of comparing a fifty-one-year-old who asks hard questions in design review against a twenty-six-year-old who closes more tickets, and the comparison usually does not go the way the older person hopes. I have seen this before.

In the eighties, when General Motors started measuring its plants on Harbour Report hours-per-vehicle, the foremen who had run lines for thirty years stopped being consulted about what the line should do and started being audited about what the line was doing. A lot of them retired or were pushed out. The plants got more efficient for a while. They also lost the kind of person who could tell you why a particular weld was failing on second shift in January and not in July.

That kind of knowledge does not show up on a dashboard until it is gone, at which point you cannot get it back by hiring two new people. The subscription model did not kill the software career on its own. It made the career legible to a class of manager whose job is to run the numbers against legible things. The people who got trimmed away were the people whose value was hardest to measure on a two-week cycle.

The numbers underneath this are clearer than the layoff press releases suggest. Meta cut about twenty-one thousand jobs across late 2022 and 2023. Mark Zuckerberg called 2023 the Year of Efficiency. Google cut twelve thousand in January 2023 and continued in smaller rounds through 2024.

Microsoft cut ten thousand in January 2023, then another nine thousand in 2025. Salesforce cut roughly seven thousand in early 2023. Stripe cut about eleven hundred in late 2022. Through 2024 and into 2025 the cuts continued, but quieter, two hundred here, four hundred there, framed as restructuring rather than layoffs.

The trade site layoffs.fyi counts about a hundred and fifty thousand tech-sector job losses in 2024. The number for 2023 was over two hundred and sixty thousand. The standard explanation is that companies over-hired during the pandemic and corrected. That is true and incomplete.

The over-hiring happened because zero-interest-rate money made growth cheap, and the correction happened because rates went up. But the deeper thing is that subscription companies discovered, around 2022, that they could keep revenue flat or growing while shrinking headcount, because the revenue does not depend on the engineering output of any particular quarter. It depends on customers not canceling. And customers, especially business customers locked into multi-year contracts with switching costs, do not cancel quickly.

You can run a subscription software company on a skeleton crew for two or three years before the product erodes enough to show up in churn. By that point the executives who made the decision have moved on. This is the part that catches me. The career that ended in that parking lot was not killed by AI, though AI will accelerate what comes next.

It was killed by a financial model that made experience expensive and continuity invisible. So what does the rebuild look like. I am going to be careful here, because I have read enough trend pieces written by people who left town before the second act. I am going to stick to what I can see from where I sit, which is a college town with a lot of software people in it, some of whom drink coffee at the same place I do, and a few of whom will talk if you do not push.

The first thing I see is that the subscription model is starting to break in places, and the breaks are interesting. Figma’s pricing change in 2023 caused a small revolt among its customers. Unity’s runtime fee announcement in September 2023, which would have charged game developers per install, was so badly received that the CEO was gone within a month and the policy was substantially walked back. Adobe tried to acquire Figma for twenty billion dollars and was blocked by European regulators in late 2023.

The pattern in each case is that customers, especially professional customers with long memories, are noticing that the subscription bargain has gotten worse over time. The price goes up. The features that drove the original purchase get unbundled into higher tiers. The features you actually use get worse because the team that built them got reassigned to something the dashboard cares about more.

When customers start noticing, alternatives appear. Affinity, made by a small British company called Serif, sells a Photoshop competitor for a one-time fee of about seventy dollars. Canva acquired Affinity in March 2024 for an undisclosed sum reported to be in the hundreds of millions, which is small by software-acquisition standards but tells you the model has value. Blender, the open-source 3D tool, has gone from a hobbyist curiosity to something used by Ubisoft and Epic.

DaVinci Resolve, sold by Blackmagic Design for a one-time three hundred dollars, has taken meaningful share from Adobe Premiere in film and video post-production. None of these are going to topple the big subscription companies next quarter. They do not need to. They need to exist as evidence that the model can be different, and to keep existing long enough that the people building them get good at it.

The second thing I see is that the unit of software work is starting to decouple from the employer. There is a category of person now, mostly between thirty and forty-five, who left a large company in the last three years and is doing what used to be called consulting and is now called something vaguer. They have two or three clients. They build specific things.

They are paid between two hundred and four hundred dollars an hour. They are not on anyone’s dashboard. The clients are often mid-sized companies that cannot afford to hire a full team and have figured out that they do not need to, because the work is bounded. I know four people in Iowa City alone who fit this description.

One of them used to work at a company called Pivotal. One of them used to work at a logistics startup in Chicago. They do not call themselves freelancers, which they associate with web design in 2008. They do not call themselves consultants, which they associate with PowerPoint.

They have not landed on a name. To put it less politely, the name will be assigned to them later by someone writing a trend piece. What matters is the practice, which is that a senior engineer with twenty years of experience can now make a living by selling that experience directly to companies that need it, without the intermediation of a large employer who would rather have two cheaper people instead. This is not a new pattern.

It is what doctors and lawyers and architects have done for a long time. Software is, with a lag, professionalizing in the way those fields professionalized in the early twentieth century. The lag is because software, for forty years, had so much capital chasing it that the salaried-employee model could absorb almost everyone. That capital has thinned.

The model is shedding people. The people are figuring out other ways to work. The third thing I see, and this is where I have to slow down because I am less sure of it, is that the customers themselves are getting tired of the subscription stack. A friend who runs a small architecture office in Cedar Rapids told me last fall that he pays about fourteen thousand dollars a year in software subscriptions for a four-person firm.

AutoCAD, Revit, Bluebeam, Adobe, Microsoft, a project management tool, an accounting tool, a CRM, a backup service. He remembers when the same office’s software budget was maybe three thousand dollars a year in licenses plus a thousand in maintenance. The work the office does has not changed much. The software is, in his estimation, better in some ways and worse in others.

The cost has roughly quadrupled in a decade. What I keep coming back to is that this is a tax on small businesses that did not exist fifteen years ago, and that taxes of this kind create their own opposition over time. The opposition will not look like a software movement. It will look like accountants and architects and dentists getting together at trade conferences and asking each other what they are paying for and why.

It will look like a forty-person law firm switching from Office 365 to a self-hosted alternative because the bill got insulting. It will look like a hospital system, after the third price hike, asking a regional contractor to build them a tool that does the specific thing they need, owned by them, running on their hardware. Some of this is already happening. Most of it is not visible from the coasts because the coasts are where the subscription companies are.

If the customers start preferring tools they can buy outright, or tools they can host themselves, or tools built by a contractor they know by name, then the work comes back to people who can build those tools. Not at the scale of the 2015 hiring boom. There will not be a billion-dollar Series B for a company that sells a one-time-license CAD tool to architecture firms. But there can be a profitable thirty-person company doing it, paying its engineers well, keeping them for ten years, letting them get good at the brush engine or the constraint solver or the structural analysis.

This is the rebuild I think is possible over the next ten years. It will not look like the boom. It will look more like a trade. Smaller firms.

Longer careers. Tools you can name and price. Customers who know the people who built what they use. Engineers whose value is not legible to a dashboard because no one is asking the dashboard.

I want to be careful not to romanticize this. The trades I covered in Rockford were not paradise. The pay was worse than software pay, the bodies wore out, the benefits eroded year by year. What the trades had that software has been losing is continuity.

A guy who knew how to set up a particular kind of grinder, knew it because he had done it for twenty years, was valuable to his shop in a way the shop understood. The shop might lay him off in a downturn but it would try to get him back, because the alternative was training someone for three years. Software has been organized for the last decade as if training someone for three years was free, because the supply of people willing to be trained was enormous and the work was packaged so that no individual piece of it took three years to learn. Both of those conditions are reversing.

The supply is contracting because the layoffs and the AI tools have made the entry-level path less appealing. Computer science enrollments at large universities are starting to flatten. The work is becoming less packaged because the easy packaged work, the CRUD apps and the dashboards, is exactly what the AI tools do well. What is left for humans to do is the work that takes years to learn, the work where judgment matters, the work where you have to know why something is failing on second shift in January.

If you are fifty-one and were just laid off, none of this is comforting on the timescale of your mortgage. The honest thing to say is that the next two years are going to be hard, that the salary will probably not come back to where it was, and that the work, when you find it, will look different. The hopeful thing to say is that the difference might be in the direction of being treated like the senior person you actually are. Not by the company that just let you go.

By a smaller customer who needs exactly what you know, who is willing to pay for it, and who is not running a dashboard. The guy in the parking lot did not ask me for hope. He asked me whether I thought the building he had worked in for twenty-two years would still be there in five years. I told him I did not know.

The company is profitable. The subscription revenue is steady. The product is fine. There is no obvious reason for the building to close, and also no obvious reason for the people who used to fill it to come back.

Buildings like that, in my experience, sit half-occupied for a long time before anyone admits what has happened. He nodded. He said his daughter was looking at colleges and had asked him last week whether she should study computer science. He told her to study whatever she wanted, and to learn something else on the side that she could do with her hands.

He said it without irony. We stood there for another minute. Then he got in the truck and drove off, and I walked back along the river the way I had come.