The software industry is writing two obituaries at once.

One is for SaaS. The agents are coming for the application layer, the per-seat model looks finished, and the public markets have already flinched. The other is for open source: the models trained themselves on the commons, and now there’s nothing left to make the commons regrow.

My instinct is that both are overblown. I distrust the instinct, because I’ve spent the better part of a decade as exactly the kind of builder who needs both of them to be wrong. But the longer I sit with the two obituaries, the more they look like they’re describing the same death. And it isn’t the death anyone seems to be mourning.

Start with the strongest version of each, because the weak versions are easy to beat and dishonest to enjoy beating. Late in 2024, on a podcast with the investors Bill Gurley and Brad Gerstner, Satya Nadella described business applications as “essentially CRUD databases with a bunch of business logic,” and predicted the logic was “all going to these agents” until the back ends themselves got replaced. He never said SaaS is dead; the headlines said that for him. What he actually said is more precise and more threatening: the part you pay for is the part that’s easiest to move.

Sequoia’s Julien Bek put the business end of it cleanly. Sell the tool and you’re in a race against the model; sell the work and every model improvement makes your service cheaper and harder to beat. The next great software company, he wrote , is “a software company masquerading as a services firm.” And notice what he concedes in the same breath as the threat: gross margins go from ninety percent to seventy. That is not a death notice. It’s a repricing, and worth remembering the next time you watch a basket of public software companies shed a fifth of its value and hear it called a massacre.

The open-source obituary cuts closer, because its bodies are specific people. Daniel Stenberg has maintained curl for nearly thirty years — the small data-transfer library that runs in basically every connected device you own — and last year he shut the project’s bug bounty because AI-generated vulnerability reports had made it unworkable. “We are effectively being DDoSed,” he said . The project “has never received a single valid bug report that was generated using AI.” Beneath the slop runs a quieter argument the lawyers are having: copyleft, the license mechanism that forced anyone building on open code to give their own changes back, doesn’t trigger when a model launders the provenance one token at a time. The legal engine that made the commons self-sustaining may simply not apply to a machine that has already read all of it, and a US appeals court is, as I write this, deciding a version of that question.

I won’t argue with any of this directly, because it’s true enough. The margins are compressing. The maintainers are drowning. The legal scaffolding is shakier than it looked. The doom is overblown only if you read it as wipeout, which nobody serious is doing. They’re describing erosion, and the erosion is real.

The same knife cuts the optimists, though, including the optimist I usually am. The reassuring story is that this is just creative destruction: the bloated incumbents lose their fat margins, the lean and fast and cheap builders inherit the ground, and more gets built by smaller teams. I want to believe it. The problem is the weapon. What destroys the incumbent’s margin is that the thing they sell got trivial to reproduce, and that weapon does not check the size of the company on its way in. If a model can clone what Salesforce sells, it can clone what the scrappy upstart sells too, the morning after they ship it.

I wrote about Tailwind once before , and it’s the example I can’t put down. Tailwind CSS was not a bloated incumbent. It was the archetype the optimists are betting on — small, beloved, technically excellent, a tool that genuinely took over its corner of the world. Usage kept climbing. The business collapsed anyway, revenue down nearly eighty percent and three-quarters of the team gone, because AI made it trivial to get what Tailwind sold without passing through the part that paid for it. In the one story everyone reaches for, the lean small builder is not the survivor. The lean small builder is the corpse.

The optimists do have a real rebuttal, and it belongs to Jensen Huang. In February, on the morning of Nvidia’s earnings call, he told CNBC the markets had got it wrong — that agents won’t replace the tools, they’ll use the tools, and use far more of them than any human ever did. He’s almost certainly right about the usage. But usage was never the question. An agent hammering your software a thousand times an hour is not an agent buying a thousand seats. The tool can win on every metric except the one the business model was actually built on.

So if the software itself was never really the moat — and after Tailwind I don’t think it was — what was? My answer is captivity. The fat SaaS margin was a rent, and the rent got paid for two reasons: most buyers couldn’t realistically build the thing themselves, and once they had bought it, leaving was made expensive on purpose. AI is dissolving the first reason. Building is no longer the wall it used to be, which turns “make versus buy” back into a real decision instead of a foregone one. A rent that depends on the buyer having no alternative does not survive the buyer getting one.

There’s a fair objection here, and it’s the one I find hardest to wave off. Building was only ever one of the two locks. The other is switching costs: the data inside the system, the integrations wired into everything around it, the compliance certifications, the year procurement took to sign off. AI does not obviously pick that lock. But look at what those costs protect. They hold the customers already inside. They do nothing to win the next buyer, who can now build instead. And the lock itself is mostly the same grunt work AI keeps getting better at — migrations, integration shims, the plumbing that made leaving expensive in the first place. So the captivity rent does not vanish everywhere at once. It retreats to where the switching costs run deepest: regulated data, decade-old integrations, the genuinely large enterprise. And it thins out everywhere else, fastest at the edge where the small teams and the side projects and the greenfield builds live.

And where it does drain out, look at what’s left. Not a captured rent that the small builder pockets in the incumbent’s place. A wage. If I can stand up in a weekend what a fifty-person company used to sell — and for a growing list of things, I can — then I cannot charge the fifty-person company’s price for it. Neither can the next person who can do the same. We compete it down toward the cost of doing the work plus a living, and there it stops. The rent doesn’t change hands. It gets competed out of existence, and what survives on the other side is a salary.

This is not the same as claiming nobody charges for software anymore. Linear sells seats and grows, because the product is genuinely better and not because anyone is trapped. Aiven runs a real business hosting open-source databases its customers could, in theory, run themselves. What dies isn’t the price. It’s the portion of the price that was a toll. Quality still gets paid for. Toll collection stops.

I didn’t see the next part until I’d followed this all the way down. The world I have just described — software made and operated close to cost by people who needed it anyway, charging a wage instead of a rent — already exists, and it has a name. It’s open source. That is what open source always was: software produced for reasons other than margin, by people whose living came from somewhere adjacent to it. The “end of SaaS” I keep reading about isn’t software disappearing. It’s software being pulled down into the economics open source has been running on for forty years. Aaron Stannard, whose essay started me down this whole road, argued that AI would amplify open source rather than kill it. I think he was more right than he knew. The thing AI amplifies is the mode of production itself, and it shoves that up the stack into the territory SaaS used to own.

The end of SaaS and the end of open source are the same event, seen from opposite ends.

The rent economy of software is collapsing into the wage economy that was running beside it the whole time.

I’ve spent years building and operating real production infrastructure mostly by myself — keeping systems running that a lot of people quietly depended on, at a scale that was supposed to need a team — back before any of these tools existed to make it easy. So the claim that one person can’t do this doesn’t worry me. I know it’s wrong, because I spent a long time doing it the hard way. Which is also the reason to discount what I’m telling you. I want this future to be real, and I’ve built my working life around the bet that it is.

Underneath everything, the argument balances on a single figure: the person who builds software for a reason other than margin, because they need it and they care how it works. Everything good in my version depends on that person being more durable now, not less. I think they are. But the open-source obituary is, under all the licensing and the slop, a story about that same person breaking. Stenberg is that person. The maintainer buried under generated bug reports is that person. I’m betting on their resilience on one side of the page while the other side watches them buckle.

And the version I find exciting carries a cost it would rather not say plainly. “One person can now do what fifty did” is the same sentence as “forty-nine people don’t.” Cheap, honest, independent software, built by whoever needs it — from almost any other seat, a contraction. More gets built. Fewer people are paid to build it. Whether that adds up to something better than what it replaced, I doubt anyone knows yet, least of all the people it has rewarded so far.

So it comes down to one person: whoever keeps doing the necessary, unprofitable work for less than it’s worth, and whether they hold up or wear down. My whole argument is a bet that they hold up. I still don’t know if they will.