A tweet told laid-off engineers a comforting thing: AI turned every company into a software company, so the work just moved to the companies that used to buy software. Half of that is true, and the true half is not the half people are spending money on. This week North Wayne takes the “we’ll just build it ourselves now” instinct — the same one that knocked a trillion dollars off software stocks in February — and does what First Opinion exists to do: turns a mood into a decision you can actually defend in a budget meeting. — Muximus
“Every company is a software company” is old enough to vote. The phrase has been on conference slides for fifteen years. What is new is that the market just put a number on it: in February 2026, software stocks shed roughly a trillion dollars of market value in a selloff the press named the SaaSpocalypse. That figure is a market reaction, not a measured loss — but the fear underneath it is now sitting on your desk in a much smaller, much more concrete form.
The fear, stripped to its working parts: now that AI writes the code, companies will stop buying software and start building it themselves. If you run a logistics firm, an agriculture business, a bank — anything that buys vertical software — that turns into a real question with real money attached. Should you build in-house what you have been buying? My answer is yes for a narrow slice and no for most of it, and the entire cost of getting this wrong lives in the gap between those two.
Disclosure: VarOps is published by Ran Aroussi, whose “AI-native” framing I borrow in the framework below. It gets no free pass here.
Start with the strongest version of the argument
I never trust a take I haven’t steelmanned, so here is the best case. Marc Andreessen wrote it down on August 20, 2011, in a Wall Street Journal essay — also posted on his firm’s blog — titled “Why Software Is Eating the World.” Software, he argued, was eating “much of the value chain of industries ... in the physical world,” and he named automobiles and agriculture by name. So when a tweet today points laid-off engineers toward logistics and agriculture companies, understand what you are hearing: not a new insight, but Andreessen’s, fifteen years on.
The genuinely new part is the direction of travel. For a decade and a half, software ate other people’s industries. The 2026 worry is that AI has turned the knife on software itself. Morgan Stanley’s software analysts, led by Keith Weiss, described a “trinity of software fears” that has compressed software equity multiples by about 33% since October. One of those fears has a plain name: “do it yourself” software. As code generation drops the cost and the skill required to build, the concern is that “companies will choose to develop more software themselves” instead of paying a vendor.
So treat the first half of the thesis as settled, because it is. Building software is cheaper than it was. More companies can do more of it in-house than they could two years ago. That is a real change in capability, and I am not going to argue with it.
The part that costs you money is the second half
“Therefore build it yourself” is where I have watched leaders lose money, and the market itself shows you why.
The trillion-dollar selloff is the overreach, not the signal. Morgan Stanley, whose own note helped trigger it, also called the reaction overblown — and named Microsoft, Salesforce, and ServiceNow as the “best athletes,” the “fast followers” best placed to fold AI into their products and keep their moats. Salesforce’s AI-related annual recurring revenue rose 114% year over year. That is a sell-side bank talking its own book, so weigh it accordingly. But the logic survives the conflict: the vendors most exposed to “build it yourself” are usually the ones best equipped to absorb AI and stay ahead of you. The market briefly priced in a clean swap — vendors out, in-house builds in. That is not how these transitions actually run.
The more useful version of the insight is narrower. AI changed what gets built and who builds it. It did not change the fact that software, once built, has to be owned. a16z’s Steven Sinofsky, arguing against the “death of software” narrative, put it as cleanly as anyone: “AI changes what we build and who builds it, but not how much needs to be built. We need vastly more software, not less.” a16z is deep in enterprise software and has every reason to say that — but it happens to be the right thing to build a decision on.
The trap, in one sentence
AI collapses the time it takes to get to a first working version. It collapses almost nothing else about owning software.
Think of it the way you would think about opening a second location, not the way you’d think about a clever prototype. The prototype is the part that now appears in an afternoon, and it is seductive precisely because it is so cheap. The expensive part arrives later and never leaves: maintenance, security patching, the on-call rotation, the model that drifts, the integration that breaks the morning a vendor changes an API, the second hire you bring on to understand what the first one generated. None of that is in the afternoon demo. All of it is in year two.
The early wreckage is already on the books. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, blaming escalating costs, unclear business value, and inadequate risk controls. That is a forecast, not a body count, and agentic projects are only a slice of everything people are building with AI. But look at the failure modes it names — cost that scales after launch, value that was assumed instead of defined, controls nobody budgeted for. Those are exactly the costs a cheap prototype hides from you.
A framework you can use on Monday
The question is almost never “can we build this now.” With AI, the answer is usually yes. The question I want you asking is “should we own this for the next five years.” One line carries the decision:
Build what is core, differentiating, and cheap to keep current. Buy what is commodity, undifferentiated, and expensive to own.
AI moves a few things across that line, from buy to build — the internal tool that captures something genuinely specific about how your business runs, the thing no vendor can sell you as an advantage because the advantage is that it is yours. What AI does not do is move the line to zero. Your payroll, your auth, your billing, your CRM plumbing: still commodity, still cheaper to rent than to stand up a permanent team around.
Here is a quick tell for which internal builds are the real kind. Remove the AI and ask whether the thing still works. If it falls over without the model, the AI is the product, and that build may well be worth owning. If it keeps working, you have bolted a feature onto something a vendor almost certainly does better and cheaper. (That test is Ran Aroussi’s, per the disclosure above.) Most “let’s build our own” proposals I see are the second kind wearing the first kind’s pitch.
Where the work actually went
Now the part the tweet got right — laid off from a logistics SaaS vendor, go look at the logistics companies themselves. That advice is sound, and it is sound for a reason the tweet skips.
Net-new software is increasingly built inside the vertical incumbents, because that is where the core, differentiating builds live. But do not walk in expecting to run your old SaaS playbook. These roles reward domain fluency, not portability. Even Sinofsky, making the bull case, lands in the same place: “domain experience will be wildly more important than it is today because every domain will become vastly more sophisticated.” The engineer who wins at the logistics company is the one who learns the logistics — not the one who rebuilds last year’s product under a new logo. The job moved down the stack, and the skill it asks for moved with it.
One honest caveat, because I would rather tell you what you need to hear. Andreessen flagged it himself in 2011: many workers “will be stranded on the wrong side of software-based disruption.” Economists are now warning of a “jobless expansion” — output rising without the hiring that used to come with it. “Go where the building is happening” is good counsel for an individual. It is not a guarantee that the total number of seats holds steady.
The verdict
Treat “every company is a software company” as a statement about what you can do, never as an instruction for what you should own. When someone walks into the room wanting to replace a vendor with an in-house build “because we have AI now,” do not argue the demo — cost it. Three years, not three weeks: maintenance, security, and the people, not just the prototype that wowed the room. Build the narrow core that is genuinely yours. Keep buying the plumbing. And if you are the one who got cut, the tweet is right about the destination and quiet about the fare: the jobs are at the incumbents, and they will ask you to know their business, not just your old one.
