Let us speak of the dead.
Software 1.0 was born sometime in the late 1960s, depending on who you ask. Dijkstra had opinions. So did Knuth. They disagreed on the details but agreed on the broad strokes: a human thinks about a problem, a human writes code to solve the problem, a different human reads the code to check it, and then everyone argues about formatting until the heat death of the universe.
It was a good run. Fifty-some years. We got operating systems out of it. We got the internet. We got a lot of COBOL that nobody can maintain and everyone is afraid to turn off.
"It was always a compromise," Justin said. He was standing at the kitchen counter in the San Mateo office, making pour-over coffee with the focus of a man defusing ordnance. Jay sat on a stool across from him, laptop open but untouched. "We compromised on coverage because exhaustive testing was too expensive. We compromised on review quality because thorough review was too slow. We compromised on architecture because refactoring was too risky. Every decision in Software 1.0 was a triage decision."
"That's not how people talk about it," Jay said. "People talk about craftsmanship. Artisanal code."
"People talk about artisanal bread, too." Justin poured the water in a slow spiral. "That doesn't mean it's the best way to feed a city."
Jay thought about his years writing Go at previous companies. The code reviews. The style guides. The arguments about error handling conventions that consumed entire afternoons. He'd loved it, honestly. There was a satisfaction to a clean pull request, to a review that caught a subtle bug, to the rhythm of human collaboration over a shared codebase.
But he also remembered the other part. The PRs that sat unreviewed for days. The tests that nobody updated. The technical debt that accumulated like sediment. The on-call rotations where you paged through code at 3 AM written by someone who'd left the company two years ago.
"The thing about compromise," Justin said, testing the coffee temperature, "is that you stop noticing it. You think the constraints are inherent. You think slow review cycles are physics. You think incomplete test coverage is the natural state of software. And then one day the constraint lifts and you realize—oh. We didn't have to live like that."
"So what lifted?"
"October 2024. Claude 3.5, second revision. Long-horizon agentic coding workflows that compounded correctness instead of error. That was the moment the compromise became optional."
Justin handed Jay a cup. The coffee was precise, like everything Justin did.
"I'm not saying Software 1.0 was bad," Justin said. "I'm saying it was the best we could do with humans as the bottleneck. Now the bottleneck is gone. Deliberate naivete. Throw out the old habits. Don't optimize the old process—replace it."
Jay sipped the coffee. It was excellent.
"So this is the eulogy," he said.
"This is the eulogy," Justin agreed. "Software 1.0 served us well. It is survived by its legacy codebases, its style guides, and a generation of engineers who will spend the next decade unlearning its habits."
They clinked cups. It felt appropriately ceremonial.
The COBOL line got me. I literally maintain a COBOL system that everyone is afraid to turn off. It processes payroll for a state government. Nobody knows how it works. We are terrified.