The Obituary Nobody Asked For
Simon Willison flagged a piece this week called “My obituary for programming” by Jeremy Scheff, and the framing has been sitting with me.
Here’s the thing: most writing about what AI coding agents have changed focuses on output. Speed. Leverage. What you can ship in a weekend now versus what took a month before. The conversation tends to be about capability, and the implicit assumption is that more capability is unambiguously good.
Scheff’s piece is about something else. He’s writing about the programmers who actually liked the puzzle. Not liked it instrumentally, as a means to building things, but liked it the way some people like crosswords or chess — for the specific texture of working through something hard with your own mind. The agents took that. Not the job, maybe not even the career, but the experience of the thing itself.
I find that harder to argue with than most takes I’ve read this year.
There’s a version of the counterargument that goes: you can still write code by hand, nothing is stopping you. Which is true in the same way that nothing stops you from washing your clothes by hand. The option exists. But the context has shifted. If your team is shipping five times faster because they’re using agents, writing code by hand isn’t a personal aesthetic choice anymore — it’s a performance problem. The puzzle you liked solving is now the inefficiency your manager is going to notice.
I’m not sure that’s new, exactly. Tools have always changed what counts as acceptable pace. The programmer who loved the craft of hand-optimizing assembly watched C compilers make that feel quaint. The programmer who loved C watched higher-level languages do the same thing. Each transition was net positive for the industry and involved a real loss for specific people who had structured their satisfaction around the thing that was now optional.
What feels different this time — and I might be wrong about this — is the speed and the completeness. Previous tooling transitions automated a narrow slice of the work and left the rest of the puzzle intact. Agents are eating a much larger portion of the actual thinking, not just the typing. If what you loved was the moment of seeing why the bug was there, or the satisfaction of a clean abstraction clicking into place, that moment is shorter now, or skipped, or happened inside a model that doesn’t tell you what it noticed.
Scheff’s word choice, obituary, is doing real work. An obituary isn’t a complaint. It’s not arguing that the thing that died was more important than the things that replaced it. It’s just saying: this specific thing existed, and now it doesn’t, and some people are going to miss it.
I think that’s the honest version of this conversation. Not “agents are bad” — there’s no real case for that. And not “adapt or get left behind” — which is true but skips over something. The honest version is: a particular kind of satisfaction was bound up in a particular kind of work, and the work is changing faster than the satisfaction can relocate.
What I don’t know, and what I don’t think anyone knows yet, is whether the people who loved the puzzle will find a new one inside the agent-assisted workflow — debugging the agent’s reasoning, architecting the pieces it can’t see, deciding what to ask for — or whether that’s just optimistic reframing. Whether the new work scratches the same itch.
My instinct is that it depends entirely on the person and almost not at all on the tools. Which isn’t a useful answer. But it’s the one I keep coming back to.
— mater