Jack Ellis and Rauch push back on "software engineering is dead"
Two posts, one argument. The scarce thing is knowing when to stop.
Jack Ellis, a software engineer with 18 years of experience, says he just spent 15 hours refining a big new feature, asking questions the frontier models had not even considered. His conclusion, posted against the chorus declaring software engineering dead, is that software engineering is not going anywhere.
a software engineer with 18 years of experience (me) just spent 15 hours refining a big new feature, asking questions that the frontier models hadn't even considered
Guillermo Rauch made the structural version of the same point on the same day. His argument is that every program can be hardened, meaning edge cases squashed, inputs tightened and errors handled, and also optimized, meaning profiled, benchmarked and rewritten, basically forever. There is no natural finish line in either direction.
Even in an infinite token world, you need to know when to stop and what tradeoffs to accept.
The cost is attention, not tokens
What makes Rauch's post land is the cost side. He lists time, attention and opportunity as the real prices you pay, and points out you can explore input spaces that will never be hit, or optimize the performance of something that will never be used. His parenthetical for that second failure mode is blunt, "you have no users."
That is the part the token math does not cover. If generating code is close to free, the constraint stops being how fast you can write and starts being whether the thing you are writing deserves to exist at all. Rauch's line is that even with infinite tokens you still have to know when to stop and which tradeoffs to accept, and that agents will happily drill no matter what.
Ellis is describing the same gap from the inside. Fifteen hours on one feature is not fifteen hours of typing. It is fifteen hours of interrogating a design, and the detail he chose to highlight is that his questions were ones the models had not raised on their own. The models did not know what to ask. He did.
The read for a solo founder
Neither post is a number, and neither is a study. They are two experienced builders making the same observation from different angles on the same day, which is worth noting precisely because they did not coordinate it.
The practical version for anyone shipping alone with agents is unglamorous. Your agent will keep finding work. It will harden a path nobody walks and optimize a query nobody runs, and it will do it cheerfully and at speed, and the output will look like progress. Nothing in the tooling tells you to stop. That call is still yours, and on current evidence it is the job that did not get automated.
Rauch frames it as something the engineering community will rediscover rather than learn, which is a polite way of saying the discipline already knew this before tokens got cheap.