Our experience is very similar except we didn't really have a review process before, and now LLMs find bugs before PRs get merged in main.
We had 5x-100x speedups in some legacy but important pipelines, with no regressions (validated after extensively by humans).
It's not that the code was actively bad. It's just only 1-5% people in the local SWE market would be able to write code that runs so fast and efficient and benchmark it correctly.
We found a subtle correctness bug that was in production for half of the decade (both GPT-5 and Claude Opus were able to find it), confirmed by human after.
And we keep finding subtle bugs that have been introduced by humans before (despite the human reviews, the particular domain is just difficult no matter how many docs and comments and tests one writes)
They need to figure out ISRU for Starship methane refueling on Mars, but I don't think it will be ready for the cargo launches. Probably can expect it to be deployed before the first Mars crew launches though
> It seems we need to find a completely new means of propulsion (or offsetting with carbon capture) before we could have a green space industry.
Space elevators would be a cool solution, but they are in the realm of science fiction for now. So deploying at least some carbon capture solutions before space industry begins to scale actively is a good idea
Financially and ecologically this is not worse than manufacturing products in China and shipping them all over the world, or usual stuff like tourism, cloud computing etc.
I would argue that Starship is even more ecological than the other rockets, as it requires a cleaner fuel than Falcon (methane vs kerosene) and should be fully reusable (i.e. no tossing the rocket in the ocean and hoping it will sink).
If we manage to bootstrap space manufacturing, benefits for ecology, biology and material science are also self-evident.
Our experience is very similar except we didn't really have a review process before, and now LLMs find bugs before PRs get merged in main.
We had 5x-100x speedups in some legacy but important pipelines, with no regressions (validated after extensively by humans). It's not that the code was actively bad. It's just only 1-5% people in the local SWE market would be able to write code that runs so fast and efficient and benchmark it correctly.
We found a subtle correctness bug that was in production for half of the decade (both GPT-5 and Claude Opus were able to find it), confirmed by human after.
And we keep finding subtle bugs that have been introduced by humans before (despite the human reviews, the particular domain is just difficult no matter how many docs and comments and tests one writes)