A huge amount of it is also the amount of training data. Agda will have little training data so the result will not be nearly as good.
I did a shoot out of making AI make the same simple desktop app from a SwiftUI reference for 30 different language & desktop framework combinations, and by far the best implementation came from the electron web typescript one. The least amount of LoC, the best and most complete implementation and the fastest to implement.
Uber has fairly large golang and java codebases so they have more of an apple to apples comparison here since they are both GC languages of a similar performance class. And if a large population tends to make more mistakes with 200hp sedan A vs 200hp sedan B, there is probably something up with the design of sedan A.
The big thing about monorepos in this case is making a change atomic across a set of projects. Multirepos can't do this 99% of the time and often add a lot more procedure to something close. It's inherently async.
> The big thing about monorepos in this case is making a change atomic across a set of projects.
The thing microservices is supposed to solve is eliminating the need to make atomic changes across services. There is an inherent overhead in doing that, like making your service support both new and old APIs as long as any service exists using the old API. This is the cost of microservices.
The world isn't just backend. There are mobile projects, games, native desktop apps which act together as a suite and web frontends. Also microservices bring up their own host of issues and extra labor. Both are essentially the semantic versioning library update problem, and when you need to propagate a breaking change it create a whole bunch more labor that is significantly reduced in a monorepo vs. a multirepo/microservice world.
The tradeoff is you need to invest into repo scaling, much like a backend service that needs to invest in scaling itself too.
IMO AI agents harnesses (claude code, codex, etc) haven't added monorepo features yet as of a couple of months ago and thats what makes them painful. Basic things like, only apply these skills to the subdir that the .agents/skills directory exists in would go a long way. Or even reading the skills in a subdir .agents/skills directory. Or the ability to specify the basic monorepo custom VCS and other actions in a way that isn't limited to fragile skills and AGENTS.md specs that can get forgotten or unused as the context windows grows in a session and so on.
It's very similar to being an EM, and if you think EMs are useless, the vast majority EMs have decided it's not worth their time to educate you, and the few who have tried to gently probably went over your head.
Funny he said "Good APIs matter more than ever". My software for one these past few weeks has been my mega archive project, making an APIs & UIs out of 50+ providers whether they like it or not. I'm calling it clump chart. I even got into my turbotax TaxML files.
I've tried designing information-dense things for colorblind coworkers, and they seemed a bit disinterested in testing it out with me. Even with tools that simulate it, you can still be off, I've found.
There can be some sensitivity about trying to figure it out with them. I've added little affordances here and there, and ironically, I rely mentally more on color coding things because I am bad at finding things in a visual field than most.
I've also found that colorblind family members and friends just never tell you and they tend to suffer in silence. Even my own half-brother (which I have a 15 year gap with) didn't tell me he was colorblind until recently.
I did a shoot out of making AI make the same simple desktop app from a SwiftUI reference for 30 different language & desktop framework combinations, and by far the best implementation came from the electron web typescript one. The least amount of LoC, the best and most complete implementation and the fastest to implement.