@leerob I would call value engineers instead, builder reflecting the work rather than outcome.
The specific job title should have changed though, and that's not related to simplication
@NateWatkin I think the problem is the opinion onto that aesthetic is too strong, while it makes most of things it touches quite pleasing, it's quite hard to steer that away off that art style.
but, if you use to visualize some brainstorming concept, chefkiss right off
is this me or codex gpt 5.5 is really eager to make change, a lot of time without confirmation.
Good
- really strictly follow overall instruction
Not good
- quite eager to finish the task, and a lot of time asking, do you want me to
well, dude...
not sure if that's better or worse. Code is code, test is test, and no matter if you write, or LLM write, the quality is just as strong as what you instructed/and reviewed. (hint: different level of engineer quality)
you are just as strong as your weakest link, that applies to both, human written or LLM written
@CasJam have to agree, while I also spend quite sometime with Codex; reasoning, planning still prefer Claude Code over.
Codex can surely be used for review (adversity review) and code review, do coding is fine, but there's surely some shortcoming here and there
I main windows, with WSL, and Windows Terminal, what doesn't work for you?
My setup tho
- everything is in Debian (codebased, git etc)
- Docker in Debian
- claude and codex in debian
- only VScode in windows, but I nolonger use Vscode
I used mcp on windows host, it works
also expose claude code as mcp to use in Claude desktop, kinda work but not my usecase
best thing, that you can use claude code and point that, you are in wsl, feel free to use Windows app, it works brilliantly
@trashh_dev golang more likely, easier to be traced-back, vendor-easy, strict enough compiler, fast iteration loop with test
typecheck in ts can be bypassed in a lot of cases, and that makes code generation to be more of defensive programmed
I'm just a novice user of Claude Code, but I pay the Max plan with surely an assumed expectation of predictability, nobody wants to pay to the max and don't get what's they assumed to receive.
Why? Because the model is definitely renting time, doesn't matter if it's reflecting to token usage or not, it is at the end of the day, within my reach, I expect to use the service within the commited SLA.
Those recent incidents defintely impact the SLA. As such, would love to have a more official support from @AnthropicAI regarding the SLA
- Service down -> less time-leased
- Cache-missed -> less time-leased
- Token cost faster -> less time-leased
While I'm a big fan of Claude Code, I do expect this to be better, the execuse should not really be, it's hard to keep up with the load, sure, but then don't do it on a ground of false promise.
@trq212@bcherny (sorry you guys, just because I don't know anyone else of the company :D)