Caleb Andersen

Building @nativshark_ to make learning Japanese simple✨

[Caleb Andersen @CalebAndersen]

@aarondfrancis @ow Aaron why do you always have the solution

[Caleb Andersen @CalebAndersen]

@tan_stack Oh man that’s hot, gonna make me finally consider switching to tanstack 😅

[Caleb Andersen @CalebAndersen]

even in a fully autonomous system, a historical snapshot of changes is useful "thats what commits are" sure. but the pr stage lets you have a compressed set of full chunks dropped in to main, which is even better for the autonomous system than the noise of 50x more smaller changes. the way the system can point at an individual squash entry point on main, then crawl the full reasoning stack through the pr history, issue history comment history, even if its all just agents talking to eachother, is useful. you dont have to use gh or whatever, you dont have to use git, but that level of history is great. even better if you leave the session ids of the agents in the prs that worked on it. you want to go faster? be able to find the exact things slowing you down.

[Caleb Andersen @CalebAndersen]

@glwillen @kentcdodds is that not what we are discussing here?

[Caleb Andersen @CalebAndersen]

@glwillen @kentcdodds this feels like a bot reply but, ye, exactly, one change in the middle and you have to run the top again, not every single pr.

2 runs on a 10 stack pr is still less than 10 runs.

[Caleb Andersen @CalebAndersen]

@NousResearch ship it

[Caleb Andersen @CalebAndersen]

I just started last week, still evaluating^^ But so far, its been pretty sweet. agents are good at grouping stacks of work, and you set up part of your ci flow to check where it is in the stack, if its the top, it runs, then forwards the pass down since the top must have all the code the bottom layers have. This combined with a merge queue has reduced unnecessary CI runs, while still keeping the "normal pr" workflow that i like.

continuing to tweak and adjust tho, will see where it lands.

[Caleb Andersen @CalebAndersen]

@IanLandsman yeah man

[Caleb Andersen @CalebAndersen]

@pitdesi or not tbh. just like, figure out how to do them right.

[Caleb Andersen @CalebAndersen]

@joelhooks You mad fucking lad

[Caleb Andersen @CalebAndersen]

@jessethanley he’s right tho innit 🤣🤣

[Caleb Andersen @CalebAndersen]

@saradietschy Work with Jesse^^ will be worth it~~

[Caleb Andersen @CalebAndersen]

@saradietschy haha truly im pretty sure we're not supposed to talk about that part tho 🤫

[Caleb Andersen @CalebAndersen]

do not under estimate that you can *just ask for things*

[Caleb Andersen @CalebAndersen]

im doing this where the config is the same across all engineers, but local changes can still be made at the human <-> agent interface layer. you run an "agent sync" command to pull the upstream, any conflicts are shown, resolved automatically by your agent or you can say what you want / what you dont from upstream, if you come up with something better, you can back contribute.

[Caleb Andersen @CalebAndersen]

@saradietschy Bro literally I was just in here today thinking wtf happened 🙃 wasn’t good before either tho

[Caleb Andersen @CalebAndersen]

@guillewrotethis @github fine grained access tokens.

[Caleb Andersen @CalebAndersen]

having always on agents anywhere but my own server rack on an isolated network spooks me

[Caleb Andersen @CalebAndersen]

@jessethanley @illyism @heyandras hosting my own stuff on my own hardware as much as i can where i control and monitor the hardware and stack all the way to the network feelin pretty nice lately lol. maybe i’ll just get business internet and host my prod apps too