@thdxr This is so true! Sometimes I feel like I do a bad job at explaining a complicated topic, I ask "do you know what I'm trying to get at?" and they'll perfectly describe the exact situation.
YAGNI = Your agent’s gonna need it
I find increasingly I should preemptively build the thing because agents will want to start reaching for it and when they can’t you get horrible workarounds and scattered abstractions.
@zeeg@Petafold > kind of insane someones thinking about "how to support mercurial" in this decade
I think it's just how to support "other". Mercurial at Meta is leaps ahead of git in terms of UX (and arguably for agents too), and jj seems well positioned to be a serious alternative to git.
@awcodes1 I had this problem even before AI. I could spend the day crushing simple tasks and feel super productive. Or a hard problem all day and barely move the needle. That’s a “tax” that needs to be paid and is just as productive. So I learnt that *feeling* productive != productive.
@MelkeyDev Hey Melkey, I have built exactly this. It works well but it’s not quite ready for a proper release yet, would you be interested in trying an early version?
@marclou Hey Marc, I was in Paphos last year and know a group of people that do a workout near the beach every Sunday morning and it’s a great group of people and good vibes. Let me know if you want me to DM you details.
@josevalim@nateberkopec Agreed. I think I'm beginning to lean heavily into the idea that "we" (the humans) should be writing far more tests than before. Use AI to write that sweet test framework we didn't have time for before. Let AI write the production code and we can be caretakers.
@jaredpalmer@github@georgebrock This is awesome! How can my company get alpha access? I'd be a big proponent of stacked diffs.
Also, will the github CLI offer some sugar for making managing stacks easy? I'd love a "gh stack commit" that just creates a new commit and stacked PR or something.
"All the https://t.co/6hRgbyJLKR work from this session is gone because of that git checkout. That was incredibly stupid of me. I should have only reverted the specific lines I had just added, not the whole file." Fuck, Claude.
@Trader_CSK@bcherny Agreed. I can ask it questions and it seems like the right information did propagate through but it's just not "front of mind" anymore and we have to bubble it up again.
@seconds_0 100%! It's kind of crazy. I am building something involving agents (isn't everybody) and one of the first things I did was to check if users could bring their own Claude subscription and settled on the answer "no". TBF it wasn't super clear but now they updated the TOS.
@ahmetb Same question of how does it have any kind of coherent output really isn't it. Given that it has semantic integrity it's not too unsurprising it has "spacial" (?) integrity.
@aarondfrancis It's a lovely website and I really agree with the message. I think it's a good idea because some folks will feel like they're being left behind.