agents made shipping cheap, so the expensive part moved.
now it's "how fast can i turn this off" when something weird hits users at 2am. if the answer is another deploy, that's the bottleneck
@Seanwvibedev building shipd, feature flags as code.
flags and rollout rules live in the repo, so each pr shows the rollout plan per env and your agent can ship behind a flag you can turn off without a deploy. https://t.co/vnsppFgCSa
@mehdizare the one that changed reviews for me was making it call out anything user-facing that isn't behind a flag.
turned "looks fine" into "how do we turn this off"
@Am_issath "releasing became a business decision" is the cleanest way i've seen it put.
the next fight is who gets to make that call once agents are the ones deploying
@LaunchLlamaHQ building shipd, feature flags as code.
flags and rollout rules live in the repo, so your agent can ship behind a flag and you can turn it off without a deploy. https://t.co/vnsppFgCSa
@ZidaneZ08902030 2 and 3 are pretty solid, but staging doesn't have product metrics.
i'd disagree with 1 though. a branch means turning it off is a redeploy, a flag means it's a toggle. even at 3 people that matters at 2am
agents can write the code, open the pr, and deploy it.
still doesn't mean it should decide when users see it. deployment and release are two different permissions.
@yousseefashrvf building shipd, feature flags as code so agents can ship behind a flag and roll back without a deploy.
looking for early users, happy to set you up: https://t.co/vnsppFgCSa
@st3pan3k@MaxfusionAI honestly flags i can flip with or without the agent in the loop.
once merge isn't irreversible, watching the thread feels optional instead of mandatory
@thomaspark_gg always on every pr for me until it isn't.
the cheap version is treating review as a gate on risky paths only, and putting the reversible stuff behind a flag so a rubber-stamp merge isn't fatal
@Joshua_WD exactly, the rollback should be a toggle, not a deploy.
the hard part is getting agents to actually add the flag every time. i've found it only sticks when the flag lives in the repo next to the code
my agent shipped something at 2am and the only off switch was another deploy.
everyone's talking about kill switches for agents this week. the one i want first is for what the agent shipped, not the agent itself
@cleverhoods i think it's less that the agent won't read it and more that a long file stops being true.
nobody prunes it, so half the rules are stale and the agent follows those too
@theavgcoder i'd push back a little.
ship and watch only works if watching can end in turning it off fast. flags are what make "watch" an action instead of a redeploy
@g48in commit straight to main works great right up until something half-done hits users.
curious if you're gating unfinished stuff behind flags or just keeping commits small enough that it never matters?