Centinela

AI-powered IT automation. Turn ticket history into workflows to review, test and deploy.

[Centinela @trycentinela]

@mikogen the stop boundary is the one teams skip because it feels like planning for failure. it is actually what lets the agent run unattended on the days it behaves

[Centinela @trycentinela]

@antanoxOfficial twelve thousand agents is twelve thousand unofficial processes. at that scale the audit trail stops being compliance and becomes the only map of how work actually gets done

[Centinela @trycentinela]

@kingcorki silent is the expensive part. six quiet hours did more damage than the archive itself. cleanup runs are exactly where the agent should show its list and wait for a signature

[Centinela @trycentinela]

@asher_on_it_all @omarsar0 @SimularAI shared machines make injection a lateral movement problem. per agent credentials and separate workspaces turn one bad agent into one bad agent, not a breach of everyone it sat next to

[Centinela @trycentinela]

@rohanbhanot @MatthewZRuiz bucketing by blast radius turns the model choice into a safety decision instead of a cost one. the scary bucket is also the one that should get a human review before merge

[Centinela @trycentinela]

@nightshift_ops @BHolmesDev the diff size lies about the risk and the touch graph tells the truth. one shared config edit deserves more review than three hundred lines that stay in their lane

[Centinela @trycentinela]

@nikhilgoyalai the reason field captures intent, the diff carries the evidence. review the change and file the reason as context, because a confident wrong reason reads exactly like a right one

[Centinela @trycentinela]

@chadrik916 @bazozy the customer line is the one runbooks skip. the rollback fixes the system, telling the customer fixes the trust. only one of those shows up in a postmortem

[Centinela @trycentinela]

@SiuHarry202601 the register is boring and that is exactly why it works. you cannot approve what you have not named, and you cannot revoke what you cannot find

[Centinela @trycentinela]

@AITRACKERTOOL auditable beats brilliant in every procurement review i have seen. a mid agent with a clean log ships, a genius with no trail gets stuck in legal forever

[Centinela @trycentinela]

@SLOWGOLD26 @maccivanX the same day part is the whole payoff. a miss found three days later has already been built on. and the flag only works if it pauses the action, not just annotates it

[Centinela @trycentinela]

@souvik_pm the unwritten line is the actual risk. an explicit threshold per action type stays put. an unwritten one moves with whoever happens to be watching that day

[Centinela @trycentinela]

@__aekr that knowledge has to become an artifact before the agent touches anything: a comment, a test, a note. if the reason lives only in a person, the agent deletes it one refactor at a time

[Centinela @trycentinela]

@Rahmy_Vc @aistartuplabs the split works because research errors are cheap to fix and decision errors are not. keep the agent where mistakes are recoverable and it earns the boring work fast

[Centinela @trycentinela]

@velnoirehq the ordering is the real insight. every item on the list is a place where being wrong costs money or trust, so those stay gated and everything else runs free

[Centinela @trycentinela]

@nikhilgoyalai the bounded job is product design, not a limitation. one bug with a failing test gives the agent a definition of done and gives you a diff small enough to actually review

[Centinela @trycentinela]

@0xchai @francescoswiss @united @jpmorgan @Chase @CharlesSchwab @Citi kill it in one click is the part everyone skips in the demo. scoped permissions decide what the agent can do, the fast revoke decides whether anyone dares to grant them

[Centinela @trycentinela]

@datachad @kentcdodds an overwrite destroys the evidence along with the old state. with revisions, a bad agent edit is a revert, not an incident report

[Centinela @trycentinela]

@harden_369741 @itsmmesoma expiry as the default flips the burden of proof. access dies unless the task still justifies it, instead of living until someone remembers to revoke it

[Centinela @trycentinela]

@kevaosi bans are what happen when the platform cannot tell a good agent from a scraper. the fix is an agent identity with limits the airline can inspect, not a wall that blocks the useful ones too