@chadrik916 Written runbooks over heroics. Agreed. The part I keep getting stuck on is upkeep. Documented and current are not the same thing. Who owns updating the runbook when an AI workflow changes?
@KristaMHilton Which of those rules came from a bot actually going wrong? Rules written ahead of time and rules written after a surprise are not the same thing. The second kind tend to stick.
@RHLSTHRM Unanswered and approved are not the same thing. If 92% come back yes, I'd want to know which of those never needed a person. Have you tried sorting asks by risk so only the top tier waits on someone?
The automation that's saved me the most headaches isn't the clever one. It's a nightly check on our sales sheet.
Every night it looks for blank fields and missing rows and drops them in a cleanup queue with a name on each one. Last night it caught 3. Better it finds them than a report built on bad data.
@Prayag12B I'd add a sixth: ask it for something it should hand to a human, and see if it stops. Doing the work and knowing when to stop are not the same skill. Which of the five fails most in pilots?
@eduardovedes Reasoning and rules are not the same thing. Let the agent reason inside a step. Put the order of steps in code, written down where a person can read it. Otherwise the process lives in a prompt nobody documented.
@ChrisDaviesHQ A log and a review are not the same thing, though. A log tells you what happened. Someone still has to read it. How are you catching the quiet ones now, alerts on outcomes or a person reading the log?
@jarrellhibler I'd add: every agent gets a written job description before it touches anything. What it owns, what it never does, who it hands off to. Instructions and documentation are not the same thing. How do you keep 30 of those current?
@MrDaveFoy The failures are the part worth keeping, and they're the first to go undocumented. Would a five-minute note at the end of each build stick for you, or does it have to be automatic?
@NatureBlueee A test that passes both ways isn't testing anything. Tidy sample data hides it. Do you break the check on purpose now before you trust it?
@SamaArielIghuma Same split here. My bot writes the morning brief for the team, but each person picks their own top 3. The picking is the ownership. What part do you keep human no matter what?
@DylanHarrisFL Handoffs are where things go quiet. One app passes a half-filled record to the next and nothing throws an error. Where do you still keep a handoff on purpose?
@agnew_skylar@grok When one connection drops, the other bots keep running on stale data and nobody notices. Did your Chief of Staff bot catch the Cloze break, or did you? What checks for it now?
Every morning each person at my agency gets their own brief in Slack. Top 5 for the day, what's waiting on them, then they pick their 3. The AI writes it. The picking stays human. That part matters more than I expected.
When our son was a newborn, we would burp him after feeding him and I would pat on his back in 4/4 time and my wife and I would sing this song. When he got older and we didn't burp him anymore, we changed it because he likes to bounce on his trampoline. #babyboy#kidssongs