We're bringing "set it and forget it" to AI analysis.
Quick question for this community before we launch: how do you currently keep tabs on your numbers day to day?
Most people we talked to do one of two things. Either they open the same dashboard every morning out of habit, even when nothing's changed. Or they don't check often enough, and by the time they notice something's off, it's already cost them a few days.
We built Routines to fix that. You tell our AI analyst, Genie, what to check and how often. From then on, it runs on its own, no re-prompting, and tells you what actually changed, good or bad.
We're launching it here on Product Hunt on September 4.
Curious to hear from this community: if you could put one recurring check on autopilot right now, what would it be? Drop it in the comments, might end up being exactly the example we show on launch day.
Replies
I think this could save a lot of time on routine dashboard checks. It’s nice to focus on changes instead of checking numbers constantly.
Databox
@oliver_graf1 Exactly the shift we're going for. Checking numbers "just in case" eats more time than people realize, and most of the time there's nothing to see. Would rather Genie tell you the one week something actually moved. What's the routine dashboard check that eats the most of your time right now?
Cost per finished piece of work, not cost per generation. We can see GPU spend and we can see revenue, but the number that matters is how much compute someone burns to get one thing they'd actually send to a client. It drifts when a model changes underneath us and nobody notices for a week. That's the check I'd want shouting at me, because it's exactly the kind I'd never remember to open a dashboard for.
Databox
@asadmalik901 That's a great example, and it's exactly the kind of check that gets skipped because nobody remembers to open a dashboard for it. Routines won't catch a silent model swap on its own, but "cost per finished output" as a weekly instruction to Genie would work well here. It reads the numbers, tells you what moved and why, so a drift shows up in the notification instead of sitting quiet in a chart for a week.
The check I would put on autopilot is direction, not level.
A sign amber for six weeks and one that went green, green, amber, amber over the same six weeks produce the identical marker today. Nothing in the format holds the second fact, so reading the trajectory is left to whoever remembers the last four updates, which in practice is nobody.
On the run I checked, eight of eight signs were worsening and not one of them was red.
Databox
@cmumulle This is the exact failure mode we built the notification subject around. A single reading always looks fine on its own, it's the trend across readings that tells the real story, and that's easy to lose when someone has to hold four weeks of updates in their head. Genie writes the subject line based on the trajectory, not just today's number, so "amber, amber, amber, still amber" would actually get flagged instead of blending in.