A pattern for data-aware artifacts, adopting the three-bucket pattern from my earlier Doorbell in the Jungle:

1) A bucket for active artifacts.

These are artifacts that you have proactively chosen to have active.

They can show you if refreshing them would pull in new data from upstream sources.

This allows you to preview updates without actually cascading them through the system, and possibly squishing out in some shared downstream artifact.

But they don't update by default unless you tell them to, preventing the chilling effect fear of "what happens if an artifact I shared updated in the future in some indirect way to share something embarrassing"

2) A bucket for archived artifacts.

Artifacts that you used to run, but aren't currently running.

These are artifacts that you've "deactivated".

Makes it less scary to deactivate an artifact, because it's easy to find it again.

3) A bucket for suggested artifacts.

These are artifacts the system thinks you might find valuable.

They might be shown off to the right: easy to ignore if they aren't a good match, but easy to see in your peripheral vision.

These artifacts would show a preview of what they'd look like applied to your data.

If you like one, you can "pin" it by moving it to your bucket of active artifacts.

Pinning is an extremely high intent act, an application of human judgment of quality that can be used to improve the quality of the overall system.

An artifact that another user decided to pin, all else equal, is way more likely to be useful to another user too than a random hallucinated artifact.

The self-steering metric to optimize is "maximize the absolute number of suggested artifacts that a given user chooses to pin".

Because you can anonymously aggregate preferences of many users, this quality curve has a network effect; the quality gets stronger the more users use the system.

At first, the suggestions won't be particularly good and people will actively ignore the suggestions bucket.

But over time as the quality improves, the suggestions bucket will be how more and more users get their tasks done.

A secondary use case that could theoretically grow to overshadow the primary use case of recipes conjured up on demand.

This is the differentiated upside of the system.