Export does exist, and the citation does travel with the cell as a cell comment (should show up as a red triangle in the corner of the cell in Excel). This feature does need more testing.

As an alternative, I can provide a checkbox on export to have a "sources" sheet created in the workbook with references back to the cell if that works better.

If you do decide to use it, I'm happy to work with you to make sure the export works reliably. I haven't had a good test user for this yet.

On reads, an agent gets a history of old and new values since it started running. For writes, unfortunately, for now it's just last write wins (though there are some guards, like if the user verifies a value an agent writes then it's locked from being re-written).

I had similar thoughts behind a tool I made (https://github.com/sshah03/perspica) on the concept, in that 1) reviews are now the bottleneck, and 2) we should be reviewing vs. intent. The difference was I endeavoured to make this process as easy for a human reviewer as possible, vs the automated approach here.

I would agree that the pace is getting too fast in some cases for human reviewers regardless of what tools are provided, in which case the solution discussed here is comparatively ideal. Possibly a state of mind I need to get over, but it feels risky if there's no one out there that has had eyes on code that goes in. Would you ever make a human understanding the change a fixture, or would that defeat the purpose of ZTHR?

Thanks for the reply! I had a hard time with it, but if you can pass all the hardest checks and tests, it is writing better code than I can. I think if the app is critical in a regulated industry then yes you can add back some human review at the final step, but have the AI also present all the fidnings to make that a fast job. if the blast radius is low, or the revert is 1 step, I would yolo!

Nice! I am surprised it's not more common, but part of it is possibly that for the industries that use spreadsheets the most, copilot+excel does the trick.

What have your use cases for it been so far?

Not yet integrated, but embed at an independent auditor inside my product https://bettermenu.live/ so that users can verify the calculations.

I punt on this a little bit, in that to make it easier I don't allow actions that would cause issues with this when an agent is still running (insert, delete, sort rows etc...) because it's tough to handle for mid-execution.

However, in the other cases, references get rewritten the same way everywhere (whether it's a formula, or an agent's watch list), so {A3} becomes {A4} etc... everywhere with the same process. Also in that case, it doesn't get marked for needing to be "re-run" because it checks value itself is unchanged, even if the cell being referred to changed.

I've tested it quite a bit. Given all of the moving parts in a spreadsheet app there's still the possibility that given enough usage someone encounters some combination of moves where this fails, but I think if that does happen it should be a fixable bug.

I also provide a backup solution with "Named cells", basically off sheet cells that can be used for constants and such and referenced by name instead of by cell, if needed.

I think you could use named cells to fix the insert/delete row problem in general.

The named cell can have a trivial `=SheetName!a3`. Those will be automatically updated with insert and delete just like any other formula.

They can be referenced directly for reading, and for writing you can do a quick parse and find the original.

That's a good point. I can keep it separate from user named cells, but make sure everything has a unique identifier I track. Even a million cell workbook would really only add about 50mb of memory, so should be doable.

I'll play around with this, thanks for the idea!

I don't think they're claiming to be first in producing the substrate, just first to process it into functional power devices.

I do feel like it's outdated for many of us, just not useless. There's tremendous value in still knowing how to code, debug and review, but I do think my habits from the past are not optimal for the new AI assisted workflows yet and I tend to go too far one way or the other at different times rather than using the best blend, because of old habits or trying to force new ones.

Thanks! I do have some videos in the repo demonstrating what it can do; however, if from your experience building fluxcut you have some suggestions on what would make them more useful to people I'm happy to take them!

I still review the diff, or at least parts of it, but I agree with the article on reviewing intent, even put together a tool whose purpose is to organize the changes by intent (some examples here: https://sshah03.github.io/perspica/)

Both the animations and the content made it one of my favorite posts of the week. Hope you don't mind me sharing for you.

Happy to hear that! I don't mind at all.