My grandmother hated throwing away the boxes things came in. There was a mountain of them in the storage closet by the kitchen that threatened to topple over every time we needed something from the closet. She argued that keeping the boxes was practical. There’d come a time when we’d need to move house or sell something, and the boxes, with their perfectly molded Styrofoam would keep their contents safe and presentable.
On rare occasions, a box would get pulled out for use. There was the time a neighbor borrowed the Preethi mixer, and the box, with its yellowing cardboard and crumbling Styrofoam came in handy for transport. And the other time when the Dyson vacuum needed to be repaired and was taken in for servicing.
No doubt, the stacked-up boxes took up space. But there was no harm in keeping them. We clearly had room to spare and there were a few, albeit rare, occasions when they got used. It was hard to argue with my grandmother’s logic, as absurd as I found it.
It took me years to recognize the same instinct in myself, though my behavior is considerably less tangible.
I keep my conversations with agents, prompts and all.
Not all of them, but a large portion. Especially the ones I think might come in handy someday. Sometimes, it’s a conversation that dove deep into an idea I was exploring; other times, it’s prompt to gather work I did for clients that I copy/paste into Notion for invoicing later on.
There’s a specific peace of mind that comes with saving things. Once something has been saved, you no longer have to wonder about it until the time comes to retrieve it again. If at all.
Except, a prompt is not like a box. Or rather, it is, just not in the way we think. A box is useful because it’s perfectly molded to its contents. A mixer from 10 years ago is still a mixer 10 years later. A prompt is molded to its contents too. It’s shaped by a codebase or a project. But codebases and projects are always changing, so a prompt just as quickly goes out of date. It’s a perfectly shaped box for a thing that may no longer exist tomorrow.
But that doesn’t make it entirely useless. Past prompts and conversations tell us what we tried to do, what went wrong, and how we got to a solution. The value is in knowing how to make sense of it all.
We’ve never been particularly good at “showing our work”. We used to work and rework an idea across RFCs and branches, then squash everything in the name of a clean, legible history. Now, we’re saving everything on the off chance we or our agents will reference and use the context again. We went from preserving too little of our process to preserving more than we know what to do with.
Perhaps the real problem is that we’re treating conversations as records of work when they’re just raw material for constructing one. There needs to be a better way of turning conversations into something worth keeping around. A change log, a notebook of failed attempts or a decision tree. Maybe even another prompt to synthesize and update it as things change.
Until then, I’m keeping my boxes around, since I can afford to. Though it might be worth labelling them to start.