On Tooling with Conjoined Perspectives
I like the oft-used term, "Load-bearing".
My favorite examples of ideas that were built and realized often came from the simplest sources. The opaque C code for generating of a string, or artfully rendered opaque programs that produced demoscene masterpieces in 64KB.
In modern LLM times, a prompt is a modern piece of C code equivalent: you can unpack an entire system or outcome from one sentence or more (in theory at least). Thus comes the art.
I've been working on a few ideas that I've found neat, but always thought were obvious to others.
Over the years of LLM advances, I've tried one particular idea a few times, and found it to have increasing efficacy. I wont reveal it here yet (teaser alert!), but I've been building it out into a workflow that is akin to speckit--or spec-driven-development in general--and the outcomes have been increasingly powerful to my eyes.
To tangent slightly, Claude Code with worktrees has become one of my favorite environments. I have almost achieved real flow, like the prior programming era, such that I'm thinking in terms of features and micro-PRs (supported by my team of course). Its allowed me to split off sub-ideas or minor enhancements into PRs, even as I lean in hard into refactors, major additions, and prototypes alike.
The more threads the better.
Which brings me to this particular idea. I've been calling it storytime. The simple crux of it is to have specs like a narrative that you can read like a story. Anyone can get the gist, see who made what decisions, and compile a more unified bundle of docs that actually add value to a repo, instead of rotting after the feature is merged and dispersed into the code.
I'll stop there. I'm hoping to dogfood this pretty hard soon. So in an ideal world, i'll release a Claude Code plugin plus demonstration/etc, enough to show what where and why of the thing.