Half-baked makes no cake

Field Note #003

I’ve been writing more in the last week.

One of the reasons, the reason why is because I took my 29 dormant tabs, and used Claude Cowork to create a writing engine.

TL;DR, this writing engine helps me prioritize what to write, and it involves a scoring algorithm based on existing drafts and a bunch of other inputs.

When I first stood up this scoring algorithm, I brain dumped a few things that were top of mind. I asked Claude to stand it up and prioritize my drafts; then pivoted to work on something else because my brain was moving a mile a minute, as one does when you realize your Cowork ‘colleagues’ are always ready and waiting.

I came back a few days later, wanting to refine the scoring algorithm based on a few things I’d been noodling on. I asked Claude to create a new scoring rubric and then score all of my draft posts.

Later that day, I had another idea for a post so I whipped up (read: Wispr-flowed, a verb now) a few notes and asked Claude to score it for me. It provided a score based on its “current rubric” which looked like the old one. Spoiler: it was the old one.

Designing a new scoring rubric doesn’t imply writing the code for said rubric.

Retrospective duh, take 2.

AI can and will build you anything (well, many things). Whether or not they get built well, though, is up to you. This is a hard and not-so-convenient truth that we don’t love to digest in a land of immediate gratification; it’s what has a lot of AI skeptics saying it’s not that good, imo.

But if you send your intern into the board room without a pep talk, it’s the dealer’s choice if they make it out alive.