Dan Ubilla retweeted
*deep breath*
people think I'm joking when I tell them that to develop taste for software they need to have brunch with their friends, watch old movies and find old records, walk around metropolitan cities and observe how the building styles lend to the personality of the people around them, but I'm so fkin dead serious. where else is this mythical "taste" going to come from? what are you doing to parse the human condition? mf you can barely decide what discomfort you're subjecting yourself to in finding out what you like/dislike, and I'm supposed to trust you to make decisions for millions of people? go do some pottery and come back to me and describe the joy you felt without faffing about curves or whatever, I need you to feel the clay under your fingernails and fall in love with mud. even steve had the decency to drop acid on a sweaty beach in goa to find himself, the least you can do is listen to your friends talk about their lives over a meal without getting distracted by your phone.
*exhale*
Dan Ubilla retweeted
I’ve noticed my interactions with agents are getting more and more conversational.
I still say “Good job, continue” a lot.
But two phrases are showing up more and more:
“What do you think?” I explain my concerns, then ask for their opinion before moving forward.
And “Don’t touch the code, let’s think about this together first.” I want them to discuss it with their teammates, come back with their thoughts, and we figure out the best solution as a team before they implement anything.
It’s shifting from “do this” to “let’s figure this out together.”
Probably the scariest thing about Bret Easton Ellis’ The Shards is how he makes mixtapes with multiple songs from the same album.
Few joys in life compare to finding something, holding it above your head, and humming the little Zelda item discovery jingle.
Would love to be able to flip from Fable to Opus or Sonnet without needing to worry about existing prompt caching or context.
I built a small weekly review agent to summarize my week. It takes input from a few specialist outputs and feeds an agent that runs Sunday evenings.
But the seven day feedback loop was brutal. So I started looking into evals.
The fun part about working with this strategy was the mocking. I mocked various failure modes with JSON fixtures, then adjusted the main agent. Fixtures like: a sub-agent returns an error, a sub-agent returns an empty set, one input dominates the others.
I wrote up the strategy, mental model, and mocking process. Really fun! danubilla.com/blog/engineeri…
