Cantilever Holdings LLC · Applied AI & Systems Thinking
For two years I hand-picked a quote every weekday morning and sent it to friends, family, and colleagues. This is the record of rebuilding that habit into something that runs itself — including the three approaches that failed first, and the moment the whole strategy changed mid-conversation.
Every weekday morning, before anything else, I browsed BrainyQuote, picked something that felt right for the day, and sent it out over iMessage and WhatsApp. It worked — people looked forward to it — but it lived entirely in my head and my thumbs. It broke down when I traveled. It had no record. It cost roughly ten to fifteen minutes I never got back.
The goal wasn't to replace the taste, just the labor: keep the eclectic, non-cliché quotes and the consistent visual identity, and remove myself from the critical path.
Each one was a reasonable bet given what was known at the time. Each failure was diagnosed from evidence — logs, error messages, isolated tests — not guessed at, and each one directly shaped the next attempt.
A scheduler outside any personal device triggers a cloud job, which generates and delivers the graphic with nothing left running on hardware I have to keep awake.
Once delivery was reliable, a quieter problem surfaced: Claude kept converging on the same “canonical” quote for a given theme — the same William Blake line every time “Energy” came up, regardless of time of day. This isn't a bug or an infrastructure fault in the usual sense — it's a property of the model's default behavior that has to be explicitly steered.
The fix was prompt engineering, not code: a randomized seed per call, explicit instructions to avoid recently-used or over-cited answers, and a wider net of source figures beyond classic literary quotes.
The interesting part of this build was never the final script — it was the sequence of falsifiable hypotheses tested along the way. A pilot reveals constraints that weren't visible at the proposal stage. The right response isn't to abandon the initiative; it's a deliberate platform decision based on what was actually learned.