I often describe UI design feedback to Claude Code to make several visual detail fixes, sometimes with a screenshot. What I really want is to record a quick video pointing and describing the changes I want, and have that automatically converted to a detailed prompt +

Relearning to create content: Before: Stare at blank page → Write slowly → Edit forever → Publish when exhausted. Maybe. Now: Raw idea to Claude → Ask for specific improvements → Edit for voice → Publish something I like. Still tiring, but I like this better.

Really enjoyed the Billy Joel doc on HBO. A lot about his early career I didn’t know.

Artists/bands that last decades (rare in itself) tend to create their best stuff early on and churn out mediocre stuff later. Joel is a rare exception. The incredible songwriting just kept coming.

Agent OS now comes with 5 subagents for Claude Code, built-in. It's way more efficient at creating specs and following them. context-fetcher file-creator test-runner git-workflow date-checker

Checking stats on day one of a YouTube video launch feels both fun and actually helpful in this game. The ranking against other videos this many hours into their lifetime is a fun horserace to watch! Interesting to see how the algorithm and comments pick it up early (or not).

Cursor's recent improvements don't get enough love. The tab completion has become noticeably smarter and more accurately predictive lately. Just a few tabs with no fixes can sometimes be a replacement for a prompt to have AI build something (small).

YouTube is the first marketing channel that feels worth the hard work (for me). The algorithm rewards creativity, empathy & authenticity with repeatable exposure to the right people. It's driving leads & sales too. Everything else felt like luck or short-lived tactics.

this will be the longest 15 minutes ever.

Just dropped a new episode on my personal private podcast. Talking through ideas for monetizing my new business. If you're interested in my updates, you can get in here: ripple.fm/private-podcas…

Starting to think they just show this message 100% of the time.

Thinking out loud (literally) takes some getting used to. Also, my family finds my muttering technical jargon to the computer pretty weird.

Hit the Claude limit when it was 20% finished building a feature. Hopped over to GPT-5 in Cursor to finish the rest. Very solid job for it's first run. Not perfect, it fixed the minor issues nicely too.

I'm a heavy AI-first builder now, but I still don't build anything (real) in just one day. I spend days/weeks on planning, architecture and design, while agents execute most coding. Still 50x faster to ship than before.

OK I'm callin' it a day. It's weekend time. No, not you Claude. You'll need to pull a late one tonight. 😉

First day back to walking after surgery. 👌