The Build Log.

Every weekday: what shipped, what broke, what it taught. The posts live on LinkedIn; this is the archive of the days that went up, newest first.

Day 8

Claude can run Premiere Pro now. So we tested it on real work.

Tested Claude running Premiere Pro to open a new project, import media, lay 26 clips across six tracks, and save. Every keystroke worked. First six clicks landed in the wrong place due to screen-scaling on the operator's side. First two imports failed. Learned: the AI operated the software, but the decisions are the job.

Day 7

I asked one AI whether to install the other one's review tool.

Tested a code-review plugin by planting three bugs in one file. Found two with exact line numbers and fixes in two minutes and twenty seconds. Missed the third. Adopted it anyway because two out of three in two minutes is a reviewer worth keeping. You still read the code yourself.

Day 4

Google read the site in minutes. Nobody could find the login.

Rebuilt a friend's tree-service website. Google re-indexed from 8 pages to 23 pages in minutes. But nobody—not him, not me—could find the Google Business Profile login. That one's on me. Should have asked day one who holds every login. Also caught one of the article's two sources was a dead link before it went live.

Day 3

My AI said the snowflake melted. It fell off the rock.

Working on the animated short EMBER in a hotel lobby. Hadn't touched it in three weeks. One shot: a snowflake lands on a warm stone and melts. My AI operator said it was done after checking every half second. I watched it. The snowflake fell off the rock. First redo, a drop of water appeared. Third take, asked for less. It just melts. Learned: look at the work yourself. The mistake was hiding between the checks.

Day 1

I moved in August and still haven't finished setting up.

A friend's small business website finally went live this week after months of delay. I kept putting it off and paid for another month of the old website builder because of it. The quote form failed its very first real test, even though every test before that passed. Cut the page from over 15 megabytes to under half a megabyte. PageSpeed went from 75 to 97. Learned: the boring stuff costs money while it waits. A test only counts if it runs where the real thing runs.

The Build Log card for day 1.
The card that went up with it.