For Steve Watts, CEO — Pacific Glazing Corporation
Stop collecting AI news. The Submittal Tracker is already on the site. It already takes a spec PDF and builds a checklist. What it does not do is read a spec the way PGC reads one.
It does not know that lite count is not optional, that wet vs dry glaze changes the package, or that structural silicone and shop drawings are what actually get a submittal rejected. The job this week is to teach it one of those rules and see if it files a real spec without someone sitting on the checklist.
One hour. One real architectural spec — a recent job, not a sample. Run it through the Submittal Tracker. Sit with the output and write down every miss: wrong, incomplete, or silent on a glazing-specific requirement (lites, wet/dry glaze, structural silicone, shop drawings, mockups, warranties).
Then write those corrections as a reusable skill — a SKILL.md or the equivalent — sitting next to the tracker. Not a longer checklist. Not a second tool. How PGC reads a glazing spec, in a file the tracker can apply next time.
If the hour is gone and you do not have one durable rule, stop. That is the result.
A second spec. Same tracker, same skill, no hand-edits to the checklist. At least one previously missed PGC requirement shows up on its own — lite count, glaze method, structural silicone, shop drawings, whatever the first hour taught.
The hour produces notes that only apply to that one PDF. If you cannot write a rule that would fire on a different job, the tracker cannot be taught this way. Say so. Do not generate more briefs.
No software-watch list. No 15-trend scan. Those machines are off. The product is the skill next to the tracker we already shipped, not another briefing.