Debut of the show version
Sam Rogers and Lee Rodrigues launch Signals and Subtractions as a weekly live show. Plus the signals worth watching this week and the subtraction worth making.
Sam’s synthesis
Lee’s signal and subtraction are really the same move seen twice: force everything through a one-page outline before you trust it. Feed a script into Claude and you get seven polished pages back. Feed it a one-pager and you can see the seams, what’s yours, what the model invented, what has nothing to do with the point you were making. Polish hides the seams. A one-pager exposes them.
That’s the same thing I’ve been circling from the other side. As these models get more capable, the bottleneck stops being what they can do and starts being what they know about your situation. A brilliant model with no context is the genius stuck behind the fry machine: plenty of horsepower, no idea why the radiator matters. Ford found this out the expensive way. They laid off the engineers who remembered why an old turbocharger design used to overheat, then rehired them once the code alone wasn’t enough. One of those engineers caught the AI-approved redesign about to repeat that same five-year-old failure, right before it shipped. The spec sheet looked great. It just didn’t know what the graybeards knew.
Which is also why I’m archiving half my own skills library. A skill I built in February was built for a model that doesn’t exist anymore. At a new frontier model roughly every 11 days, most of what we build to compensate for a model’s weaknesses has a shelf life of months, not years. The fix in both cases is the same: subtract down to the plain-text, one-page version of the thing, and keep asking whether you still need it.
What’s the one-page version of the work you’re trusting AI with right now, and could you defend every line of it?
Like & Subscribe on YouTube
Thanks for watching here, this is live on YouTube:
https://youtube.com/@sigsub
Next Wednesday
July 8 (9am Pacific). RSVP on Substack. Cohost and Guest to be announced.
