Most companies hire a second support agent when tickets pile up.
I built a system instead.
50 tickets/day. Escalation rate dropped from 70% to 20%. Time to build: 2 weeks.
Here's exactly what I did:
An agent that can't check its own work isn't finished.
One video used to cost me 8 hours of editing.
Now AI edits it, then checks frames, cuts, text legibility and A/V sync before I see it.
I review a checked draft. Not a raw one.
Watch Later:
Where 55 hours of "I'll get to it" go to die.
Mine had 83 videos. An agent now pulls the captions, compares each idea with my notes, and keeps only what I don't already know.
Result: 3 videos to watch. 22 minutes.
99 LinkedIn posts. Median: 0 likes.
The problem: I never re-read my notes.
So I built an agent with Claude Code. Every Saturday it reads my sessions, my decision log and the files I changed.
11 minutes. ~$0.30. 7 briefs.
Notes nobody reads are useless.
I built an AI operating system that knew my priorities, my calendar, my notes.
Then kept opening Claude Chat tabs instead.
The smarter tool lost to the faster one.
Fix: make the OS faster to start than a new tab.
Friction beats features.
Hard pill to swallow:
The documents worth giving AI are exactly the ones full of personal data.
So I stopped choosing. Built a scrubber with Claude Code: swap names and emails before sending, put them back after.
First test run: 28/58 passed. Now 60/60.
Every one wants to creates features with AI.
I do too.
But the constant battles with our CTO right now are :
- Focus on our core business
- How to deploy and maintain all the new features
- How to increase our review capacity
New features = new load for our servers.
@thejustinwelsh And approval from people working from an old reality archetype
Sorry mom and dad our world is changing so quick.
Hard to explain why being a YouTuber matter so much nowadays !