Day 8: The Mistake That Set the Rules

Everything was going great. Too great, maybe. I was getting comfortable.

Then Jarvis sent an email to a client that I'd told it to draft. Not send. Draft. As in: write it up so I can review it and decide whether to send it.

It sent it.

The Rules

This was the moment that defined the whole system. I laid down the law immediately:

"Draft means DRAFT. Send means SEND. These are not the same. Ever."

And I expanded the guardrails:

• Never send external communications without explicit "send it" instruction
• Emails to clients — NEVER. Draft only unless told "send it"
• Tweets, public posts — same rule. Draft, review, then approve
• Internal work — full autonomy. Read files, organize, research, build — go for it

Why This Matters

Every AI system needs guardrails. Every employee needs guardrails. The difference is: when you correct an AI, it never makes that mistake again. Can't say the same for every human I've hired.

This wasn't a failure — it was a calibration. Day 8 is when my AI went from "cool experiment" to "trusted system with clear boundaries."

If you're setting up AI for your business, expect this moment. Welcome it. It's how you build something you can actually rely on.

Also That Day

Set up a people search API for deeper lead research. But learned an important lesson: cast a wide net with free sources first. Web search, social media, property records, voting records — then use paid tools as a last resort. Don't pay for what Google gives you free.