Block 3: AI Out of the Box Is AI Out of the Box

Month one, you were gung-ho.

You paid the twenty bucks. You typed a thing. Something magical came out. You told your team.

Month two, the output was mediocre. You said the sentence Daphne hears every week from prospects: "I can't make AI work for me. I can't get it to do what I want it to do." (Daphne quoting prospects, Daphne call 8/14)

(Brad note: I refer to team members and clients by name across these lessons. For context, these are folks I work with directly. If it sounds like a fit, apply at buildwithoptimus.com.)

Then: "Oh yeah, actually I have to do a lot of work, don't I?" (Daphne, 8/14)

And you walked.

Look — that's not a you problem. That's the framing everyone gets sold and then quietly abandons. The framing is that AI is a magic vending machine. Insert prompt, receive miracle. That framing is a lie the influencers tell to sell you the next course.

Here's what I told Daphne on August 14th, verbatim, because it matters:

"Out of the box, AI is magic, but their expectation is it just does everything for me. I don't have to think, I don't have to architect, I don't have to reason. I don't have to give it the data... and iterate and babysit it while it figures s**t out... AI out of the box is AI out of the box." (Daphne call 8/14)

Same call: "None of the s**t I've built has taken an afternoon. It's an iterative process." (Daphne call 8/14)

You are not a customer of AI.

You are its mentor.

That's the reframe. That's the whole reframe. "You're using this tool and you're mentoring and guiding it through." (Daphne call 8/14)

The people who look like they have some kind of superpower with AI don't. They just don't quit after run one. They run it, they see what's broken, they collect the broken parts, and they feed them back. Then they run it again.

I said this to the team on the 17th and I want you to write it down: "I collect all of the stuff that's sub-perfect... and then it gets better and better and better." (Team call 8/17)

That list — the sub-perfect list — is the entire method. That IS the work. If you're not building the list, you're not doing the work, you're just complaining about the output.

There's a version of this that scales past what you can babysit in a doc.

Agents make mistakes. Every one of them, every day. You don't prevent that — you layer against it. The odds a single agent misses something are high. The odds six agents stacked together all miss the same thing are very low.

That's how you manage a fleet. Not by hunting for the one agent that never errs — that agent doesn't exist — but by building a stack that catches itself. One drafts, another checks the facts, another edits the tone, another flags anything that reads like AI, another signs off. Any single one will let something through. The stack won't.

Same move as the sub-perfect list, just running in parallel instead of in series. You're still the mentor. You're mentoring the architecture now, not the prompt.

Want a better starting point than a blank prompt? Drop-in Prompts is seven battle-tested ones — start there, then run the sub-perfect loop against the output.

Do this

Take the deliverable from Block 1 — the real one you shipped to an agent.

Look at what came back. Open a doc. Title it "sub-perfect."

Write down every place the output missed. Every tone thing. Every fact it got wrong. Every format issue. Every place it read like AI.

Feed the list back in. Run it again.

Do that three times.

The third pass is a different piece of work. You did that. Not the AI — you, mentoring the AI. That's the whole game, and it's the reason people who "get" AI aren't smarter than you. They just kept the list.

Come back once you have three passes. We're going to build on this.

— B