The Fastest Way I Know to Get Confident With AI

The other day, I was listening to Zevi Arnovitz's interview on Lenny's Podcast. Zevi is a project manager at Meta, and he talked about shipping product despite not having a traditional engineering background.

What really struck me was not the tools he uses but his approach. He described running review passes on his work (for example, using /review in Cursor) and even having more than one model look at the same output.

That drove home something I wish more nontechnical users knew early on: You don't need to trust AI to get value from AI agents. What you need is a review loop.

The Problem

If you're nontechnical, AI output can feel like the flip of a coin.

The agent hands you a plan, a workflow, or a chunk of code that looks clean enough. However, looks can be deceptive, so you ask yourself, "Is this reliable … or am I about to ship something I'll regret?"

The Solution

The solution is to generate, review, and interrogate.

1. Ask one agent to do the work.
2. Ask another agent (or the same one in a fresh chat) to review it.
3. Keep asking follow-up questions until you fully understand the critique.

Most people stop at step two. They read the review, then either ship it or scrap it. However, step three is where the actual learning happens.

When the reviewer says, "X is risky" or "Y is missing," don't just accept it. Ask, "Why?" "What would you do instead?" and "What small test could I run to prove it?"

That's how you develop judgment without having to become a tech expert overnight.

Prompts

Here are some prompts you can use to refine your process

First agent:

"Create a simple plan for X. Keep it practical. List assumptions."

Second agent:

"Review this plan thoroughly. Identify what's wrong, what's missing, and what you would change first. Highlight any risky assumptions."

Follow-up questions:

• "Why is X risky?"
• "What would you do instead?"
• "What evidence would make you more confident?"
• "What's the smallest test I can run today to validate this?"

Credit

Inspired by Zevi Arnovitz's interview on Lenny's Podcast: https://www.youtube.com/watch?v=1em64iUFt3U

If you're interested in learning how to use AI agents like Zevi, join Scout Academy, and let's chat about it.