[00:01]
you how you can make more money and save more time using AI without writing subscriptions. This is going to show you how you can force AI to give you much higher quality outputs. Pretty much no matter what business you do, whether
[00:14]
you're in software or services or coaching, e-commerce, I've done all of those businesses. You could also apply this to an agency if you wanted. You're a task, for example, you know, planning out a Black Friday launch, for example.
[00:27]
e-commerce brands and the various portfolio businesses that I have. And AI It'll even go and write the plan. And it might look solid at first. But the problem is that nobody's reviewed it and nobody's checked it. This is why most
[00:42]
people call what AI creates slop. It feels like AI created it, and that's honestly because it did. And so what's the solution? We make the AI argue against itself. we have the model argue with a different model about the way
[00:57]
that it approached the problem. So for example, I might say, "Help me plan out my Black Friday launch for whatever I'm selling." You're going to tell this to your primary agent. So it doesn't really matter what agent you use specifically.
[01:09]
I typically use Claude Code or Claude in the desktop app, but I've also used matter, but this is going to be your primary agent. And so you can give pretty much any task to it. You can say, "Help me draft up a plan." You can just
[01:23]
give it a task directly and say, "You go write out these emails for me." Or it doesn't matter what type of work you're giving it. And so, for my example, it's going to go ahead and just draft out the Black Friday launch plan. Or maybe it's
[01:37]
attempt, it's just going to basically throw stuff at the wall and try and come up with obviously the best thing that it can, but it's still ultimately the first draft. And so it might look good to you initially, but is it as good as it
[01:51]
to show you how to set up is an automated review agent. And so imagine if that first agent could spawn another agent to take a look at it. This can happen in a few different ways. So that primary agent, we told Opus, "Okay, I
[02:05]
want you to go out and I want you to create a Black Friday launch plan." Well, it can have another Opus agent spawn off right after it's created the first draft of the plan to review that plan and try and poke holes in it. It
[02:18]
can do this with the same model, but it can also do it using a different model. For example, Codeex or Chatg GPT or even Grock or whatever models you might have basically going to review whatever deliverable needs to be reviewed from a
[02:33]
perspective. This is super helpful because it can attack that deliverable from a bunch of different angles without you having to think through and iterate over every angle that the original agent might have missed. Okay? If there's
[02:47]
involved, or maybe there's some risk that you didn't see at first, if you have an extra subscription laying around, which most people have you more you do if you're watching my channel right now, that review agent is going to
[03:02]
be able to poke holes in that plan. Things that you probably would have never seen on your own. This can be anything. For example, what happens if [music] now will be able to think
[03:14]
it goes back to that primary agent for revision, it's going to be able to fill in all those gaps in the plan before you go to market. You could just have these agents kind of go back and forth and continue arguing over the deliverable
[03:27]
until they agree. If you just accept the first deliverable, sure, it might be, you know, 80 or 90% of the way done, but after the loop, you're going to get a much higher quality output almost every single time. There's several ways that
[03:40]
just have a one-off job like creating a Black Friday sale, you can obviously just hand AI that job, you can say, you know, here's the task, follow this procedure, and you essentially just tell it, you know, do the work. So, create
[03:53]
the Black Friday plan for me and then spawn an agent to review the plan and then once you guys agree on the plan, you know, hand it back to me for the final review. The way that I usually use this is by building it or baking it
[04:05]
directly into skills. So for example, if we have a pipeline of skills that help me create deliverables, for example, in the software development, I might have a research and then a planning phase and I might want review agents to analyze
[04:18]
either the research or the plan that it created to pull holes in it and find areas that we can improve before we start building and shipping. Can be the sequence or doing a piece of research. So you can bake it into the skill so
[04:31]
that you can run those skills without having to sit there and you know monitor own. You can also bake it directly into your agents memory. So when you tell it you know I want to always spawn an agent when we do X Y and Z task. So anytime
[04:47]
we're drafting up an email campaign or anytime we're doing content research I want you to spawn review agents or additional research agents to make sure this from the correct angle. I specifically use this the most for
[05:00]
planning so that we can create a really highquality plan upfront so that when we actually start working on something, all of those hours are actually put towards In my software company, for example, since we want it to be as secure as
[05:14]
possible, we always do security agent reviews. So, we'll spawn multiple agents across multiple different platforms. We'll have Claude, Chatypt, and Grock agents all attack the And this is where it's really valuable to have multiple
[05:29]
agents that are going to attack your site or attack your plan from different angles, you're able to spot those issues before you ever start building and service. So, you know, if you have a customer service agent set up, which is
[05:42]
one of the easiest and highest ROI, it saves you probably the most amount of with AI because why wouldn't you? You know, you can have an agent set up to monitor support tickets and respondse to
[05:54]
them, maybe drafting replies, and then you can have review agents check those replies. And you can set up specific conditions. For example, if a reply requires a money path and it's going to be issuing a refund or something, you
[06:07]
can have another agent review those specific support tickets so that it can take you even further out of the loop. This is basically the exact flow. So, your primary agent. you don't want to have to think about it and it will
[06:21]
automatically review until they both agree or sometimes your primary agent might override the review agent if it's being a little too critical. But either way, you will almost certainly get a higher quality output every single time
[06:34]
because you've had these additional review cycles without you having to interfere here. And so you can set that up either in a one-off prompt. It can be explicit about having it do this on its own depending on what the task is. You
[06:47]
can also bake it into skills and memory. You can also have review agents do really cool because you could have, you know, the generic sort of auditor kind of taking a look at the plan or deliverable as a whole. You can also say
[07:00]
like, hey, impersonate me and act as if you were me a year from now. Does this align with my goals? Does this align with the direction that we're going? Or, you know, the new hire agent. You could be the skeptical buyer. So, you could
[07:13]
say like, you know, take the role of a person who might not want to buy off of this sales page before we draft all of the copy and actually add it to our website or, you know, what does the other side's lawyer say about this? This
[07:26]
is helpful because on the first draft of your project, you might not be thinking that's why it can be helpful to bake these in to the review agent. on a contract, for example, what would the other side's lawyer say? So, you know,
[07:40]
you might not have noticed a specific term that was slipped in there and that you're doing some negotiations, right? You can have that review agent play the person across the table. So, poke holes again and do your numbers line up. How
[07:54]
are your SOPs? You know, if the agent is a new hire, how are they going to respond to being onboarded to your company? I've also ran review agents running. And so let's say that you have a customer support agent. Well, you
[08:07]
could spawn more review agents and stress test it, saying like if someone, you know, hammers for a review, but it's been 8 months since they signed up, how does it actually handle that? So, you can just really stress test basically
[08:20]
every idea that you have, trying to tear apart any deliverables that the AI is creating. Hopefully, you save a lot more time from having to do this manually, mistakes that the AI is making up front. And if you want the prompts to set this
[08:34]
link in the description. You can have them completely for free because yeah, I you a bunch of time if you're able to use it.