Big Idea

The argument in one line.

Big agent factories start small: one director agent that only coordinates, plus separate roles with their own rules, grown by iteration instead of designed up front.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You chat with a single coding agent all day, re-explain the same expectations every session, and feel stuck in the iteration loop.
You've heard 'agent factories' and 'subagents' and want the smallest concrete setup you can copy into Claude Code or any agent client.
You're building a larger agentic system and need a way to split responsibilities so each part can be tuned and diagnosed on its own.

SKIP IF…

You want a finished framework or the author's full production pattern. He says he's saving that for a later video.
You only use AI for one-off questions and have no repeated work to codify into roles.

TL;DR

The full version, fast.

Agentic systems, or factories, look intimidating but start simply. Create a director: the agent you talk to, whose job is conversation, breaking down requests and spawning sub-agents at the right model strength, never doing the work itself. Then define roles, each a bounded domain with its own rules, so you can tune and diagnose one piece at a time. In a live poster-making demo, the director builds roles, runs jobs in parallel, and accepts changes to work in flight. Keep adding roles as you notice repeated requests, and collapse crowded areas behind sub-directors.

Free for members

Chat with this breakdown — free.

Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.

Create a free account →

Chapters

Where the time goes.

00:00 – 02:08

01 · Agent systems, and the main-loop trap

A complex factory diagram up front, then why a single agent keeps you stuck in the iteration loop.

02:08 – 06:44

02 · Asking for a director

In an empty poster-maker folder, he asks for a director that converses, delegates, tunes sub-agent intelligence and reports status.

06:44 – 07:12

03 · Why separate the work

Splitting research, writing and building into roles lets each be tuned and diagnosed independently.

07:12 – 12:05

04 · Defining the roles

He asks for artist, numbers, colors and words roles, and explains a role is a domain you can write consistent rules around.

12:05 – 17:08

05 · The roles at work

The director spawns sub-agents in parallel, picks models per role, and builds poster concepts for Paris.

17:08 – 19:40

06 · Changing work in flight

He adds an art-direction system, an engineering role and a three-poster gallery while work is still running.

19:40 – 23:00

07 · You build it as you go

Queue more cities, set roles by model effort, and turn repeated corrections into new roles.

23:00 – 27:03

08 · Directors of directors and the gallery

Departments collapse behind sub-directors, the poster gallery is shown, and he points you at this video as a starting director.

Atomic Insights

Lines worth screenshotting.

A director's job is to talk with you and delegate. The moment it starts doing the work itself, you're back in the single-agent loop.
Separating work into roles gives you one place to tune colors, words or research depth without touching anything else.
A role is a context or domain small enough to write consistent rules around, not a job title borrowed from a human org chart.
You don't need the brilliant design up front. Run one thing, see what's wrong, and add the role that fixes it.
Tell the director what outcome you want and let it coordinate. Hard-coded 'this calls that' workflows are what you're trying to avoid.
A director can steer work already in flight, so you don't wait for one task to finish before changing the plan.
Specify model effort as low, medium or high in a role rather than a model name, so the role works on any agent platform.
Every time you say 'why did you forget this again', that's a rule that belongs in a role definition.
Memory won't track your expectations per role as the system grows, so write them down in durable role files.
Even tiny chores like 'verify you've pulled recently' can be roles the director decides to invoke.
Once a department gets crowded, collapse it behind its own director. Directors talking to directors scales the pattern.
Roles aren't only for code. A role can manage slides, do research or triage the five most important emails.

Takeaway

Start with a director and a few roles

WHAT TO LEARN

You don't design the whole agent system up front: install one coordinating agent, add roles as repeated needs show up, and let the system grow.

Make the agent you talk to a coordinator that delegates and reports, so it stays available instead of getting buried in the work itself.
Let the director choose how much model intelligence each sub-agent needs, small for light jobs and stronger for creative or reasoning work.
Define each role as a small domain with its own rules, so you can tune or debug colors, research or engineering in one place.
Describe roles by effort level (low, medium, high) instead of specific models so they carry across agent platforms.
Don't hard-code workflows between roles. Give the director the outcome and let it decide what to call and when to run things in parallel.
Steer work in flight by telling the director the change. It updates the running pieces without you tracking each sub-agent.
When you catch yourself repeating a correction, write it into a role instead of hoping memory keeps it.
When a department of roles gets crowded, put it behind its own director so the top level stays simple.
Roles work beyond code: slides, research, email triage or any recurring job can become a role with a one-line definition.

Glossary

Terms worth knowing.

Director: The main agent you converse with. It breaks requests down, spawns sub-agents for each role, chooses how much intelligence each needs, and reports status instead of doing the work.
Role: A bounded domain of work with its own written rules, guardrails and evaluation, such as research, colors or engineering, that a sub-agent adopts when started.
Agent factory: A system of cooperating agents, each with defined responsibilities, that turns requests into finished work with the human steering rather than executing.
Sub-agent: A worker agent started by another agent to handle one task in parallel, often on a smaller or larger model depending on difficulty.
Department: A group of related roles collapsed behind its own director so the top-level director doesn't have to manage every role directly.

The Script

Word for word.

Read-along

Don't just watch it. Burn it in.

See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.

Okay, today I want to show you agent systems or factories. You've probably heard of them called factories. Same thing, really, really the same thing.

But first, let me show you kind of a diagram of a very complex version of this. And don't get scared. This is not where we start, but this is what I'm about to show you how to build.

And it is so simple, we're going to walk through it step by step. All right, let me show you what I'm talking about. Okay, so the moving parts that you see here.

are kind of the different elements inside of an agentic system. So each part has its own responsibilities, has its own rules, instructions, evaluations, guardrails, all these other things. Each single element here has the capacity to be able to be tuned in that way.

And that really becomes the value. And when you zoom out a little bit and you start noticing that our little system here, which was complex when we looked at it, is actually just part of a whole bunch of other potential systems. that are cross -communicating, you start to see the power and why everybody's talking about agentic systems or factories and how we need to move forward into this using AI in the future.

Okay, so I understand that was a complex system. These are the kinds of systems that I'm building and the ones that I've been working in for months now. And they are wildly beneficial and you absolutely need to adopt them.

By the way, they don't need to be that complicated to be hugely beneficial. the moment you put it together, it becomes valuable. And we will start from the very beginning in just a second.

But I want to tell you, if you're having a conversation with a single agent these days, some agent system that you're asking it to do work, it does work and then you evaluate it. You'll know what I mean by being stuck in the main loop, kind of being stuck in that iteration loop. You're so required to get this work done that you can't imagine scaling broadly.

These systems will allow you to scale wildly as they end up taking a huge number of jobs and kind of codifying each one. By the way, you're the one that's describing the shape of the entire system. So this is not magic.

I'm going to show you how to get started. And I think the best way to do this is let's start with something that's just a ridiculously useless project. in a completely empty folder to get started so let's go right now what i'm going to try to do is show you the steps that i would take to starting something like this so i'm going to start on my desktop i'm going to create a new folder here called poster maker so completely empty folder of course let's open poster maker in cloud code so that we can start having a conversation with it so now we're in poster maker and let me say hello What we're about to do is ask for what I call a director.

And this director is just basically the first character in our set of roles that we'll have in this kind of agentic system, in this company, in this tool that we're going to create. So that's all we're going to ask for. Let's get started with that.

Okay, I want you to create a director pattern here so that the main agent that I start talking to is what I call a director. And I want you to use a little clapboard. Uh, uh emoji whenever that director talks to us just so that we know we're talking to the director if you don't mind But what I want the director to be able to do is its job is to be very intelligent and to be a co -partner In what we're trying to solve so it needs to be creative and interested in solving the problem But it needs to try to not do the work its job is to have a conversation with me and always be available to have a conversation with me and use sub -agent patterns to kind of create new workers as needed to do the different roles for the work that's needing to be done so when a request comes in the director will figure out what that request is and kind of break it down intelligently now the director also has the capacity to tune those different sub -agents that it's bringing up to use different levels of intelligence so it can use a very small model if it's asking to do something that's very lightweight or a more intelligent model if it thinks that it's doing something that needs some creativity or some more reasoning behind it so that's a really important aspect

of what this director does. And one other thing that it does is it reports back to me on a regular basis. While sub -agents are still running, the director thread is always running and reporting back on a regular basis what's going on and what the status of those sub -agents are.

All right, and that's pretty much it. So now we have what I call the director pattern installed inside of this folder. There's a little bit more to it if you want to get really a little more complex as we lean toward what we saw in the opening diagram.

But our job here is just to make it available to us, just something simple. And what I just said is absolutely enough to have kind of a good partner in the work that you're doing. If you create a very simple skill with just that in it that says install a director here and it does that, then...

Every time you open a new conversation in this project, you'll be talking to a director. All right, I covered this in a recent video. I advise going and watching that video if you're more interested in this.

But I just want to say, what is this director concept about? What is this director pattern? Now, when we talk about factories or we talk about any of this agentic system work, what we're really saying is we want to separate the work into meaningful parts, more atomic kind of elements doing the work, if you will.

What I want to be able to say is if I have a job in kind of this project that we're building, if I have a job that needs to do research, then I have definitions around how research works, what the responsibilities are, what guardrails might be, how it might test, all of this complexity. around how research works. And I need to be able to create that definition such that whenever a researcher is running, it uses those rules to run.

Same thing for one that might need to do writing or one that might need to build an application or run a PR or do CICD or any of the other moving parts of your system. Each one of those becomes an area that you kind of want to separate from the other areas for a lot of reasons, for at the very least diagnoses, if something's not quite working perfectly.

or when you need to tune, that separation allows you to tune each one of these parts independently. Now I showed you once again, that diagram at the beginning, very, very confusing, lots of moving parts, and they all have pretty intense kind of rules around them. This is what gets us there.

This is what gives us the possibility of putting those rules in the right place. basically without this separation you can't create a complex system that can grow and shrink and learn you only have that single agent and that becomes a problem so this one piece of separation i call it the director pattern so that i have one interlocutor to work with and it kind of communicates to the other sub -agents in meaningful ways creates more of them when it needs it runs in parallel when they can Those sorts of things is what its job is.

So that separation, all we're really trying to do is create space for us to then layer in the needs of our system or the complexity of the system in the future. And so we can start a new conversation here and just say hello to a brand new session. And of course, it comes up and says, hey, I'm your director for this project.

What do you want to make today? We haven't told it how it works or what we're doing here. And in fact, you don't even know.

And this is a silly project. You understand we have nothing so far. And we're going to build posters.

This is a system that knows how to create some kind of posters or something like that, right? So what I'm going to do is have a conversation with my director here. Okay, so listen, I am interested in this system being able to build posters for us.

And I'm interested in those posters being related to... for now maybe like an area a location let's say paris or whatever it might be and what we need to do is be able to create different roles that we define that the director can send work to each one of those work those roles and potentially work in parallel where possible so there's one artist poster generator so the one that kind of composites all the information together and works with a skill that we have here called gpt image and use that skill to create a prompt to create the image it also obviously needs to be able to evaluate the image that came back whether or not it meets its needs those kinds of things There are other roles, though.

So I think I kind of want to create a pattern of three different elements being put together creatively for each one of these posters. The next role, after our artist role, I kind of want to have one that deals with numbers. It's going to have to go do that research itself.

Same for the other two. The next one will be colors. Are there colors that are representative of the city?

And then the third one needs to be words. Dependent upon what words we choose or why we choose that that might lead our other two kind of roles to pick what they're doing So I don't I don't know how this works out but maybe we end up with kind of a the artist is the one that determines the Basic structure somebody needs to do some research to figure out what the basic concept of this poster is going to be Can you help me create those?

So, OK, you heard everything that I'm asking for. Pretty straightforward. It's a little various.

I'm not really sure on what I'm building here myself. We're just having a little bit of fun. And that kind of variability might be how you work with your agents when you're talking to them.

You might not necessarily have the absolute picture of what you're trying to do to begin with. And it really does not matter because we'll run one and we'll see whether or not it's good. And we'll come back and say, well.

Actually, what we need is an artistic director or maybe even a historical researcher that gives us some information first or an area or object researcher or whatever it might be, right? We might add more roles. All right, so a real quick concept here about roles now.

I don't typically like thinking of these things as roles. I think it just helps as a mental model for you to get started. But roles kind of ties us into something that a person would have done.

So here's my researcher role or here's my writer role. And those are the examples that I've even been giving. I don't think those are the best scenarios.

A role is really a context or a domain that someone can control. And it's a small enough piece of work or concern that you can write rules around it that all work. You might want it might get pretty big because the rules are so nested together that you want them all in the same place For example, we're doing colors and words and numbers and you heard that I was defining it and realized wait a second We're gonna get one number from a historical reference and the words from some meme that just recently released and they're not gonna make sense together That becomes this I'm separating these so so dramatically it would have made a lot more sense to say let's pick the numbers and the words and maybe even the colors all together in one role so the the kind of the definition of the subject matter of this product or this poster should have been in one role but that's fine we're just separating it so you can see the delta here and what might be different these roles can do all kinds of things like almost anything you can imagine is what you can make a role for and even if it's a tiny little responsibility like oh every time we

do something I want you to check it in or I want you to verify that you've pulled recently or whatever it might be those very simple kind of what you might think of as as hooks today those can be roles in your system and your your director can be the one that makes a determination of do I need to employ this at this moment what you'll notice that I didn't or I tried to stay away from it, sometimes hard, is a real workflow.

Saying this needs to call this and then this guy needs to call that. You want your director to kind of be able to coordinate this. That agent is actually directing things intentionally.

So all of this complexity that feels like, oh my God, how do you get to this final huge product that does all of this crazy stuff is just iteration. So I don't want you to be scared of what these roles are doing. And by the way, the roles are not so strict that it's the only thing the system can do.

I almost all of my systems these days are self -learning and they create their own roles when they identify that I'm kind of doing the same kind of work again and again. It starts writing roles relative to the kind of work. that i ask for so that it kind of will create a consistency between doing that kind of work all right let's see where we've gotten with these roles okay so here it is running along and you can see the background tasks that are happening at this point and this is good this is that example that the director knows that it can create sub -agents to do work and so its job is to try to start peeling out different sub -agents to figure out how much work needs to get done if we looked at all of the different agents that have happened so far two are running right now and the others that have finished so okay looking inside of what it's building you can see that it's come up with different um workers or basically roles what we're asking for that it might need

to use and you see that it's created a concept writer research writer artist writer pipeline writer these kinds of things and it's even maybe chosen different models to begin with i'm not sure that i would want that in the role definition but that's absolutely possible to say when using this role this is how this should this one should work if it's doing this work this is easy work always use sonnet for example that's possible to do i would of course rather just say medium so that it could be used across any kind of um agent system or any agentic uh platform that we might be running it in but so now it's just working through the rest of these roles as it's trying to write them and you can see it's just writing out the different roles and we won't go through these but you can see it's got a pretty good definition of how it's supposed to do each job and you can come in and take a look at this and say is this how i really want colors done you're welcome to read through it or just ask your director right now like we can absolutely since it's a director ask it is can you give me an abstract of how we might be handling colors what's our definition so far great and it says everything's set up so while that's sitting there it's also saying everything's ready do you want to try paris let's do try paris while it's

having this other conversation with us so that's good and then here you can see and again i won't go through this but it's kind of describing to us what's going on inside of that color system and if you want to adjust this in some given way or give it to another agent to say come up with a much better thoughtful paradigm for how we might choose color palettes etc etc update that role.

This is that separation that I was talking about. So suddenly, if I want to choose words in a different way, I have that tunability in one location that I can work with my director and say, let's just work on how we do words. Or let's just talk about how we pick numbers.

What numbers should be allowed, what shouldn't be allowed. All of that gives me separation. Or how hard is it to do?

How deep should the research go? Those kinds of things are all doable because we've created this separation. All right, and it is off.

And the Paris Guinness World Records for... It's looking in a lot of different places for Paris concept. So the concept roll is looking through and trying to figure out what the poster should be about before the other rolls pick up.

So great. That was our concept roll kicking off. And now it'll go into all of the individual rolls to get the numbers and the colors and everything else.

Of course, this is not something that I've defined. And if I don't like how it's working... As mentioned, it's just a conversation away.

Okay, so while that one's still working on Paris, let's add some more kind of responsibility to this system. Real simply so that you can see, how would I add more responsibility? This thing's even running.

I have a director who's sitting there that will continue to give me updates to what's happening. That's one of his responsibilities. But otherwise, he's actually kind of idle.

He's there and available for me to work with him. Hey listen, I think building these posters is kind of interesting. I want to make sure of two things.

I want to make sure that the design of this poster that we're going to get back is highly integrated. I want it to be very creative and interesting. And I want it to feel like a travel poster or something like that.

Something that will be... more graphically rich and not realism. I'm not looking for photographs.

I'm looking for graphics and feels more like a reason to visit a place or be proud of a place or something like that. That's what I'm looking for from the art design. So if you put together some system that the different, maybe the concept role or somebody else.

kind of starts to understand and certainly the artist what our art style is i'd love that figure out how to fit that in and also i want to make sure that these get updated into a web system of some kind so i'm looking just to be able to update a website and we'll do this locally so just build a small website that can hold the different posters for us that we can go and see them you know in kind of full screen if we need them or kind of full mode or thumbnail mode and we can just browse through the items as they start landing and as they land we'd like that to update on a regular basis if possible so we'd like that built and an engineer probably needs to be a role as well so that we can continue to update that and start finding best practices eventually maybe pushing this to a cloud service or something like that so if you can get those started i'd appreciate it Okay, so here it is.

It's kicking up the art direction system and also the engineering role. Ah, good. I see you're working on these things.

Listen, I need some changes to what I just asked for. I want to make sure that we try different kind of art directions. I still want them all graphically rich, but every time we create one of these things, I think we should create three pieces of art, three different posters, and...

and i don't i i want them to allow them to be different art styles it's perfectly okay they need to be separate from one another but really as much as anything we want three options and that also means that when we display them on the website i don't want them as three different thumbnails i want kind of a stack of different thumbnails on top of one another.

So the Paris one we just created, for example, would have three little thumbnails that we would be able to go into and see kind of a grid or a composite of those three or some way to flip through them.

Okay, so what's interesting about that is I'm changing work that's in flight. So essentially, it will have to make changes to what's going on inside of this. engineering role and the gallery site that's being built as well as the art direction itself this is one of the values of what we're doing is this director has the responsibility of understanding the work around it and being able to update the work that's being done inside of those and communicate down to those sub -agents i understand this sounds a little bit nuanced but this ends up being an enormous pattern it means you don't have to wait until one piece of work is done before you make adjustments and steer it can actually steer those different elements for you and you don't even need to know the different parts that are out there doing the work and you of course won't because these systems end up being much bigger than you might anticipate and having much more kind of moving parts than you might expect so this is a really useful pattern okay so this thing is working along it's doing really well it created uh the

the gallery for us so here is the gallery so far there's nothing yet here but it has picked some of the themes and other things so we'll be able to come in and see this stuff very very cool what i want to do is one more thing as you might imagine it's fun to kind of push it around and say all right so it is not finished building the different parts that we asked for last time but in the meantime let's see if the director is strong enough to kind of stage work and hold work on Okay, so I know you're still building a lot of the moving parts and you might not know how to do this yet.

But listen, what I want to do is I want to run, let's say, Athens and let's do Cairo. And I'd like to do, I don't know, something like bicycles, airplanes, and New York City. All right, get those posters running as well.

You know, we want the final versions of this. All right, so there we go. We have all of these running.

We'll get back to those in a second. So I think what you're seeing is as we add more and more things in here, it's describing more and more elements. One of the last things that I would do is I would tell it.

Listen, I want you, when you start a subagent, I don't want that subagent to be a director, of course. So I want you to make some change in our instructions so that when you start a subagent, you can tell it what role it needs to take and what intelligence it needs. So what model or what level of effort it needs to be able to use.

And that is what it chooses from kind of our agents file today, instead of naturally starting as a director. Can you make that change, please? So if we just take a look at the background tasks that are running, there are 12 finished, and I think there's like 12, in fact, running right now.

So these are all of the things that are going on, but our agent, our kind of director, still is there for us to work, and it understands. this much work is occurring so it's really valuable to have a participant with you that understands the system and you get to strengthen that i would say that's where i spend a large part of my time is strengthening the relationship that i build with that director and by by doing that what i mean is describing the rules that it needs to follow or my expectations of it now at some point you'll have so many rules around your director that you'll realize It's actually doing roles now as well.

I need it to not do so many roles that you then will go back to the director and say, hey, look, there's things in here that are going on. I don't want you doing. You need to turn those into roles.

And when I ask you to do certain things, you give those to roles when they're needed. So that's how you build these systems going forward. All right.

So the magic of this system is definitely that you build it as you go. It's one of those things that you've been doing iteration for a long time. You're probably doing it today.

And what I would say is that is no longer the way to do it because what you're doing is you're pushing in saying, why did you forget this again? I just asked you three times to do this or go check the last conversation we had. We were doing this kind of work.

Why are we still making these mistakes? All of those statements should be landing somewhere in your head. They are landing somewhere.

You want them to. These systems have been using memory systems for a while now, so they are keeping track of some of these, and they're getting better about them, but they're not going to be keeping track of absolutely everything, and they absolutely will not keep track of your expectations per role as the system gets bigger and bigger.

Not to mention you want this thing to be durable across different changes and different systems and different environments. And those memories are not always durable in that same way. So what you're really building here is a small starting point that you can have a conversation with an agent.

And the understanding with that agent is you don't do the work, you dole out the work. And you get to choose how hard the work is. So that relationship with the agent that you keep pushing back in with the director saying, I need another role that does this.

Let's say that we wanted to make sure that we didn't have duplicates, or we wanted to do some evaluation, that the art style was up to some desire, or that all of the posters, you know, kind of looked the same that were coming out on a given day, or something like that. We can just start creating role after role, another evaluator that does that, and our intelligent director will figure out how to work those together.

What you're seeing in that opening diagram I showed, i have gotten things so complicated in some of my systems that i need to collapse those behind directors so i have a director that deals with all of the audio in my game because there's a lot of roles to do audio and i don't want my main director to have all of the roles of the entire system so i said great i'm going to have what i call a department that does audio and when my director talks to the department For audio, it's actually talking to another director.

The exact same thing you just saw. So it's kind of a recursive system down the line here. So this graphing works that a director talks to a director and says, here's what we think we need from you.

That director knows all of the roles in its area. And that continues as things getting more complex. Just creating the space to be able to do this.

is the actual answer i can't describe it enough it is this simple so before we close i want to show you some of the really cool posters that have been flashing up in my eyesight over here so let's go take a look at some of those all right so we've got to take a look at some of these some are better than others uh but let's take a look at the airplane one and of course it gives us the whole concept this was the basic concept that we got for the art itself And as we move through these, we can see the color schemes for each one of them.

This one's still working. A hero's welcome for the Statue of Liberty. I don't imagine the buildings were this tall in 1886, so I would want to put an evaluator on this saying make sure that the image that's coming back is valid.

If it's a time period piece, that it's valid for that, etc. Okay, very cool. The bicycle, 1896.

1797 the first parachute from a balloon apparently in Paris That's very cool Athens as violet crowned violet crowned All right, so those are great. Okay, so this has gotten pretty long I'll kind of keep it short here, but I hope you've seen the value of being able to create a system for you and the work that you're doing as you move forward.

You don't have to have the brilliant idea up front of all of the moving parts. That's daunting. It's an impossible task.

So just get a partner that you can work with and a couple of the roles broken out that makes sense, the kinds of things that you keep asking for. Right, you might have a front -end engineer and a back -end engineer if you're building a piece of software, or you might have someone that manages your slides and someone that does your research or looks at your emails or whatever it might be.

Those are the roles that you have in your system that even if they're just one line definition of how they work, I need you to go get my most recent emails and show me the important five. Okay, that's one of them. And another one is I need you to kind of keep track of all of the emails, whatever it might be.

It starts growing on its own. You'll see as you have that space, The separation creates the space for you to layer in your complexity and create an actual system that helps solve problems.

I'll stop now. I'm very excited about this. So hopefully this kind of resonated a little bit and that you have a little bit of a...

interest or patience to be able to just go ask for that director up front you can steal what i've said here as my director it's very easy to grab from the transcript point your agent at this video it'll grab that for you and just start there it's not at all perfect as i mentioned i have a whole pattern that i want to share with you later but i don't want to give that to you here because you'll see it as kind of a crutch to this is how i get started so we'll save that for later but Hopefully this helps you get started on your journey of creating agentic systems.

It really will change everything. I almost guarantee it to you. All right.

Thanks for coming along for the ride on this one. The very, very long ride. And I'll see you in the next one.

The Hook

The bait, then the rug-pull.

The big diagrams people call agent factories look intimidating, but Matt Maher shows they start with a folder, one director agent and a handful of roles. He builds one live, as a poster maker.