/roast-room: How I Use Claude to Think Like the Top 1% — Transcript
Full transcript
- 0:00So, right here you can see that I've got
- 0:01Jeff Bezos, Paul Graham, Alex Hormozi,
- 0:03Gary V, Dan Martell, Gino Wickman, and
- 0:04Warren Buffett all discussing a business
- 0:07question that I had. You basically shoot
- 0:08off a question with different options,
- 0:10and all of these different council
- 0:11members will kind of come together,
- 0:13roast your idea, and help you think
- 0:15about it in a different way. Because
- 0:17each one of these seats has one specific
- 0:19question that they're focusing on. And
- 0:20now you can see what's happening is
- 0:21they're actually debating with each
- 0:22other. So, Jeff Bezos was debating with
- 0:24Gary V. Same thing with Paul Graham and
- 0:26Gary V. Now, I've got Alex Hormozi going
- 0:28after Gary V, too. And well, everyone's
- 0:30deciding to pick on Gary V in this roast
- 0:32room. This is crazy. There we go. Warren
- 0:34Buffett goes after Hormozi. But, the
- 0:36point is, like I said, everyone has
- 0:37their own question they focus on. So,
- 0:39Warren Buffett's question is, "What does
- 0:41it do to cash, risk, and long-term
- 0:43value?" Alex Hormozi's question is,
- 0:45"Will people pay? How much? And how do
- 0:48we close?" And Gino Wickman's question
- 0:50is, "Can the business run this without
- 0:52the founder holding it all together?"
- 0:53Now, after all three of those rounds
- 0:55happen, we can see that we have a final
- 0:56verdict, which was option C. And then we
- 0:58can also download a PDF, which allows us
- 1:00to read this, and we can sort of look at
- 1:02this more in depth, and we could also
- 1:04share this with our team if we wanted
- 1:05to. Now, this is really just a visual
- 1:06layer on top of what actually goes on in
- 1:08Claude code, which is basically just
- 1:10running a bunch of sub-agents based on
- 1:12these different council members that I
- 1:14have created. So, you don't need this
- 1:15localhost open in order to run and use
- 1:17this, but it's just a nice visualization
- 1:20of what's going on. And this is
- 1:21completely customizable, meaning you can
- 1:23switch out different council members,
- 1:24and you can build other ones, too. In
- 1:26fact, inside of this repo that I'm going
- 1:28to give you guys, there already are a
- 1:29few other ones, too. Hey Claude, what
- 1:31are the other council members inside of
- 1:33our roast room? And so, that's Glydo
- 1:35actually shooting off a prompt right now
- 1:36to Claude code, as you can see, and then
- 1:38it'll show us who the other council
- 1:40members are. Claude code comes back and
- 1:42says there are 11 seats. We have the
- 1:43default seven that you're seeing here,
- 1:44and then we also have Naval, Cody
- 1:46Sanchez, Sara Blakely, and Leila
- 1:48Hormozi. But, of course, you can build
- 1:50as many other ones as you want. So,
- 1:52that's a new feature of Glydo where
- 1:53we're able to actually orchestrate
- 1:55different chats between different coding
- 1:56agents. If you want to try it out, you
- 1:57can get a free month using the link in
- 1:59the description. But anyways, just
- 2:00wanted to show this off to you guys and
- 2:01let me break down how this actually
- 2:02works. So, this is a repo called the
- 2:04Roaster Room, as you guys can see. And
- 2:06basically, we have these seven different
- 2:07advisors as well as four extras, and I'm
- 2:09probably going to keep adding more in
- 2:11the mix as well. But what I wanted to
- 2:12show you is the advisors. What are these
- 2:14council members? Where do they actually
- 2:16live? They're actually just a bunch of
- 2:17sub agents. So, if I go into the dot
- 2:19Claude, actually let me zoom in a little
- 2:20bit more. You can see that we have
- 2:22agents. And in here, we have one
- 2:23markdown file for each of these sub
- 2:26agents. So, Bezos, Blake Lively,
- 2:27Buffett, Gary V, Graham, and all the way
- 2:29through. And if we open one up so you
- 2:31can see what it is, it's literally just
- 2:32a markdown file that holds information
- 2:34about the sub agent. So, for Jeff Bezos,
- 2:36for example, how he thinks, what are the
- 2:39sources that he references, how he makes
- 2:40decisions, all that sort of stuff. So,
- 2:42that when I actually shoot off a prompt
- 2:44to the Roaster Room, it's going to go
- 2:46through the different council members
- 2:47and it sort of adopts their personality
- 2:49of how they think and how they make
- 2:50decisions, what they, you know, and it's
- 2:52able to pull from things like their
- 2:54actual conversations, their lectures,
- 2:56their books, their principles, things
- 2:58that they put out there publicly,
- 2:59articles, stuff like that. So, what
- 3:01you'll notice is they will all have
- 3:02their own agent file, which isn't really
- 3:04that long. It's some things like how you
- 3:06debate, how you talk, your lens. But the
- 3:09really interesting stuff here is the
- 3:10brain, right here. So, we can see all
- 3:12the sub agents. But if I actually go
- 3:14into the skills in this repo, we can see
- 3:17that there is a Roaster Room folder that
- 3:20has brains. And once again, there's a
- 3:22folder for every single council member
- 3:23that you want to spin up. So, if we for
- 3:26example, open up Leila Hormozi's brain,
- 3:29we can see that she has hot, books,
- 3:31frameworks, positions, principles,
- 3:32sources, stories, and voice. And it's
- 3:34pretty much the same thing that all of
- 3:35these other council members have as
- 3:36well, but they all obviously look
- 3:39different. The content is different. So,
- 3:40if we open up Leila's hot, what happens
- 3:43is we can see her pillar is can your
- 3:45team and your leadership carry this? And
- 3:48who has to change? It shows us what this
- 3:50was built from. So, for example, Leila
- 3:51was built from 48 sources, about half a
- 3:53million words from 2021 to 2026, 14
- 3:56interviews where she's the main guest,
- 3:58and 34 videos of her own on leadership
- 4:01and things like that. Every person here
- 4:03has eight rules they decide by,
- 4:04frameworks, decision loops, things like
- 4:06that. Obviously, any books that they've
- 4:08written, frameworks that they talk about
- 4:10frequently and that they quote, their
- 4:11positions on certain things, pricing,
- 4:13discounting, raising money, debt, their
- 4:15principles, different sources that they
- 4:16actually pulled from, specific stories
- 4:18that they might talk about in videos,
- 4:20interviews, things like that. And every
- 4:22single one of these brains for the
- 4:24individual people is built the exact
- 4:25same way. If we open up Dan Martell, we
- 4:27can obviously see that he's got books in
- 4:29here as well. We can see that he's got
- 4:30frameworks that he's using all the time.
- 4:32You can see the buyback principle, the
- 4:33buyback rate, time and energy audit, the
- 4:35drip matrix. All of this stuff should
- 4:36sound familiar to you guys if you have
- 4:38followed, you know, Alex Hormozi or Dan
- 4:40Martell, value equation, grand slam
- 4:42offer, all of these things that they
- 4:44actually talk about all the time. A
- 4:45quick break to tell you about the
- 4:46sponsor of today's video, Fish Audio.
- 4:49Now, anyone who's tried AI voiceovers
- 4:51knows that the words come out right, but
- 4:53the delivery doesn't always because it
- 4:55might read a joke or a warning and an
- 4:57apology in the exact same tone of voice.
- 4:59So, Fish Audio has a new model that is
- 5:01their most emotive yet, and it's called
- 5:03Drama 3. You direct it the way that you
- 5:06would direct an actual voice actor,
- 5:07which means that you can write the
- 5:08direction in plain English inside the
- 5:10text like slowly realizing something's
- 5:12wrong, and everything after that tag
- 5:14will follow that direction. So, one line
- 5:16can start calm, and it can end in a
- 5:18panic.
- 5:19>> Thanks for calling. Your order's on its
- 5:21way.
- 5:22Wait. That's not the right address. Who
- 5:25approved this workflow?
- 5:27>> And the tags work in other languages,
- 5:29too. It also generates full
- 5:30conversations. You can give each speaker
- 5:32a voice, you can mark who's talking, and
- 5:34every character keeps their own voice
- 5:36across the entire scene.
- 5:37>> Just run it one more time.
- 5:39>> Of course. That'll be attempt 47.
- 5:44>> So, if you're putting it on video, the
- 5:45streaming endpoint sends word-level
- 5:46timestamps back with the audio. So, your
- 5:49captions line up exactly with the
- 5:50speech. Now, on the build side, you set
- 5:52the model to Jama 3 in the Fish Audio
- 5:54API. They also have an MCP server, too,
- 5:56so you can connect it to Cloud Code or
- 5:58CodeX, and they also have an official
- 5:59N8N node. So, if you're building
- 6:01something that talks, like a voice agent
- 6:03or a game character or just narrating
- 6:05your videos, then try Jama 3 on the Fish
- 6:07Audio API with the link in the
- 6:09description. Huge thanks to Fish Audio
- 6:11for sponsoring this part of the video.
- 6:13Now, let's get back to it. And then, I
- 6:14added this new skill called New Roster,
- 6:16which basically just lets you build new
- 6:19rosters. So, for example, if I came in
- 6:20here and I was like, "Hey, new roster,
- 6:23and we wanted to do, you know, Elon
- 6:25Musk." And we shot that off, this will
- 6:27basically create Elon Musk's brain and
- 6:30his agent file and all of his sources.
- 6:32It will build him out and also build him
- 6:34a little cartoon character that will
- 6:35show up in the actual roster room
- 6:37itself, exactly the same way that all of
- 6:39these other roster room council members
- 6:41have been built out. So, you can keep
- 6:43scaling this thing up, and obviously,
- 6:44I'm going to let this thing build out
- 6:45Elon, and then I'm going to add that
- 6:47into the repo so you guys can all use
- 6:48Elon as well. So, while this is building
- 6:50out Elon, as you can see, let us do a
- 6:53new session. So, open up the roster
- 6:55room. That's pretty much all we have to
- 6:56do is ask for it, and I'm just going to
- 6:58switch this back to Opus here, and it's
- 7:00going to open up the roster room. We'll
- 7:01open it up in the local browser inside
- 7:03of the desktop app, and then we will
- 7:04shoot off a prompt so you guys can see
- 7:06how this actually works. Now, once
- 7:08again, you don't need this visual
- 7:09interface opened up. It will obviously
- 7:11run the sub agent roster room just the
- 7:13same, but it's nice to see. But, one
- 7:15thing that I will say about actually
- 7:16spinning up a roster room is typically,
- 7:18you want to put in a decision. You want
- 7:20to put in options, so two or three
- 7:22options. You want to give it context so
- 7:24that they can actually reason through it
- 7:26and make a better call, right? So, let
- 7:29me just shoot off a quick example. We
- 7:31are looking at hosting another in-person
- 7:33event in the summer of 2027, and we're
- 7:36trying to figure out where to do it.
- 7:37We've been, you know, analyzing
- 7:39different options like San Francisco,
- 7:42Salt Lake City, Austin, Chicago. We
- 7:45obviously want to be able to get good
- 7:46sponsors there, local incentives, get a
- 7:49lot of people there, and you know, so
- 7:51airports should be important, um,
- 7:54public transportation, good restaurants,
- 7:57good activities to do outside of just
- 7:58having the event, things like that. So,
- 8:00help me figure out where this should be.
- 8:02And do this on the roast room council.
- 8:04So, this natural language request should
- 8:06be just fine, but you could also
- 8:07couldn't come in here and do a {slash}
- 8:09roast room, and then you could just
- 8:10invoke this like that. But, we will go
- 8:13ahead and just get rid of that for now
- 8:14and shoot that off, and it should be
- 8:16good enough. One thing that this will do
- 8:17is it will help you figure out what
- 8:18would be the best council members to
- 8:20have for this specific debate. As you
- 8:23can see, it's going to load the skill,
- 8:24it's going to gather event and business
- 8:26context, so it will ask you questions if
- 8:28it doesn't feel like it yet has enough,
- 8:29and then it will actually go ahead and
- 8:30run the debate for you. So, you can see
- 8:32it is going to ask me a couple questions
- 8:34here. It has four questions that is
- 8:35asking me. And by the way, we already
- 8:37did book this event in Salt Lake City,
- 8:40June 16th through 20th. So, if you guys
- 8:42want to attend, then definitely check it
- 8:44out. But, for the sake of the example,
- 8:46I'm just going to pretend that none of
- 8:48this has happened yet. So, I'm just
- 8:49going to say not signed, open choice. It
- 8:51says, "What size and budget should the
- 8:52council plan around?" And I actually
- 8:54just skipped through those questions,
- 8:56but the last one says, "Who sits at the
- 8:57table?" And it gave me a recommendation,
- 8:59and it also let you play with it if you
- 9:01wanted to put different people in. So,
- 9:02that's part of it is thinking about who
- 9:05should actually sit on the council for
- 9:06your specific debate. But, now it's
- 9:07going to be it. It's basically going to
- 9:08run, it's going to process this. Even if
- 9:10this was closed, like I said, we would
- 9:12have sub agents that would be going
- 9:13through and doing this no matter what.
- 9:16And the reason I decided to do this with
- 9:17sub agents because when I actually
- 9:18started doing these roasts, I started
- 9:20doing this with agent teams, but the
- 9:22reason I'm doing sub agents now is
- 9:23because I found after doing lots of
- 9:25testing that agent teams was more
- 9:26expensive for a minimal amount of actual
- 9:29value add. So, doing sub agents works
- 9:31pretty well because they start to
- 9:33actually talk to each other and stuff
- 9:34like that either way. So, it works just
- 9:37fine. Now, in this case, you can see
- 9:38that we took out Dan Martell because his
- 9:40is more about like protecting the
- 9:41founder's time and we put in Naval. And
- 9:43his question is, does this build
- 9:44leverage and equity you own or does it
- 9:46just rent out your time? So, you can see
- 9:48it was great that we brought in Naval
- 9:50for this specific type of roast room.
- 9:52And obviously, last time you guys saw
- 9:54the PDF that it creates, which is pretty
- 9:56high-level, but all of this stuff is
- 9:57stored locally and you're able to dig
- 9:59in, you can ask more questions about it,
- 10:01you can inspect all that stuff if you
- 10:02want to look at the debate more deeply.
- 10:04So, it's not lost data, of course. Show
- 10:06me where all of the debates are held
- 10:09later if I want to be able to dig in and
- 10:11ask more questions about it. So, as
- 10:12they're going through round one and
- 10:13writing their opening positions, I can
- 10:14just show you real quick. If I go to the
- 10:16files and I go down here to projects and
- 10:20then we scroll down to a project called
- 10:21Founder Council right here. You can see
- 10:24that there is different sessions in here
- 10:26and these sessions will show us
- 10:27different roast rooms that we've run
- 10:29since creating this visual interface and
- 10:31we're able to actually then dig in. So,
- 10:33if I look at this one, we can see that
- 10:34there was a brief, there was a round
- 10:35one, round two, and a verdict. And this
- 10:38will hold all of the session logs from
- 10:39each of these people. Because as they
- 10:41start to write their positions, we'll
- 10:42see that all of their reasoning comes
- 10:44through even when they start to debate
- 10:46and we can view it right in this
- 10:47interface as well. So, now you can see
- 10:48they're starting to come through with
- 10:49their opening positions and we'll see
- 10:51below their name what they voted for.
- 10:53So, Gary came in with voting for A and
- 10:56gave it an eight out of 10 score. Gino
- 10:58did the same and you can see A is Salt
- 11:00Lake City and B San Francisco, Austin,
- 11:02Chicago. This one, Alex Hormozi, also
- 11:05voted for Salt Lake City. So, everyone
- 11:06voted for Salt Lake City except for it
- 11:09looks like Naval went for B, San
- 11:11Francisco. And then like I said, we
- 11:13could click in and we could look at
- 11:14exactly why, we could see the risk, we
- 11:16could see what they would do, and then
- 11:17we can keep seeing what everyone's going
- 11:19to do. And now what's going to happen is
- 11:21they're going to read each other's
- 11:22positions and then they'll start to go
- 11:24after each other and try to poke holes
- 11:26in each other's opinions. In this case,
- 11:28they all are in pretty much good
- 11:29agreement, which is nice, but if they
- 11:32did all completely agree, then the
- 11:34chair, or sort of like Claude Code, the
- 11:36facilitator of this debate, would make
- 11:38one of them argue for a different side.
- 11:40So, they kind of play devil's advocate,
- 11:41and we get to see more perspective as
- 11:42well. But, in this case, we didn't have
- 11:44a unanimous decision, so the debate will
- 11:46continue as expected. Now, what I think
- 11:48is really fun about this is we're able
- 11:51to leverage the subject matter expertise
- 11:53of other people. And humans are really
- 11:56good at their domains, right? So, like
- 11:59for example, not build a world-class
- 12:02copywriting agent. It would be tough for
- 12:03me because I don't have a copywriting
- 12:04background. But, what I could do is go
- 12:06find someone who I think is the best
- 12:08copywriter in the world. I could extract
- 12:10their frameworks, examples of their
- 12:11copy, how they think, how they decide
- 12:15what to write, and then I could put them
- 12:16into this council. And when I have
- 12:18questions about copy, or you know, how
- 12:20to market something, I could leverage
- 12:22them inside of this roast room. And it
- 12:24doesn't mean it's going to be perfect,
- 12:25right? It doesn't mean that whatever
- 12:26this roast room decides, I blindly take
- 12:28and accept. But, just getting in this
- 12:30mindset of thinking about who has done
- 12:32something that you want to do. Who has
- 12:34done something and proven it. Now, how
- 12:36can you go out there and study what they
- 12:38did and how they got there? And
- 12:39basically just try to follow that path.
- 12:42I think it's a lot better to do that
- 12:43than just trusting Claude's gut, or
- 12:45trusting your gut sometimes. Like,
- 12:47following someone that's actually walked
- 12:48the path you want to walk is a pretty
- 12:50good signal, right? So, doing these
- 12:52councils helps me just look at things
- 12:54through different perspectives, and you
- 12:56know, think about things that I might
- 12:57not have thought about because Paul
- 12:58Graham brought up this idea, or because
- 13:00Naval brought up that idea. And so, even
- 13:02outside of the bounds of this roast
- 13:04room, for example, I think that just
- 13:06thinking like that when you're building
- 13:08out other skills, or you're trying to
- 13:09automate different processes, I think
- 13:10thinking like that
- 13:12is a really good way to think. But,
- 13:13anyways, you can see they're actually
- 13:14coming through now. They're deciding on
- 13:17final votes. And you can see that we're
- 13:18probably going to win, you know, the
- 13:20Salt Lake City. It's 7.9 out of 10 right
- 13:22now, and that is probably the verdict
- 13:24that we're going to get. But, of course,
- 13:25we could go through and we could read
- 13:27everyone's key points. As you can see
- 13:28here is Hermosi's key point, here's
- 13:30Wickman's key point. You can see here
- 13:32Bezos conceded to Gary Vee and Graham
- 13:34about them being right on certain
- 13:35things, but also disagreed the most with
- 13:38Naval. So anyways, that's how you could
- 13:39sort of start to read through some
- 13:40things in more depth. We can see that we
- 13:42have the PDF now, which I'll go ahead
- 13:43and download real quick. And now we're
- 13:45able to look at the fact that Salt Lake
- 13:48City won, which is actually pretty cool
- 13:49because that is where we decided to have
- 13:51this first in-person event. Now anyways,
- 13:54that is done. We could obviously do more
- 13:57rounds if we wanted to. We could bring
- 13:58another council members. And now going
- 14:00back to check on Elon, you can see that
- 14:01it created the actual portrait for him
- 14:03and it created these different views so
- 14:06that we can see what it will look like
- 14:07when he's actually talking. And then if
- 14:09I go to the repo, we can see that we now
- 14:11have Elon added in here. If I go to the
- 14:13brain, or not the brain, the agents, we
- 14:15can see right here we have Musk. And
- 14:17then of course in the actual brain,
- 14:18because it's supposed to do that through
- 14:20the new roster skill, we have a brain
- 14:22for Musk. Once again, it has hot books,
- 14:25frameworks, positions, principles,
- 14:26sources, stories, and a voice.md. So we
- 14:29are now set up and we are able to use
- 14:31Elon Musk inside of this Roaster
- 14:33Council. If you guys want to grab this
- 14:35GitHub repo and play around with it,
- 14:36then just go to my free school
- 14:37community. This will be linked in the
- 14:38description. Go to the classroom and
- 14:40then grab it in from the AI skills or
- 14:42from the all YouTube resources. All of
- 14:44my other free GitHub repos, skills,
- 14:46prompts, guides, everything is in the
- 14:48free school community. So anyways guys,
- 14:50that is going to do it for this one. I
- 14:51hope that you enjoyed and you learned
- 14:52something new. And if you did, please
- 14:53give it a like. It helps me out a ton.
- 14:55And as always, I appreciate you guys
- 14:56making it to the end of the video and
- 14:57I'll see you on the next one. Thanks
- 14:58everyone.
About this transcript
This page contains the full transcript of /roast-room: How I Use Claude to Think Like the Top 1% by Nate Herk | AI Automation, generated from the public captions YouTube serves with the video. The transcript has 3,577 words across 498 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.