I Built the Ultimate AI Board of Advisors (free claude skill)
What this video is
⚡ a 15-minute video, readable in 60 secondsThe creator built Roast Room, a Claude Code skill triggered by a slash command he calls "slash roast room," that runs a real business decision (an Austin conference plan with four options) through a simulated three-round debate among seven advisor personas, Jeff Bezos, Paul Graham, Alex Hormozi, Gary Vee, Dan Martell, Gino Wickman, and Warren Buffett, each built from markdown "brain" files of that person's public work and run as sub-agents reporting to a chair sub-agent. The system scores and weights the options, lets advisors argue and vote, and outputs a verdict PDF with next steps and kill criteria; a separate web interface shown on screen is described as an optional, cosmetic way to watch the debate, since the skill can generate the full report without it. The creator demos the system live on his own decision and also shows a library of past sessions he can replay, indicating this is a system he already uses rather than a single one-off demo. No model name beyond "Claude Code" and no on-screen prompt template text were given in the source.
System does [00:00]: Roast Room is a web app the creator demos on the "Openings" step of a decision session, showing seven caricature advisor avatars: Jeff Bezos, Paul Graham, Alex Hormozi, Gary Vee, Dan Martell, Gino Wickman, and Warren Buffett.
Key takeaways
- System does [00:15]: The debate topic shown is a real decision, an Austin in-person conference planned for June 2027, with four options (A, B, C, D) the user submitted.
- System does [11:02]: Claude Code already has the user's context loaded into their "AIOS" and will ask clarifying questions if it needs more before sending the decision to the council.
- System does [12:25]: The Roast Room is set up so the AI does not just agree with the user, since each persona has its own principles and way of thinking, so it may validate the user's lean or try to convince them otherwise.
- System does [05:00]: Users can view past sessions, see all PDFs created from prior roasts, and replay the council on past decisions.
+ 47 more takeaways
- Stack [00:27]: Roast Room runs as a Claude Code skill triggered by a slash command the creator says as "slash roast room," which he says he will give away.
- Stack [02:35]: The architecture uses sub-agents (not an agent team), where each advisor sub-agent talks to a "chair" sub-agent that facilitates the debate.
- Stack [07:51]: Each advisor has a sub-agent file defining how he thinks, who he is, the question he owns, and how he makes decisions, plus links to "brain" files covering his eight rules, named frameworks, principles, positions, stories, voice, and sources.
- Stack [10:20]: Jeff Bezos's brain has a "_hot.md" file covering his worldview, his eight decision rules, and his positions on certain things.
- Stack [10:27]: The council member files are markdown files created by having agents go research each person.
- Stack [08:21]: The Hormozi brain example was built from about 124 public sources totaling over 1.3 million words.
- Stack [09:19]: The repo is described as having seven council sub-agent files, but this point names only six: Bezos, Buffett, Gary Vee, Hormozi, Martell, and Wickman (Paul Graham not listed in this point).
- Stack [10:48]: The skill can generate the full report on its own without the visual interface; the visual interface is purely optional for watching the debate.
- Stack [06:50]: An Excalidraw-style whiteboard slide shows the pipeline as: user decision, Claude Code acting as Chair, seven advisor sub-agents, saved round files, a live room app, and a final Verdict PDF, looping through 3 rounds.
- PROMPT: Not stated, no verbatim system prompt or on-screen template text appears in the extracted points; the only command reference is the spoken phrase "slash roast room" [00:27].
- Seats [01:06]: Jeff Bezos's persona focuses on one question, is this the right bet and how reversible is it.
- Seats [01:11]: Paul Graham's persona focuses on whether something is what people actually want and whether it's growing.
- Seats [01:21]: Warren Buffett's persona focuses on what the decision does to cash, risk, and long-term value.
- Seats [07:45]: Alex Hormozi's sub-agent is focused on offer, pricing, and sales.
- Scoring [01:46]: In Round 1, Bezos scored option B 7/10 (stance conditional yes; option scores A 2/10, B 7/10, C 4/10, D 4/10; pillar score 7/10; confidence 4/5; ruin risk none; door one-way), Paul Graham gave B 6/10, Hormozi gave B 8/10, and all seven advisors backed B at an average of 7.1/10.
- Scoring [02:35]: Since Round 1 unanimously backed option B, the chair flagged it as a warning and assigned Buffett and Martell to argue as hard as possible for option C in Round 2.
- Scoring [03:32]: A quoted line from the debate reads, "I've noticed that people who are right a lot change their mind a lot."
- Scoring [08:53]: Options are scored and weighted by relevance before being debated.
- Scoring [07:01]: The debate runs three rounds: opening positions, cross-review of each other's positions, and a chair-facilitated argument round.
- Scoring [07:08]: In round three the advisors vote and everything is saved into a final verdict PDF.
- Verdict [04:18]: The final verdict is to run option B as a $3,000 annual plan plus an Austin tier capped at 120 seats, with nothing signed until 50 members have paid $150,000 by January 15.
- Verdict [04:26]: The verdict can be downloaded as a PDF showing scores and the reasoning from each round.
- Verdict [04:33]: The downloaded report scored all seven advisors, and all seven voted "Amend."
- Verdict [04:33]: Paul Graham's watch-out states the presale email is the decision, if members won't pay before the venue exists, they don't want the room.
- Verdict [04:33]: Next steps include naming an ops backfill by Oct 16.
- Verdict [04:33]: Next steps include sending a $3,000 offer to all 2,500 members by Oct 23.
- Verdict [04:33]: Next steps include a checkpoint on Nov 22 at 40 paid seats.
- Verdict [04:33]: Next steps include signing the venue by Jan 15, 2027 if the gate cleared.
- Verdict [04:33]: Kill criteria are under 40 paid seats by Nov 22, under 50 by Jan 15, or churn up more than 1 point before Jan 15, any of which triggers a refund and staying virtual.
- Wiring [12:46]: You can ask Claude Code to help turn your input into a Roast Room prompt, then run the skill to start the three rounds.
- Wiring [11:35]: When asking the council a question, you give it the decision, at least two or three options, real numbers like revenue and team size, limits like deadlines, hours, or budgets, what success looks like, and which way you're currently leaning and why.
- Wiring [05:16]: The system dynamically decides which advisor personas to use, for example running only Buffett and Martell, or in another case only Graham, Hormozi, Martell, and Buffett for a "first sales hire or keep selling" decision.
- Cost [07:14]: Sub-agents were chosen over agent teams because they let one chair facilitate and are cheaper, while giving answers very similar to agent teams.
- Access [04:48]: A free version of the Roast Room setup is accessible via the creator's Skool community, under Classroom then AI Skills.
- Access [14:31]: The creator says this agent-debate approach can be found for free inside his free Skool community.
- Also in video [05:54]: A separate free offer, "The First Client SOP," promises a first client should cost you money and shows ROI math making a free project worthwhile.
- Also in video [05:54]: The SOP includes a 5-minute video answering "can this person actually deliver" before any money changes hands, with a minute-by-minute script to record it in one take.
- Also in video [05:54]: The SOP addresses having zero case studies, claiming you can build an alternative in an afternoon.
- Also in video [05:54]: The SOP gives two copy-and-paste messages to send to 10 to 15 people you already know to land your first project.
- Also in video [05:54]: The SOP teaches a "50-5-2-1" cold outreach formula and a four-line proposal structure positioned against a generic "Dear Sir I am an expert" opener.
- Also in video [05:54]: The SOP claims three to five retainer relationships, not 30 to 50 one-off gigs, can become $9K to $25K a month within a 90-day timeline.
- Also in video [06:06]: The free offer is called "The First Client SOP," delivered instantly by email as a PDF for $0, with a link in the video description.
- Related idea [13:35]: As an example, if building an app without security knowledge, you could build a security agent, tell it to research articles, books, and YouTube videos, and store that knowledge in a skill or agent brain to query later.
- Related idea [13:54]: Specialized agents outsource thinking, but recommendations they return should still be processed and understood, not blindly accepted.
- Related idea [14:09]: Building specialized agents increases confidence in answers, especially when having them debate each other.
- Related idea [14:17]: As an example, a security agent and a CTO agent could discuss a topic together and bring a verdict back to the user.
- Related idea [14:25]: The stated goal is figuring out how to leverage other agents and other perspectives to help make a decision.
![System does [00:00]: Roast Room is a web app the creator demos on the "Openings" step of a decision session, showing sev](assets/nRXHWckVxoM/frame_0.jpg)
![Stack [06:50]: An Excalidraw-style whiteboard slide shows the pipeline as: user decision, Claude Code acting as Chair, s](assets/nRXHWckVxoM/frame_411.jpg)
![Wiring [11:35]: When asking the council a question, you give it the decision, at least two or three options, real number](assets/nRXHWckVxoM/frame_701.jpg)
How this brief was shaped: Personal AI Build / System Demo · confidence Medium
Transcript describes a custom Claude Code skill called slash roast room that runs multiple named advisor personas as agents to debate a submitted business question, and OCR shows the actual system internals (agent counts, commands run, model Opus5.5High, session cost, seat scores) confirming a built AI system being demoed rather than a straight tutorial.
The lens sets this brief's structure, never its facts — every claim is held to the same citation and fact-check standard.
Jump to a moment
Their links, sorted & clickable
🏛️ Communities & courses2
Get the FREE Roast Room councilHow I built an AI agency from 0 to over $1 million per year📢 Sponsored / affiliate1
Code NATEHERK for 10% off VPS (annual plan)💼 Sponsorship & business1
Sponsorship InquiriesWas this brief useful?