# AI Boardroom — Agentic Society Members · Clean Transcript **Date:** 2026-07-31 · **Type:** AI Boardroom **Location:** Industrious East 7th, 924 E 7th St (Conference Room A), Austin, TX **Duration captured:** ~2h51m (in-room, one shared mic + two individually tagged voices) > **How this was built:** the room was captured on a single in-room mic, so > "who said what" was reconstructed by triangulating Zoom's recording against > Granola's independent notes, the host calling on people by name, and two > voices Zoom was able to tag individually (Austin Distel, Roman Bediner). > Names are corrected against the calendar invite and the member directory. > A handful of lines the room mic caught but no source could confidently > attribute are rendered as "a member" rather than guessed. ## Who was in the room | Name | Business | Notes | |---|---|---| | Austin Distel | The Agentic Society (host) · The Tox | facilitator | | Justin Day | Day by Day Productions (Local SEO) | dashboard demo | | Roman Bediner | Agentic Society (fractional COO) | dashboard architecture | | Mike Couvillion | Alexa Guru | dashboard demo | | Oscar Hidalgo | CG Service Pros · Vindex Consulting | new Agentic Society team member | | Prashant Vanka | Agentic real estate, Austin | | | Will Sayasenh | Four28 Creative / Agentic Society video | | | Cristian Aguilar | Agentic Society (social/content) | | | Darby Rollins | Agentic Society (Chief Agent Officer) | confirmed for the next session | --- ## Transcript ### [00:08:51] Why today is about agentic dashboards — Austin Distel Austin opened by framing the day's main topic: agentic dashboards, starting around 10:15 after housekeeping. "I'll bring up why I think that this is gonna be really important, and why I think that we should have a discussion about this: as founders, CEOs, executives, your decisions that you make are wildly going to impact the direction and size of your company. And if you make those decisions without being informed correctly, then you're gonna make maybe the wrong decision." He put the moment in context: "In the past, it used to take an operations engineer to build out Looker dashboards and to really bring all of those APIs and different tools into one place. Now you can do this in half a day. And in the past, you used to be limited by just what these other tools could do — now you can build a dashboard that gives you the one thing, whatever system you follow — EOS, YPO, or any other KPI-based system." He polled the room on who already had a hub dashboard built, then set the format: "This isn't a prescription. I'm going to share how we did ours, and what I recommend, but also everybody here can learn from each other. The beauty of the mastermind is steel sharpens steel." He also told the room what this very document would become: "By the end of today's discussion, the transcript that's being recorded right now will end up becoming the ultimate playbook for you building agentic dashboards. From every tool that's mentioned, every recommendation, everything that you've ever encountered, overcame — so that you can learn from everyone's collective experience and solve this for yourself way faster." ### [00:11:18] A lighter Rose & Thorn — Austin Distel Before the announcements, Austin changed the room's own format, live: "I have heard feedback from — and you guys know I'm constantly listening to your feedback, and I'm documenting all of it clearly, and implementing it — is the Roses and Thorns is really cool, but the most valuable part is knowing who's in the room and their businesses, how they're using AI. The three-part system of Roses and Thorns is a little long, now that we have so many members. So, rather than spending that much time going around — it ends up being an hour at this point for 12 members — now we're gonna do one win, keeping it to about two minutes." With a full agenda ahead, the room moved from there straight into program updates rather than a full individual round this session. ### [00:21:29] The new cadence: 1st and 3rd Fridays — Austin Distel Austin thanked Lauren (not in the room) for feedback on scheduling that "improved the model for everybody," then walked through the fix. The old pattern of two boardrooms a month collided with the calendar: "The way some months work out, there's three Fridays, four Fridays, or five Fridays. It created, basically, an overlap and an inconsistency in the scheduling over the course of a 90-day period." The new rule: "Boardrooms are on the first and third Friday of every month. It's not saying it's every other week — it's saying it's always on the first and third Friday of every month. There are certain months that will have five Fridays. That means that on that month, there would be a two-week break. But it's always the first and third Friday." What surfaced along the way: "I actually can do a boardroom in the morning and the mastermind mixer in the afternoon." On 1st Fridays the boardroom ends at 12:00–12:30 and the public Mastermind Mixer runs downstairs after — members are invited but not required. "That also makes it easy on everybody. That means you only have to come here twice a month instead of three times a month, if you wanted to go to all three events... if you wanted to go to both of these, you save yourself an entire drive, and you can work remotely on a Friday twice a month now." He asked the room directly whether that aligned; the answer back: "That works." One member added, "I was definitely getting FOMO sometimes." Two logistics decisions got settled in the room: - **Start time.** Austin called a show of hands on 9:00 vs 9:30 AM; the room voted to keep 9:30. "9:30 is a good time to start. Got it. Cool. Thank you for making that decision." - **The room itself.** "We're moving downstairs to the same boardroom that you see downstairs that has 40 seats. This 16-person room — it's even cramped today, right? I see y'all's elbows are hitting each other." The move starts the following week. RSVPs now run on two tracks: the boardroom lives on the Google Calendar invite ("make sure to say yes I'm going or no I'm not going, which just means physically you're going to be in the room... If you're joining online, it's okay to say no — you're always going to get the recording"), while the Mastermind Mixer runs on Luma, publicly, with members getting first dibs on seats. The mixer now requires an application: "We have now over 400 people that have requested to join these... Not just anybody can come to these. They have to apply, and then I review their application, and then they get accepted, and that retains their seat." A consistent show-up rate of roughly seven in ten against 45 released seats keeps the 36-seat room full — and yes, "I will be providing lunch for the mastermind, so if you do attend that, there's tacos for you." ### [00:25:25] Cohorts, sprints, and Parkinson's law — Austin Distel Onboarding is moving to a monthly cohort model — new members officially join on the 15th of each month and move through a 12-week cycle together. "We create a starting point. We say, hey, you're at level one right now, let's get you to level two in 90 days or less. We need to build you a company brain, a dashboard, we're gonna start building AI agents, get your team set up on Claude Code. By the end of this 12-week period, you have at least one AI agent built, solving a problem in your business, a main constraint. And you have now created an ecosystem of which to build upon." Why 90 days: "Quarters work for a reason. It's long enough to see results. It's short enough to put pressure on getting to the goal. Parkinson's law says that getting work done grows or shrinks to the time allotted to it. If you say you're gonna get your chores done in the next two weeks, it'll take two weeks... So 90 days is the time allotted for your growth in AI." He framed the levels like belts: "Everyone's going to be at a different starting place, whether you're a blue belt, brown belt, black belt. Some of you guys are already black belts, and you came into this because you want to surround yourself with other black belts. At the end of 12 weeks, we hope to level you up multiple belts." The container structure got a clean summary later in the discussion: virtual members get the online-only **Agentic Academy**; society members get the in-person sprints, the twice-monthly boardrooms, co-working, and everything launched into the community — the playbooks, the Agentic OS, the AI agents the team builds; and private VIP clients of the new **Agentic Agency** get one-on-one installs ("outbound prospecting agent, an SEO agent... you are in the room with us, collaborating, because you need to know how to manage the machine"), with the explicit rule that everything learned flows back down: "I would never accept a private client that wouldn't allow me to prepackage whatever we result to the community... If something's out of alignment, it's not worth my time to distract me from the core, which is the society." ### [00:28:45] Hackathons: two days, three tracks, demo day — Austin Distel "It's one thing to say you're gonna get your work done — it's another thing to show it to the world. We know accountability drives outcomes. So, what would it look like if we had a two-day hackathon for our membership?" The shape: one hackathon every 90 days, two days, downstairs in the 40-seat room, with likely two to three tracks — "Hey, I want to build my company brain in two days or less. I want to build an outbound prospecting agent in two days or less. I want to automate my content workflow in two days. And then you'll collectively get together, and you build side by side." Day one runs from a 9 AM start: "We get together, and we talk about the menu items, who's going to be doing what. By lunchtime, you have defined the scope of work — you've gotten really clear, and I'll help you with some exercises on that, of what definition of done looks like, of what recommended tech to build it on... you create a PRD so that you know what you can give your agent and scope it out. And if you ever need support, just raise your hand, and then we walk over to you. That's the value of us being in person." His staff — and possibly guest facilitators from Austin's AI-builder scene ("you may see that I'm recruiting") — circulate to help install and customize. Day two is polish, ending in a two-hour show-and-tell: "Demo day at the end is all about — you get five to ten minutes to showcase what you built, how you built it. And guess what? The entire room is recorded." The recordings compound two ways: "What everybody else has built [goes into] the intelligence of the playbooks... and it gives a menu of additional options of what y'all could build. You get inspired by what Justin built, you get inspired by what Prashant built... and now everybody in the room that has already built it is also available for support. The beauty of a mastermind, right?" The room's reaction: "Love it." One member pitched an addition — rotating people's laptops up on the big screen while they work, so everybody's involved. Another reflected on the prior week's session: "Last week was really impactful for me... everyone working on a similar project, the company brain... It saved me a lot of work this week." For some members a hackathon will be a kickoff in their first month; for others it's a graduation moment in their last. ### [00:34:23] Going all-in on Skool (and what Slack becomes) — Austin Distel "Ninety days ago, I asked, do you guys want to be a part of Skool or Slack as a main communication channel? I haven't seen more or less engagement in one versus the other. So I'm making the decision to go all-in on Skool, because it helps our team scale, keeps everything centralized." Slack isn't closing — it stays open as a members-only social space with elevated access to the team — but official posts move: "Going forward, you can just kind of see that me and my team are always posting now inside of Skool. We're going to be doing announcements inside of Skool" — including the recording of this very session. He walked the channel map: the mastermind channel is the general thread; **Wins and Gems** is the bragging channel — "You're building some cool shit, and people should see that. Post a Loom video, post a link to the project that you just got completed." His examples from the same week: Oscar landing significant new business "by having an AI agent do outbound prospecting on a list that he bought — and now landed a couple apartment owners in Austin to redo their HVAC system," and Justin's build: "Justin built a hive mind dashboard as a central operating system, like a Jarvis, for his business. Screenshot whatever you want in that and say, guys, this was a total unblock for me." "I would love to see, in 90 days, a frickin' huge stream of everyone's biggest wins. Constantly... Ultimately, I want you to feel like this is that place that people get you — they get what you're working on, you have a shared language, and you're sharing your results." **Questions** is for strategic asks, not quick lookups: "This isn't just something you would ask Claude. This is like, hey, I'm thinking strategically... what platform do I build my dashboard on — Vercel, Lovable? I'm going into a big project, how do I want to build this? Hey, why are you guys switching over from OpenClaw to Hermes? Great place to talk about that." ### [00:37:39] The Agentic Show — Austin Distel "I'm launching the Agentic Show — clearly AI made my slides," Austin joked, before explaining the gap it fills: "This space kind of doesn't work that well for demonstrating what you've built. And the downstairs mastermind doesn't really either — it's a lot of talking, but it's not a lot of showing. So the Agentic Show is literally designed to be podcast-style, in a nice space, with a high-quality HD screen share — a 30-to-60-minute masterclass with our community members and with influencers in the AI space to showcase what they're building. These become masterclasses and playbooks inside of the Academy, and I can crank these out." The pitch to members was identity-first: "I want to put you all on a stage, a pedestal — the Agentic videographer, the Agentic accountant, the Agentic whatever... When you get results, when you have case studies and you want to share that with the world, this is a platform in which I invite you onto stage." He was explicit about the underlying philosophy: "I am not trying to build a guru-based business... Y'all are the stage. This is the really frickin' smart community... 'Here's how I made my restaurant agentic' — and now other restaurateurs can maybe look at you and say, that's a really cool model, maybe I could work with him as a consultant." ### [00:41:30] Housekeeping: August 21, and testimonials at noon — Austin Distel Two last items. First: "Darby will be leading this room on August 21st" while Austin travels — he'll be back in the seat in September. Second, testimonial videos, and the marketing philosophy behind them. Austin told the story of Proof, the company he took through Y Combinator before Jasper: "The whole idea is the best marketing doesn't say 'I'm the best' — it's where you put your clients on stage... The best copywriting is actually your client's voice, not your own copywriting. We built a software that shows [social proof] in real time on your website, and that alone boosts conversions by 15%. It's a psychology principle — social proof. A line outside your restaurant says, hey, we're popular." He pointed at the coffee shop on South Congress that deliberately pushes its line outside, and at Proof's most famous customer: "Oprah Winfrey was a client of ours, and she showcased her meditation daily with Deepak Chopra — 80,000 people are meditating with Oprah right now." "The more that I can put my customers on stage, instead of making our company the hero, y'all are the hero. So my ask is: if you would like to be one of the faces of Agentic Society, to share your story, to share your results, me and Will are gonna be downstairs after this today from 12:00 to 12:30, recording testimonial videos... Five minutes is all the time it takes. The goal is that we're sharing outcomes, not activities. Results, not just theory. That's the whole premise of Agentic Society: what the heck are people building in Austin, Texas, to grow their businesses with AI?" Several hands went up; one member asked for a two-week rain check: "Next week, I'm gonna have my company brain built out — I don't want to sound dumb on it." (Austin: "Don't worry, I'll go after you. You'll feel smart.") ### [00:45:40] Open floor: alignment, hiring freezes, and who's becoming what — full room Austin opened the floor for reactions. One member called out the alignment piece specifically: "I love the alignment piece... that could easily be something where you go, hey, listen, I'm going to do this on a different tranche. But to include everybody — that is a great opportunity." Another endorsed the hackathons: "The hackathons are gonna strengthen the community, because we're all going to be in the trenches together, building something cool." A third was already planning: "I'm building my bank of questions for Justin." Austin restated the ownership philosophy underneath the whole product: "As a philosophy, you own your shit. You don't leave it inside of Claude — you own your repository, you own your skills. You can migrate from Anthropic to OpenAI. You own your company brain — it's not hosted on someone else's land... Building AI employees is the start, it's not the only part. A company brain is an AI system that you need to have. An agentic dashboard is the system you need to have. It doesn't necessarily mean that's an agent, always — but it means that you're using AI to power your business." Then he probed the room: **who's on a hiring freeze?** The answers drew a nuanced picture rather than a purge: - Oscar, from the HVAC side: "It becomes to a point where even industries — for my example, an HVAC company — doing AI and scaling through agents, my competitors are spending four times as much to do the same things. At this point I see it as: are you embracing the internet, or are you embracing social media? We've had these spots where it's necessary." - One member drew the line by function: "I have a hiring freeze on operations, execution, and internal things like that — but sales, that kind of part, still needs people. We're not saying replace everybody. It's refocusing — understanding where the leaky buckets are, and increasing your margins with the power of data." - Justin is hiring *because* of AI, not despite it: "Our velocity of building product is held back by how quick our developers can actually create code. Now we're using agentic to amplify that. So now I'm scaling my velocity, versus trying to reduce costs — more product out the door. I have five job positions open right now. The most important one we need is a React developer, if you guys know anybody." - Another member, who runs software consulting and also serves as COO of an interior design firm ("as non-technical as you can get — they're all creative types"), described cutting operations costs by a third with AI: "Procurement used to be five people, it's one person now... it's just that one person is scaled up to do the work of five, and if we hire a second, it's just gonna be because we want to let her go on vacation." The mechanics were unglamorous and effective: industry software with no API, only Excel exports, so "I stood up a web service that actually has an LLM on the back end — she just throws it in a Slack channel, and two minutes later it comes back and says, here's your file." Project managers who used to handle three new-build projects per designer now handle about six: "That means we can double our revenue without increasing our headcount... it's never a cost, and it's always going to be covered by revenue." Austin then ran a quick identity poll — CEOs, COOs, CTOs, CMOs — and laid out where the community's playbooks are headed: "The Agentic CEO means that you're doing CEO things, but you now have orchestration... The COO tends to be an integration type; the way they'll build with AI is totally different. Roman is my COO, and he's built so many internal systems to grow our company — project-manager Slack agents, agentic dashboards, onboarding, structure. I'm building the company, he's building the structure... You need an Agentic CEO, CMO, and COO, because that trifecta will grow your company outstandingly." His diagnosis of most marketing teams: "So many marketing teams are still very focused on being the individual contributor on their team, and not scaling the organization." He grounded it in his Jasper years: "We had a 12-person company doing about fifty million dollars a year in revenue. We really liked that size... it was very much, how do we add on another marketing channel in the same amount of work week? We added on SEO, paid traffic, affiliate marketing, SMS, email, webinars, content, YouTube — all with two people. How do we do this? Because we build a system first, and then we scale it." A member summed up the ambition back to him: "A hundred million dollars with 20 people or less." Austin: "That's the plan." The challenge he left the room with: "What would it look like if I were to hire five people that happen to be AIs? If I were to post five job posts, here's what these five job posts would be that would give me freedom, that would help me scale, that would help me double or triple my business... What are some of the things that you can implement in your business that would double your output, increase your performance, without having to scale up your team — while maintaining what you love about your company?" ### [01:00:09] Will Sayasenh's Clipper Austin turned to his own team first, as proof the philosophy applies internally: "With every person that's on my team, they all know they have to be using AI tools in their business — and then they also have to surface the ways they're using that in the community. Will, how are you using AI in videography and content marketing?" Will: "Currently building a Clipper — very similar to Opus Clip, but I'm coming at it in a way where it's reading more than just the transcript itself. It is reading tonality, which is the biggest piece that is really missing — and I'm pretty close to nailing that." The pipeline is fully hands-off: "Anytime Austin were to drop, for instance, this Zoom video — as soon as that drops, we created a trigger where it'll send it to a Google Drive that I have access to. From there, my dashboard will recognize that, upload it to my drive, and start clipping it and reading all the transcripts." Austin used it as a management lesson: "Give your teams the vision of who they're becoming — the Agentic videographer... We have a big project, we come in AI-first: creating the framework for how are we going to create, capture, edit, produce case studies at scale. Now that's on his roadmap, where he's building the case-study clipper that is automatically producing, branding, all of that — and so now he's able to scale up more volume than one person should be able to handle." ### [01:02:29] Cristian Aguilar on running social with AI Asked how his social workflow changed over the last 30 days, Cristian described a system built on the community's own knowledge base: "I built these different workflows to help me come up with all the creatives. I used our company brain — that has all the context of everything, all of our members within there — to help me create a 30-day calendar. I have full stories, carousels, reels, all planned out and thoroughly mapped out, and campaigns all thoroughly put out there. It helps me create shot lists of everything, so I pretty much have clarity on everything that needs to be created — and the rest is just creating it." On top of production: "I'm also using plugins for social media so I can get a weekly report of how everything is performing — what do I need to double down on to improve. And I'm also able to manage all the conversations, within Instagram and GoHighLevel, all within Claude — all the content and all the leads coming in." Austin asked what changed when he took on the identity of "the agentic social media manager." Cristian: "Before, it was just every day, on the day, what I need to create. Now, by owning that identity, I'm thinking differently on how do I become that. I'm asking different questions, and different outcomes are coming from those different questions." Austin's observation from the outside: "He's systems-oriented first, thinking in frameworks — what's on-brand, what's off-brand, what's in alignment. Rather than trying to crank out content, he's thinking about: how do I build a system that cranks out content?" ### [01:08:04] The dashboard block begins — Austin Distel "Okay, I'm off my high horse," Austin said, and reset the room for the next hour. "This is the small hinge that swings the big door — because you're in this seat, you have this type of role in your business, you get to make decisions. Not everybody in your business gets to make the decisions that you do. That's why it's so important that you make the right ones. How do you make the right ones? You get the right data. In the past, you had to have a RevOps person on your team that would pull in the right KPIs — this is your cabinet, this is your council." Before anyone shared a screen, he set the container: "I know that your dashboard is going to surface some private things... What I can promise you is that it will be clipped from the actual published video. I'll have the transcript, but it will be redacted of any private information. We're going to create the recommended system for how we build agentic dashboards — how we build it, what's on that — but not what yours looks like. We're never gonna surface your revenue or the way that you do things." (This document is that redacted transcript.) Members joined the Zoom room from their laptops so their screens could hit the big display, and Austin queued the order: "I will go last — and part of my goal here is to not only say why this is helpful, why I built this, what this solves, but then at the end, how I built it, how you could too, what I learned along the way. So Justin, could you kick us off?" ### [01:13:14] Justin Day's "Hivemind" — a second brain for the whole business Justin shared his screen and started at the architecture: "At a high level, this is the entire operating system as it sits. I have my second brain here — it has data sources from all the different sources of data and platforms that I use, that feed into different processors, that then feed into an intelligence loop, that then feed into an action layer. Each part of the brain has something that's processing it and then feeding it into its own dashboard. And then we have these different delivery layers, and then a feedback loop that becomes self-learning, self-growing, self-optimizing." His core design position: don't build one god-view. "Each one of these things really should have its own dashboard and its own view — and I think that goes down to what we've talked about here, having role-based AI employees. So what I've done is created modules, or dashboards, for each one of those different things." The tour, module by module: - **The Action Hub** — "This is the thing that we look at on a daily basis that tells us what we need to work on, day-to-day... it tells us what we need to do next, where the pipeline might be stale." A projects dashboard shows what he should be working on in real time. - **Executive AI** — "This pulls into my Gmail, my calendar, really everything in real time, so that I can manage everything — and then also it feeds into the brain." - **Ad intelligence and social** — Facebook ads and every social account integrated live: "showing what's working, what's not... literally every single dashboard you could possibly have, I've got it all in one place, feeding into an action hub." - **Data processors on Hormozi's money models** — "This is all based off of Alex Hormozi's money models... it takes all this data in real time, processes it, tells me what I need to be working on based off of the leads, or the offer. This one's on the lead side, this one's on the money models, this one's on the growth hub." - **A sales layer on GoHighLevel** — "This brings in all of my pipelines, all the contacts... bringing all the data in here so that it feeds into the brain, and I can ask everything in one central place." - **"Justin's brain"** — the strangest and maybe most valuable module: "I ingested all of my Fathom, my Wispr Flow, my Gmail. This is who I am at my core — what my decision patterns are, my voice, my alignment, my learning log, my handbook — so that I can take this dashboard and feed it into a Slack channel where people can talk to me virtually in real time without having to actually come to me." - **An ops monitor for his overseas team** — built on Time Doctor: "It sends me a Slack message every day based off of what people actually worked on. It actually reads the video screens... it tells me exactly what people are working on, and flags blind spots — if they're on YouTube, or if people are overworking or underworking." In total: "I've got probably over 30 or 40 modules here, and each of these have submodules. I can go for days on all this." **Q: How do you approach this — when do you look at it, how often?** "I used to look at it every day, and it's good to have, but what I found is what's more helpful is having all of this feeding into my Slack. It just shoots me a message at the beginning of every day of what I need to do. This actually helps me get out of email now — I don't even open up my Gmail anymore. Anytime I have one of my whale clients or something that's important, it'll shoot me a Slack message so that I can go back to email. This is the engine, and the output I just stick with is Slack, so that I'm not constantly in this and I can work on other things. It's the input-output." **Q: What did this replace?** "This is going to replace my COO at some point... this has replaced several employees from an execution standpoint, for sure. It's replaced the need for me to go in and run reports — [our CSM] runs our customer success, and now I just have automated reports go out. It gives her our custom dashboards for our clients. It also has a video script that automatically sends to her, so she just reads off of that. It's as easy as it could be. So it's empowered the CSM role and replaced my execution side... I'm just chipping away at where the lowest-hanging fruit is." **Q: How much of Hivemind is a repository for loose ends you want built, versus your operating system?** "I first had it just kind of be a data set... the first step was adding all data, understanding leaky buckets. But then I started feeding it the modules and the actual things that are running the company for the offers that I'm launching, so that I have a central brain that's self-learning, and it goes into this feedback loop. Every facet of my business is in this second brain and in this dashboard... I don't want to have separate brains or separate dashboards or anything. The whole goal is for the agency to be completely agentic, and the only way to do that is to have every piece of the business in here." **Q: Do you have a framework for build versus integrate?** "I've done this with so many things I've tried to recreate, and you unleash a monster of technical debt that I don't want to get into. So it's really just taking the time to slow down and have AI analyze for yourself: is this something that I need to actually create as a standalone, or is it something that I can just leverage from their platform and integrate? It's a mindset — and then leveraging AI to help you determine whether or not it's something that you create." His concrete example on each side: he won't rebuild GoHighLevel — "I'm sure I could do that with Opus 5 or Fable or whatever, but that would take so much time... that's weeks or months of a build" against a subscription of a few hundred dollars a month — but he *is* rebuilding Leadsy, the tool that one-click-connects his clients' Google Analytics and WordPress accounts, "just based off of getting those integrations, and it's an easier thing to do. What's the lowest-hanging fruit that will make impact, that won't take days or weeks, and doesn't have major complexity? Those things I'll attack." His time-horizon rule: "Is this a now thing, or is this a later thing? A year from now [you'll] one-shot build almost anything, based off of the trajectory of all these models." He also visualizes the plumbing itself: "Have it prove to you and show you how it's looping, how it's connected within your brain... you can tell these things what to do, but unless you're having it prove to you that it's doing the things, and showing you when it's broken, what needs to be connected or unconnected — that's really been a big part of growing this thing." ### [01:26:57] Defensibility, and the due-diligence story — Mike Couvillion joins in Justin's demo raised the obvious question, and a member teed it up with an anecdote about a big venture firm that keeps "a SEAL Team 6 kind of unit that will try to recreate a company they're doing due diligence on, just to see if they can — as a simple way of seeing, is this defensible or not?" Austin brought in the room's expert: "Mike does due diligence for software companies on the technical side, to make sure that what they're buying is worthy of buying." Mike explained his angle: "It's on the technical stack itself, but the part that I add to it is I also look at the behavior patterns of the engineers who built it. I can go back through all of the history of the code — if they're using GitHub or any of the Git-based repositories — and I can see: does your senior programmer check in code three or four times a day, three to four days a week, which is a behavior pattern of a good engineer, or are they just phoning it in once a week? If they have good behavior patterns, they test their code, they have unit tests — you can make a reasonable assumption: I don't have time to deep-dive now, but that's probably not as risky as I'm afraid of. Here's another section where I'm seeing a lot of anomalies, a lot of bugs, a lot of rewrites — let me go deep-dive into that." Then the story. "I had been flat-out asked: are we reproducible? That is what a PE firm is coming in and asking me. Do I adjust my valuation because I'm going to have to drop a couple of million dollars in upgrading the stack, because they haven't updated anything in two years? Or — no, this can be redone by AI in a couple weeks. Literally just had a client that we wrapped up at the end of April where, in the two weeks that we came in and knew nothing about what they were doing, we were actually able to replicate an entire product line that they had. Because it just wasn't that hard — and we were able to actually add additional features that their team was saying would take a year. And we did the whole thing in two weeks. So they were like, yeah, we're actually just gonna close that product down... they would get replaced by their own customers within a year. And so the PE firm actually backed out of the deal because of that. It is a real thing." Justin connected it straight back to his own build: "The biggest push for me on building this Hivemind and this dashboard — my goal is to sell this company, and for this to be a turnkey solution. A brain that I can just — boom. You don't need me anymore, this is self-sustaining, it's fully agentic." Austin added the buyer's-side view: "I bought a business last year, and I've sold a business — key-man risk is one of the biggest swings of that valuation. How many agencies out there are agentic? A very small pool — so they deserve a higher valuation, and easier to plug and play." ### [01:31:53] The Hormozi caution: just because you can, doesn't mean you should The counterweight came next, via a quote from Alex Hormozi's recent Diary of a CEO appearance: "People are building with AI things that they should never have built in the first place... just because you can, doesn't mean you should." The teaching example that followed (told dollar-free here): a company realized it was paying virtual assistants a modest yearly sum to do manual data entry — and then spent roughly ten years' worth of that cost, up front, building custom software to replace it. "Does that make sense? No. It's literally ten years of that spend up front... don't let the tail wag the dog. Let's not just recreate GoHighLevel to save a couple hundred bucks when that spent you a week of time as a CEO — and how much is your time worth? Do the math, it's a lot." Austin added the economics of saving versus earning: "For every dollar of expense that you save, it costs you several times that to make the same amount as profit. You could either go and make a hundred, or you could save twenty-five, and it's about equivalent." And he pointed back at the framework the room already knew: "It's that impact-and-effort matrix that I walked you all through. How often are you experiencing the pain of this problem? And when that pain shows up, how painful is it? That's the same kind of thinking that I would approach a dashboard with." **Q (to Justin): How do you think about saving versus earning in the way you look at your dashboard?** "It all goes down to the revenue perspective, exactly like you're talking. If it's not making me money, and if it's not actually helping me uncover things, then why am I doing it?... Let's get as much data as possible so that we can understand what's going to move the needle based off of where you're in the business. And if you're building something just to build it — a fun widget or module — this system should stop you from doing that. You could even have, inside of your Claude, a skill trained to tell you: does this actually make sense to build? And stop you from moving on those things." There was also a defense of play, though. Austin: "The first three months since OpenClaw came out, I was doing a lot of 'this is cool.' Well, it took a lot of learning too — you gotta play with it a lot, you gotta fail a lot. I've heard Google has this infrastructure where every employee needs to spend a chunk of their time experimenting. That's how I treat AI: I have my thing that works, and this is what I'm doing day-to-day, but I am gonna spend twenty percent messing with OpenClaw, or downloading Hermes, or breaking this new dashboard — just so I can stay on top of what's actually working, instead of just watching YouTube videos. My first OpenClaw dashboard was probably thirty tabs — and I wouldn't go look at them again. The way you've built towards action is the point: here's what you need to do, and you do it through Slack, so you don't even have to look at the dashboard." **Q: If you were to recommend the community one thing to build in their dashboards first, what would it be?** Justin: "It all starts with lead generation and really driving revenue. Integrating Facebook ads and getting all the metrics in one place — and being able to talk to it so I can understand what I need to actually move the needle with my ads — was a huge unlock. Anything surrounding building the business and growing your revenue, that's where to start — versus just building things that are, like, replacing employees. How's your lead generation going? Can you afford to spend time on this? If not, focus on the top of the funnel and driving revenue." ### [01:39:07] Mike Couvillion's dashboards — from his own dev cockpit to a due-diligence product Mike had to leave early, so Austin bumped him up in the order ("Justin, thank you for sharing." — "Cool, dude. That was awesome. It's a monster."). Mike showed two very different builds. **Build one: the developer's cockpit.** "Take this as the point of view of a CTO who's a very active developer. These are just terminal windows where I'm running Claude on three different projects; the left-hand side is the dashboard in and of itself." Even the dashboard's own existence was a build-vs-buy lesson: "When I first started this, I was building a project — I want a website where I've got three or four terminal windows and I track all these things — and I was doing this in an open-source version of VS Code, and I said, wait a minute: VS Code already has all of this. I just need to do some plugins. So my code went from a few thousand lines to about forty." The orchestrator behind it is called Night Shift (he also calls it Nighthawk): "Night Shift is aware of all my repos. It's what runs my fleet of coding agents. It tracks how much right now is assigned and running in the fleet, and I keep an ongoing, real-time update of how much the spend is on tokens. I'm using this as the API spend, even though I'm on the max plan, because the niche that I'm filling is: I use AI tools to build non-AI tools. You don't need AI everywhere — I'm building deterministic tools." Measured against revenue, his return on that token spend is running at many multiples of what he puts in — "as long as that number stays positive." The panel he watches most: "The unassigned issues. There's a reason my fleet couldn't do one of these — that's why it's a red X. Usually that's going to be something like a major architecture change — I don't want to do this without an interactive back-and-forth question session, because it's not clear." And a panel born of self-observation: "Because I'm working on so many projects at one time, I had to go out there and say — when's the last time you put the build out? What version is actually out there? Once I ask myself a question about the third time, I figure out: let me do something that lights that up somewhere." **Build two: CycleTrack + Guardian, the CTO's due-diligence dashboards.** "This is typically what one of my clients is gonna see — a dashboard for a CTO. It goes in and scans that repository of all the code changes over whatever period... how many times a day is the team checking in code, what's the commit volume. It's my own tool, I'm the only author, so my blame line is up there." What makes it different from the standard scanners, in his words: "There's a lot of tools that do this stuff. What I've done, based on my experience as a CTO for over twenty years, is: when it finds a problem, it gives you a real-world answer — what are the immediate actions you can do to remedy that. I hate dashboards that just show me a red light — it doesn't tell me what to do about it. My philosophy as a CTO with all my guys is: if I ever see an error message of 'something went wrong,' or some nebulous error 47 — you're fired that day. Give the customer something meaningful they can take action on, even if it is: this is bad, call the 800 number and start talking to someone. I don't just tell you that this metric looks bad — I tell you here are the immediate actions, and here are the structural changes you need to make to your organization to make this problem go away." His example finding was concentration risk — one engineer carrying most of a codebase. "As a boss, you need to ask yourself: are they gatekeeping and keeping everybody else out of the code, because they're the crusty old senior programmer who won't let anybody touch anything? Or — no, they're the hotshot who's actually doing all the work, and you need to be questioning why you have all the other people. Those are real-world questions you can ask." Two more scanners round it out. An AI-readiness scan: "Is your codebase — maybe it's a legacy system — in such a condition that you could actually start using AI on it? Most of the time, the answer is no. If you do nothing else — be it Codex, Claude, or anything — have it go through your code and just comment what it thinks each one of your sections does. Just adding those comment blocks will increase the value... and this will actually measure that. That all goes into the valuation calculations." And Guardian: "It scans across multiple scanning types — security, vulnerabilities, scalability, complexity — and gives you a score. Here are the findings in this particular area; here's a high, I want to do that — and it will automatically create the issue. In this case on GitHub, so it'll actually put the ticket out there, and whatever your normal process is picks it up from there. In the security things, it will output the format in the SARIF standard, so you can pull those into any third-party security scanning tools." The positioning principle underneath both: "I'm trying to fit into an area — this actually replaces several tools that are very typical in a software development organization. But it's integrated with GitHub; it doesn't try to replace GitHub. GitHub has its job. It does that better than anyone else right now. So work with it, don't try to replace it." ### [01:48:53] Austin's revenue dashboard — what it answers and how it's wired "I'll share mine, and then we'll get into what I'm currently recommending for you all," Austin said, framing the takeaway first: "Everybody's gonna have their own thing — whatever's important to you to drive the needle in your business, whether it's a restaurant pulling in OpenTable and Toast, social media engagement, pipeline. What I hope you walk away with is at least a platform in which you can start talking to Claude, and then it builds whatever you want." His own version answers one question: "The most important thing at this stage in the business is that I know how far we are from certain revenue target goals." It pulls in "Stripe, GoHighLevel, and bank statements — Mercury is very agentic as a bank, so you can communicate directly with your bank, which is pretty cool." The live math is the interesting part: "It tells me in my pipeline how far I am away from my goal. Currently I lead sales in my company, so I need to know how many appointments I am away — with my close rate, my show-up rate... It knows that I'm two appointments away from my goal for the month, and so it will say: well, you've had four no-shows this week — you actually need to book more appointments to get to that same goal. If I hire a salesperson, I know their conversion rate is going to be less than mine. It's going to notice that change in conversion rate and automatically change how many appointments we are away from the goal." Next up: paid traffic. "What's the cost per lead, cost per appointment, and cost per booked qualified show — that will tell me how much more do I need to spend to hit my revenue goal. It tells me just the mathematics of how to grow the company." Roman built it. "As COO, his job was to surface insights about the operations of my company to me — in a way that gives me my one main thing I need to look at every day: balance of bank accounts, how much cash was collected, how far are we away from that goal." The second requirement Austin gave him was self-improvement: "Part of this requirement is that you make this system an ability for me to improve it. So he built this chat interface on the side, with a RevOps engineer" — a role that would otherwise cost six figures to hire, keeping HubSpot- and Salesforce-style dashboards dialed in. "I can just talk here with it and say: hey, what's the cost to acquire a customer in the last 45 days? Look at all of our marketing spend as a category, divided by number of clients we got. It will surface that, and I can say: now add that as a card to my dashboard, so it's a live part of this dashboard. And then it will go as a ticket for Roman to review, because it doesn't automatically push — it will add it to GitHub as a ticket, he'll review it, and then push it. The other way to do that is you would just talk to Claude directly in Claude Code and say: now push this update. That's what I'm recommending for a lot of you all that don't have a COO using AI tools — Claude Code is your interface, and you're chatting with it." **Lesson one from version one: build an app, not a page.** "The first time I built with Vercel, I just went with an Astro-framework HTML/CSS site. It's simple — but it's static. If you want to update it, it's only going to be for you; if you clear your cache, sometimes it messes shit up; if you open it in a different browser, it's gonna mess some shit up. Something that I've discovered is: it doesn't take that much more effort to make this just an app — a real app — and not just a website." And bake in multi-user from the start: "We build it from day one with roles — you, your co-founder, your executives. It doesn't mean they see everything, but they at least will be available in the future to be added as a user, to have their own dashboards, or for you to share your dashboard with them. If you get into a system where it's only for you forever, you could end up having to rebuild the whole thing, and that's annoying." (His forward vision: "I will have visibility into departments. Whenever I do my one-on-ones with my team members, I say — okay, show me your KPIs, and we open up their dashboard.") ### [01:55:56] The recommended stack — Austin Distel, with the room's technical review Austin then walked his current prescription, item by item — he's packaged it as a PRD purpose-built for CEO dashboards, written for a non-technical builder: "I want a non-technical way — because I don't know how to code, and actually nobody on my team does, which is pretty cool — to build dashboards in a way that uses all of our existing tools, where we can just give it APIs and it starts connecting the shit, and it has a backend so that it actually updates and can manipulate and transform the data over time." - **GitHub** — "Don't recreate it, keep it. That's where you host your code." - **Vercel** — "That's where you publish, because it's the agentic way to build websites right now. Claude — or OpenClaw, or Codex — push really easily into Vercel, and you can see all the logs, and version changes." - **Next.js** — the base framework. - **Claude Code** — "the interface I'm building in. Maybe y'all want to use Cursor or VS Code, but I'm using Claude Code." - **Clerk** — "for authentication, and it's free. I can one-click log in with Gmail — you know how every single app has that option? It installed that thing in 20 minutes or less, just a couple clicks. Now I have authentication, everybody on my team logs in with their own Gmail, and I can say certain people can see certain things." - **Neon Postgres** — "You need a backend database that ingests revenue over time, log changes, and can look at windows — last 7 days, 30 days, 90 days." - **Untitled UI** — the design answer. "Do y'all want to be front-end engineers? Do you actually want to have a style guide? Heck no. It's a beautiful, entire UI kit... these components render wicked fast." Price, when the room asked: "$129 — one time." ("That's not bad.") "It's automatically going to bring in all these amazing components for your entire Tailwind CSS system. It's gonna be consistent — you're not gonna have odd words falling off the page, you're gonna have really smooth animations — and all you have to do is say: install this. You never have to think about design again." - **Recharts** — "for charts. Simple." He invited pushback from the technical members, and got endorsements plus additions: "I wouldn't change anything," one said, offering to share alternatives on Slack — **21st Dev** ("a great component library") and the **MUI** components. Another asked if the room knew **Mobbin**: "It screenshots every app and every website's UI, every screen — so if you ever like any one specific piece of any other app or website, just point your Claude Code to it. I love their checkout page, I love how they did this." On design workflow, Austin went deeper for the Figma users: "With Untitled UI, it also comes with an exact replica in Figma, so you can always update your components... You put it into developer mode, it literally has 'implement this design from Figma' — you copy that prompt, put it into Claude Code, and it has access to regenerate this exactly. The way designers and product designers today can just give their engineering team this is insane. Before, you would screenshot and send it, and it would kind of get it right — now this is actually giving it the raw code that developed this. I do this with our own Agentic Society website — I'm building a funnel right now; when I'm ready, I've moved all the pixels around, I just say copy this, and then it builds my entire funnel." For non-Figma people: "Claude Design now, if you guys haven't played with it, works like Canva — you can click on it, drag things, iterate, get a good design, and then export that to Claude Code so it can follow the same plan." (The room's caveat: "It burns tokens like crazy.") On the database choice, he explained the ecosystem logic: "A lot of this is inside of the Vercel ecosystem already — Vercel is involved with Neon Postgres and Next.js, it's a combined ecosystem. Yes, you could use Supabase — for example, our Agentic OS does use Supabase; Darby had his own reasons for choosing it that way." One member vouched from experience: "I recently built a travel app using this whole system, and it's working so much better than using an Astro framework, which is what Vercel recommends for just plain websites." The room's verdict on the stack overall: "It's all super straightforward and simple... a solid baseline, and you can build upon it — maybe adding a storage of some sort, depending on what they're doing with their dashboard." The PRD ships to members: "On the boardroom page, you'll get access to this PRD. Read through it, and then you can just say — hey, Claude, give me the onboarding steps to set this up." ### [02:04:55] Sidebar: websites that AI can actually read — Justin Day Austin used the moment to ask Justin the adjacent question the room needed: what are you recommending for the future of websites being read by AI? Justin: "Everything on Vercel is fine. I've sent the SEO playbook in the Slack channel — make sure you've got your schema, it loads fast, all that fun stuff. There's a free tool, Google PageSpeed — enter your site and it'll give you a breakdown of exactly what needs to be fixed on your website, if it's agentic-ready, and just feed that to Claude and it'll fix it. But definitely don't use, like, Base44 — Google can't read that. You want to make sure everything's public, it's got a robots.txt you can follow, it's got index on. From a hosting perspective, [Vercel] or Cloudflare." On the file the bots now look for: "LLMs.txt. Just put your basic business descriptions in an MD file." One member described the payoff of migrating off a page builder: "I recently went through the process of moving my website from Webflow to this similar framework. One of the benefits, for somebody who hasn't gone through that process: I just give that SEO playbook to Claude, I change my thing on GitHub, and it gets posted automatically — now I'm not relying on an agency to make changes." And for anyone stuck on a legacy platform, another member offered a war story: "My wife will not move off of Wix for a website. I've lost that argument. So — buy-versus-build decisions. But literally yesterday, I just said: do an SEO audit on this Wix website, with a Claude plug-in extension [of marketing skills]. It opened it up, went in, and said, let me change these three or four things — here's your report, before and after. If you're stuck in that situation, there are tools that will even work with these other platforms." ### [02:08:03] Roman Bediner: self-healing, or your dashboard is already dying Austin handed the architecture segment to the person who built his dashboard: "Roman — you spent a lot of time building that agentic dashboard for me. Any learnings along the way? Recommendations for the community?" Roman's first principle, verbatim: "By default, the dashboard's just gonna be static. So you need to consider self-healing — it needs to write the logs and read the logs, and if there are errors, it needs to try and fix itself. And then I think the highest bang for the value you get is if it's connected to your other agents that you are already running, so it integrates into the whole hive-mind concept — that's really what makes it powerful." Why it matters more than another chart: "I would very highly recommend self-healing, because it connects to a whole bunch of data sources — Stripe, GHL, whatever. And those API connections sometimes just die. Sometimes you just run out of Claude usage, sometimes you have outages. Building observability into these dashboards is probably as equally important as surfacing the key metrics." Austin asked him to define the buzzword properly: "Walk me through self-healing. I know this is a hot word in agentic AI — what does that really mean to you?" Roman pulled up the flow — memory, a cron sweep, gap detection, then the heal, with constant retries — and gave his three-stage definition: "First, it needs to identify that something is broken — and there was a comment in the room earlier: don't tell me that it's broken without telling me how to fix it, and that's exactly the key. Second is: read the log to figure out why it broke. And then three: have a system in place that can actually make a corrective action on it. That's really what self-healing means. It's modeled kind of after AWS architecture in terms of self-healing — it's just a continuous loop that keeps running. Outside of that, it's just the workflow that you build. In terms of the stack, Vercel is the key — they've made a tremendous amount of progress." Then the cost architecture, which he flagged as its own discipline: "You will probably see this chatbot — this RevOps agent has intelligence to route queries to a specific model, based on cost. If I'm asking it, hey, what's the definition of this metric? — it's just gonna route it to Haiku. If I'm playing with the what-if scenarios, then it's gonna route that to a more sophisticated Opus model. Cost — if you're gonna run a bunch of cron jobs and a bunch of API calls, you want to be cost-efficient. Consider caching optimization, which Anthropic gives really good documentation on, and model routing." The dashboard exposes it: "If you take a look at the company vitals, it tells you the Anthropic spend in real time, and it gives you the cache efficiency — what you saved, and what you're projected to save. While Austin is looking at the revenue tab, I'm looking at the vitals tab and the system-health tab. Killing two birds with one stone — this tells me how efficient the cache is working and if I need to make any optimizations, and this tells me if the system's operational." He shared two design details worth stealing. The origin: "It all stemmed from one question — Austin challenged me with: how many appointments do I need to book to reach my monthly goal? It all stemmed from there." And data provenance: "Every tile here is flippable — it'll tell you exactly the data source it's coming from. And there are labels, because there are some metrics coming directly from the system you're importing them from, some metrics that are estimated, and some that are inferred. You're already doing some calculations on the data you're pulling in, and sometimes it could be wrong — so you want to know exactly which metric is derived, versus inferred, versus just a static pull." **Q: Is self-healing already built into that PRD — and how do you suggest we add it if we're building from the template?** Roman: "To be totally blunt, Claude Code is really good at it. You just start a conversation, and you'd be explicit that every service in the repository needs to be self-healing. Then you define what self-healing is with Claude through a conversation, and it builds — suggests — the automated jobs that perform those actions. It's hard to prescribe the actual methodology, because the methodology is based on your specific dashboard and your specific workflows. But in a nutshell, it's literally just a conversation with Claude Code." Austin added: "I have a playbook, I'll send it to you." ### [02:16:21] Self-learning and "dreaming" — Justin Day & Roman Bediner Austin widened the lens: "What does real agentic mean? Self-healing is one of those values about being agentic. Self-learning is another one — how do you have what it's doing also improve itself? Justin, your idea of dreaming at night also inspired Roman to build dreaming into this. Could you explain your concept around dreaming?" Justin: "What we do to self-learn is — at 2 AM, hopefully we're not all working then, sometimes we are — it has a cron job to review all the session logs that we've had for the day, and it builds out lessons that it's learned from that, and then it builds out instincts, so that it can pull out all the patterns and recognize what you've worked on that day. So you're not repeating mistakes the next day, and the system is getting smarter and smarter. But it needs a systemized process so that it actually learns — you're not just dumping session logs into your brain. At night: read those session logs, build patterns, self-improve the next day, and just continue to iterate." He's packaged it: "I have a full playbook for that," which he offered to send on the spot (one member's response: "That'll do the job"). Austin closed the loop: "Agentic means self-learning, self-improving — not just a static, one-way road. It's a loop. You gotta close that loop." ### [02:19:08] Staying current: Buzz, research agents, and a standing "State of the Union" — full room The conversation drifted — productively — into how anyone keeps up. A member asked Justin if he was testing **Buzz** yet, Jack Dorsey's new open-source release: "It's like Slack, but it allows you to actually customize your agents. You can have agents in there — call them whatever you want, they can have their own SOUL.md file — and you can choose the provider, so that agent can use Claude Code, or Codex, or whatever. It's open source, so you just download it, and it recognizes all your stuff." Another member confirmed it had just gotten deeper: "They just updated it this week — Hermes is now fully integrated. There's a side where you can communicate with your agent, so you can just tag them." That raised the meta-question one member voiced for everyone: "With so many skills out there, and resources people are putting out — what's everyone's process to finding what's out there, versus creating it yourself?" One answer from the room, an agent solving the problem itself: "I have a Hermes agent that's just a researcher. It searches through X and Reddit every day, twice a day — 7 AM and 6 PM, a cron job sends me an iMessage with the top five things that you should read. And then if I think it's interesting, I dive deeper. It's impossible to keep up with otherwise." Another example of why curation matters — a tip making the rounds that week: "The founder [in a YC conversation] a couple of weeks ago talked about deleting your global CLAUDE.md file every six months so that it refreshes... That seems terrifying, but based off of what people are saying about the rot it can have, it makes sense." The room immediately engineered it into policy: "I would think it's a self-learning type of thing too — every three months, a cron job that reads through your entire second brain and sees if the main CLAUDE.md and your agents need to be updated." "Would there be a backlog anywhere?" "Keep the other one — in case everything failed, you'd still have something to refer back to." That fed a concrete program suggestion: a recurring 10-to-15-minute "State of the Union" — what shipped across the ecosystem in the last month, what's real and what's noise — with the members applying it live. ("A filter for recent news — and what's bullshit." "One thing that has all of us in common is that we are using AI to do awesome stuff.") The room agreed the Tuesday call is the natural home. Austin described its current intent: "I named it an AI Breakthrough call — in the middle of the week, if you have a problem you're trying to conquer, you can hop on this as office hours. My AI building team is dedicated to that call; when you get stuck, you have Darby and Roman there to help support and break through." The improvement requests: a pre-call questionnaire ("a couple days before, everybody gets a notification to remember to submit your question... and the day of, everybody gets: this is what we're gonna discuss today"), and one member suggested borrowing the Level 10 meeting structure from EOS/Traction — fixed time slots: minutes for good news, a block for KPIs and the news refresh, then the issues list. Austin noted the recording habit that already works: "Not everybody joins that call, but some people watch the recording at 2x — in the car, click the recent recording in the Skool community, fly through the AI updates." ### [02:30:45] What's next in the Academy — and Oscar's agentic SEO result "The last section I want to go over is what is coming up next in the Academy," Austin said. "We brought on Oscar. He has done some great work with Hermes, so he's now our community's go-to expert on Hermes — any question you guys have that contains the word Hermes, he's gonna reply to it." Three mini-courses are coming in the next six weeks, built to stack: 1. **Setting up your harness** — "how to set up your harness, how to have orchestration — basically how to use an AI agent harness, Hermes being our recommendation right now." 2. **Agentic SEO** — built on Oscar's own result: growing his site "from 800 to 15,000 in a 30-day period." The room's SEO expert made him qualify it, in one of the session's better exchanges — "Impressions," Oscar clarified. Austin: "I think that pressure [test] is very clear, especially with an SEO expert in the room... it sounds way better when you say I actually have 15,000 people on my website." Justin: "Still very valid, is impressions." 3. **Outbound prospecting** — "as a category, we're gonna be focusing a lot on revenue." The prerequisites are deliberate: the Claude Crash Course and the Company Brain training, with the dashboard course next — "Those three things are just foundational to be able to start using agents. Get your team caught up: you know how to use connectors, you know what an MCP is, you're not afraid of the terminal. That's just foundation... and then it becomes a shared language, a shared understanding of a baseline, and now I can start building on top of that with AI agents." Austin's analogy for why the company brain comes first: "I think about AI agents like I'm onboarding employees — and if I have no onboarding slides to teach them about my business, I'm just kind of saying, hey, go for it." ### [02:32:20] Oscar's outbound agent for the rebate program — Oscar Hidalgo Austin asked Oscar to explain the prospecting agent behind the win he'd mentioned earlier. Oscar walked the whole pipeline: "We reach out to apartment complex owners and managers for taking advantage of the Austin Energy rebate program, for our HVAC company. It's a free product for them, pretty much — the city pays us to improve the energy conservation for all these apartment complexes. It involves finding the right contact to reach out to — like, who even owns these apartments? I was able to find some data in some specific apartment list providers, so I bought some data on the apartments that I want to target, filtered by the qualifications. That gave me at least an idea of: this is the company that owns it, and this is a good contact — either an operations manager, or a CEO, or the owner. "So then I have the agent on Hermes with a cron job every day. At midnight: grab from those lists of properties that I'm trying to tackle, find that contact, use the tool Apollo to find their contact information — email and phone number — then bring all that back, and also do a bit of research using the Perplexity MCP to make sure that we create a customized email for them. So now the email looks less like 'hey, have you heard about Austin Energy?' and more like: 'Hey, I know that Greystar manages 300 properties in the Austin area, you're one of the regional managers — I think this is something that will benefit your company. We as a company have done this project; here are two of our latest projects if you want to reach out to them.' Every single one of these emails is customized to whoever is reading it. "I also found a couple of cool email skills, just researching, that made the email even better — it changes the subject line, it keeps it short. A few things that we're gonna teach in the class — but pretty much automated, is reaching out to people." The outcome so far: several distinct properties enrolled into the rebate program off the agent's outreach — meaningful, six-figure-scale retrofit projects for the company, landed while everyone slept. When Austin polled who wanted the outbound-prospecting course, hands went up around the table. ### [02:35:28] The three levels, and when you need a harness instead of Claude Code — Austin Distel Austin laid out the maturity model behind the curriculum: "There are three levels of becoming an agentic company. Level one: I solve it for myself. Level two: my team is involved. Level three: agents are doing it autonomously. At level one, you need your own context set up into a repository — files, your computer, the environment. Level two is: can your team collaborate in Claude teams? Are you using things like skills, connectors, artifacts? Level three would be: I have a harness. Maybe I even have a separate computer. Maybe it has its own email, its own SOUL.md file, its own cron jobs, its own autonomous stuff. A lot of what we've talked about in the Academy so far is level-one and level-two work — now we're getting to level three, where the content we're creating is the AI agent orchestration content." (One member's testimony for why the onboarding steps exist: "I built a lot of OpenClaw without the onboarding steps... it was breaking every night.") Prashant asked the question that framework begs: "You're teaching a lot of this stuff with Claude Code — so what's the use case for Hermes versus Claude Code? How do you silo that in your brain?" Austin's answer was a checklist: "Does this system need to run when your computer's closed? Do you want to host it on a computer or on the cloud? How do you have persistent memory? What's its scope of work — is it for one business? I own a couple businesses, so do I want it to do one job, or multiple jobs? Are they sharing the same company-brain folder, and are two harnesses gonna compete to rewrite the soul, the memory loops, the instinct loops?" Then the concrete example from his own portfolio: "One thing I'm going to be automating is my receptionist at my locations of the spa. Entry-level front-desk staff are such a hard audience to keep training how to do sales — and sales grows the business. So the constraint is teaching entry-level employees how to do lead follow-up consistently, making sure that the Facebook ads I'm paying for leads every day are actually getting responses quickly, in a way that's on-brand and accurate. It never stops. That's a great example of an agent need — that's not something that Claude Code is going to do just yet... For my business The Tox, I would build a physical agent on my desk that works 24/7/365, without my computer on or off. That's the AI CMO that also does text-message marketing — I wouldn't build just 'the agent for text-message marketing.' And I would do it outside of Claude, on its own device, because leads come in 24/7." The room added the practical wrinkles. On authentication and cost: "You can't use Claude OAuth with Hermes, so you have to use [an API key] — and why that's important is cost. It's not meant to use your subscription to feed the harnesses." On consistency, one of the room's builders described the deterministic pattern: "If you give them both the same company brain, two LLMs think differently — they're gonna just mess each other up... I build a lot of automations not relying on the model. Prashant and I were brainstorming through a bunch of math problems for a real-estate analysis agent — trying to replace a whole real-estate analyst. If you rely on the agent to do it through the LLM, it's gonna give you a different output depending on which model you use. So with the best model I can get — Fable 5 — I built a bunch of scripts, just Python scripts that are running the math. And now the cheapest model available can run the same exact workflow and get the exact same outcome every single time. There's definitely ways to automate the task and not depend on the model intelligence." Austin tied it to where the space is going: "What's grown in the AI agent space is parallel agents — no longer are you just training one agent, you're training the fleet. Which further points to: build agents for the department, rather than for the individual job." ### [02:41:43] Case study in public: the AI CFO built on a layover — Austin Distel To make "department, not job" concrete, Austin told on himself: "On my layover this week — I'm kind of behind with my bookkeeper on all of her questions. She has so many questions all the time, about all the businesses. So I built, basically, an AI CFO that manages my portfolio of companies — not just Agentic Society, or The Tox, or my Airbnb. I want you to make sure that the money and the movement of all of that is accurate. So I built an AI CFO that has a large scope — and then it has a bookkeeping *job*, instead of just being a bookkeeper. It has all of the jobs that a bookkeeper would do — from reconciliation, to balance sheets, to profit-and-loss statements — having access to all of the QuickBooks. And it's able to draft replies in Gmail to my bookkeeper with all of her questions already answered. "As I approached this problem, I was like: well, I could spend these next three hours of my layover responding to all of her questions, finally, because of uninterrupted time — or I could build a system that has access to my information to answer all of her questions for me. So I said: I'm gonna answer the questions by building the system that will answer them for me. And it drafted the reply in the same amount of time it would have taken me to do it. "I say this eating my own dog food, in public: instead of me just building a bookkeeper, I asked — who would the bookkeeper report to? The CFO. And what access would they need? It has access to my S-Corp and all of the documents, the comptroller's office — things that even a normal entry-level bookkeeper wouldn't touch. So now I'm building agents thinking not just 'social media manager,' but content marketer leads up to CMO — the CMO should know how to do all of that. So I'm starting there." ### [02:45:03] Q&A: how do multiple agents share one brain without corrupting it? The question came back from the room after the CFO story: "I still don't understand how you get multiple harnesses to let them both edit the same intelligence, and that not be a problem." The concern got sharpened further: "Both of their ingestion systems and insights are gonna just overlap and conflict." The answer the room built, in two layers: **Layer one — agents append, they don't overwrite.** "It all works like GitHub. GitHub is just a folder of files. When you have multiple developers working on the same codebase, they're not going to wait for the next developer to do their part. Ideally, if you set it up correctly on the second brain, every agent is just adding a decision log, or adding a task log, or adding what they did — which is only making the other agent smarter. They're not really modifying the core files; they're adding to the brain in each of their individual ways. If I work on Claude actively, Claude at the end will log what we worked on in my second brain; if my Hermes does something automatically, then he should also log what he did. It becomes like a timeline of all the stages." **Layer two — the README is the checkpoint, and the changelog carries the why.** Austin: "Inside of that second brain, there's these README files, and those have directions — when the agent reads it, it'll say: document what you changed as a changelog below here. Ideally, whenever it enters a file, it reads that MD file first as a checkpoint, and then it follows the directions. It's also part of the self-healing, self-learning cycle: update the README file with what you changed, but also *why*. With every decision that I've made to grow and refactor Agentic Society's work, I also document in the skills the changelog of how we arrived at these decisions — so that we can learn from our mistakes, or learn from our improvements. This is history-buff stuff, but if we only know what happened and not why it happened, then we're gonna fall into the same traps again. We're gonna redo work — 'we changed this stuff years ago, why did we go back and forth on it?'" A member closed it with the org-design analogy: "It's just like employees. We have established what the SOPs are for the company. My CSR might come in and create a little process for herself to answer calls — she's adding to the brain, but she's not modifying how we do business, or the pricing. With agents, you kind of have to think the same way. It's creating a system, versus an open area for these guys to come in and mess things around... I don't know if there's a right or wrong answer, honestly — leveraging different models that cross each other's paths and check what you're doing gets you the best output — but the answer is: what does the system look like?" ### [02:49:39] Adjournment — Austin Distel "At this time, Will and I are gonna go downstairs to the lobby. We have officially 15 more minutes in this mastermind — y'all are welcome to kick up conversations. We're gonna start recording some testimonial videos; if you guys are down to record one, just come with me and Will downstairs, and then I'll come up to adjourn the meeting." Side conversations and one-on-one meetups carried the room out. --- No video replay exists for this session — the full attributed transcript above is available to attendees via the download link on the recap page.