# AI Mastermind & Mixer — Clean Transcript **Date:** 2026-07-03 · **Type:** AI Mastermind (free monthly, first Friday) **Location:** Industrious, Austin, TX **Duration captured:** 2:42:32 · Main program begins at [00:33:14] > **Timestamp note:** times below are from the RAW recording (which started at > 12:01 PM CT, during lunch). Re-map them to the published/edited video before > using any "watch at" links. > **Attribution note:** triangulated from three captures — the ceiling-mic Zoom > connection, a Granola transcript, and a tabletop Plaud recorder on the far side > of the room — plus the digitized post-it board. Speakers are reconstructed from > context across sources. `[H]` high > confidence · `[M]` verify speaker · `[V]` verify wording. Where we could not > determine the speaker, the segment is marked **Unattributed** rather than > guessed. ## Roster (in the room — ~40 attendees; Luma check-ins + spoken evidence) | Name | Business / role | Source | |---|---|---| | Austin Distel | Agentic Society (host) | host | | Oscar Hidalgo | CG Service Pros (HVAC) + Vindex Consulting | Luma check-in + spoken | | Sagun Sedai | Real Austin Report (austin-report.com) | Luma check-in + spoken | | Justin Day | Local SEO agency | spoken + prior events | | Christopher "CT" Schenk | OVAE (ovae.ai) — EBI skill | Luma check-in + spoken | | Geoff Woods | AI Leadership — author, *The AI-Driven Leader* | Luma check-in + spoken | | Christopher David | OpenAgents (openagents.com) | Luma check-in + spoken | | Chrissy McDannell | The Magnolia Firm | Luma check-in + spoken | | Lynn Hayden | Neurow (neurow.io) | Luma check-in + spoken | | Nikki Pepper | LyfeOS (lyfeos.io) | Luma check-in + spoken | | David Gonzalez | Entrepreneur | spoken + prior events | | Bryce Sandry | Connectricians | Luma check-in + spoken | | Wilson Smith | AI builder — multi-model debate loops | spoken + confirmed by Austin | | Rick Middlemass | Financial/sales reporting | Luma check-in + spoken | | Adrian | Government architecture work | spoken (verify last name) | | Chris | Web/SEO services | spoken (verify) | | Erin Kenny | The Celiac Space — functional & integrative health (RDN) | spoken + confirmed by Austin | | Rohan Karunakaran | Frontier Studio | Luma check-in + spoken | | AZ Moyer | AZ Does Taxes (CPA) | mentioned in stories | | Isaiah Zimmerman | AI builder — company-brain software | spoken + prior events | | Prashant Vanka | 8homes.ai | Luma check-in + spoken | | Lauren Goldstein | Golden Key Partnership | Luma check-in + post-it board | | Ed Vincent | FestivalPass | Luma check-in | | Brigitte Donovan | Creative Collective ATX | Luma check-in | | Toria Bonanni | Creative Collective ATX | Luma check-in | | Sarah Ahmad | ScalePartner AI | Luma check-in | | Blue Delgado | Funk It Interiors | Luma check-in | | Riley Spiller | — | Luma check-in | | Jay Douglas | KeyFive | Luma check-in | | Darby Rollins | Agentic Society (team) | Luma check-in + spoken | | Cristian Aguilar | Agentic Society (team) | Luma check-in + spoken | | Roman Bediner | Agentic Society (team) | joined via Zoom (Granola participant) + spoken | | Connor Cullip | Property-listing microsites | post-it board | | Christin Raschke | Business Foundation Blueprint | post-it board | | Peter Hewlett | — | spoken (tabletop mic; verify spelling) | | Nicole, Adam, Frank | first names heard in the room | spoken (verify) | --- ## Transcript (attributed · timestamped · signal only) ### [00:33:14] Welcome & how this event works — Austin Distel `[H]` All right — if you've been to this kind of mastermind before, raise your hand. Okay, now new hands. Love to see it. Welcome — happy July 3rd. This event tends to be laptops-closed. We're going to talk a lot about theory and a lot about how we use AI in our business, and y'all are going to participate a lot. This is not my show — but there are a lot of new faces, so something I added this time is a little about me and what Agentic Society is, because most of the time this room is full of members and people who know me. My goal for these events — the first Friday of every month — is to inspire the leaders with real-life stories of business owners in Austin, Texas who are getting outcomes with AI agents. In a little while you're going to write down outcomes you've gotten using these tools, and then we're going to vote. Not everybody has to share — you're welcome to participate by listening — but if you want to share, you'll have that opportunity today. There are three goals for this event, and really this should be your goal for any event you go to: **one meaningful connection, one brilliant idea, and one decisive action.** That framework will make sure you get the most out of the next 90 minutes. To make sure you capture that value, I've created these notebooks — write your name on the front so if it gets lost we know who to return it to. Claude helped me create the first draft and Canva finished it up. ### [00:36:38] Everything from today ships to you tomorrow — Austin Distel `[H]` When you hear an idea that resonates, write it down — because after this event you're going to get the full transcription, the recording, and all of our notes. What's personalized for you is what resonated, and the timestamp in your notebook is how you look it up in the transcription later. By tomorrow you'll get access to a website personalized for this event. It includes the recording — smile, you are on camera — and there are microphones embedded in the whole ceiling so everyone's voice can be included. What we're doing is enriching the room's ideas. This is a mastermind — think collective consciousness. We're creating, agentically, a web page full of all the ideas we generate today. ### [00:37:30] About Austin — YC to Jasper to Agentic Society — Austin Distel `[H]` A little about me, since some of you don't know who I am. I live here in Austin — moved here in 2018. Seven of us moved together after graduating Y Combinator in California, landed on 6th Street, and built a startup called Jasper. We started working with OpenAI in 2020 — GPT-2, two years before ChatGPT — and taught it how to do great marketing. As CMO I helped build it to a hundred thousand paying users and about $60 million in annual revenue before departing at our Series A. Today I've invested in and acquired some businesses, and this is now my main thing: collecting together this community of AI builders and business owners and helping them use AI agents. What I do is dummy-proof everything, because I'm non-technical — I don't know how to write code. That's actually how I feel Jasper grew: I took what developers were building and kept it in common language so marketers knew how to use it. So if you're non-technical, I hope you resonate with me. ### [00:38:43] The three levels of AI adoption — becoming the Agentic CEO — Austin Distel `[H]` Having a lot of conversations with people building with AI agents, I've discovered there are three levels to your skill set — the phases of AI adoption on the way to becoming the **Agentic CEO**. Even if you're not the CEO of your company, this is still an identity you can step into, because when you hire a bunch of AI employees you essentially become a CEO: you're learning how to onboard them, train them, get them access to things. - **Level one**, you're a solo operator using AI directly. - **Level two**, you've created a structure and organization for AI to work in. Anyone who's built with teams knows this is harder — getting your agents access to everything, and letting others work with what you built. - **Level three**, you have a team of people *and* AI agents working together to grow the company — an AI-first company. In your notebook, take an honest account: where are you right now? I believe that by the end of summer you can get to the next level if you commit to it. I wrote descriptions in the notebook to help you graduate — little bits and pieces to look up on YouTube or have conversations about. ### [00:40:25] What changed: AI got hands, a mouth, and a network — Austin Distel `[H]` What has changed in the last year — even since Jasper was using GPT-2 two years before ChatGPT — is that **AI has been given hands, a mouth, and a network.** The ability to do things autonomously, to access your tools on your behalf, and to work with the people on your team. It's no longer siloed, no longer silent, no longer held back from the rest of your systems. When you switch from chat — any chat, Claude or otherwise — to using it in code, talking to it, giving it access to your machine, it can autonomously do things. That's what we call the employee, the agent, however you want to call it. It works 24/7, 365. That's what we're all here to learn: how to use, hire, and build these. ### [00:41:24] Invest in AI like people, not tools — Austin Distel `[H]` AI is no longer just a tool with a head count. We used to engage with software tools. But if we shift the perspective to investing in *people* — investing in the intelligence, the skill sets, the access, the permissions — then trust has to be earned over time, just like a contractor who becomes part-time and then full-time. That's the path of autonomy. Think about the first week of using a new AI employee like the first week of hiring a human one. What does onboarding look like in your company? What knowledge do you give them access to? Which tools do they have permission to use? Who else works with them — the stakeholders on the team? What are their outcomes, their belief systems? When you hire that person, what's their mindset, soul, and identity? When you talk to Claude and say "I'm hiring you as my AI CFO" — do you want them to be a creative genius? No. Program them that way. Do they need access to Canva? No — that's not in their permission stack. They own Excel and QuickBooks. Call your agents by the role they hold in the business — and then give the human employee who manages that AI employee a director title. Give them the promotion, because their productivity is about to be insane. As Agentic CEOs we're leveling up our team, not just ourselves. And don't outsource this. If somebody else builds it and leaves — hopefully they loved you better than they found you — these should be owned systems, by you and your company. This is your proprietary product. Your company is becoming an entity in the truest sense: a year from now, you're in your boardroom, and the company itself is talking — because it has all knowledge across all AI agents and human employees building the company intelligence together. Who runs that? The Agentic CEO. Who builds the infrastructure? Who hires, trains, and upskills the team? Same answer. That's our role. ### [00:44:19] "One-legged soldiers" and Debbie from admin — Austin Distel `[H]` I was talking to a well-known SaaS founder — 17 years in — and he said: "I feel like my team are basically one-legged soldiers if they're not using AI agents." It's not that people who don't use AI aren't valuable — it's that we need them to evolve to be fully productive. Even the most resistant person — I call her Debbie from admin — can learn these tools. One of the most exciting evolutions at Jasper: some of our customer success grads fresh out of UT were doing cooler shit with AI than our senior developers, because the developers thought they were better than the AI at the time — this was 2022. When the CS team started using it and the developers caught on, it literally broke the frame and leveled up the whole company. So my recommendation for every leader in the room: take the person who's most resistant and least adapted to AI, and get them a quick win **publicly** — where everyone is praised and recognized for helping Debbie get up to speed. That's what shows everybody else what's possible. ### [00:46:11] The milestones: foundations → brand intelligence → company brain — Austin Distel `[H]` There are a few milestones that help you level up. **Level one — AI foundations.** If some of your people use AI and some don't, you need to get on the same page: hold a foundational meeting, make it part of team culture, and start talking the same language — because the people in this room and the people outside it don't talk the same language. Have the team go through Anthropic's documentation, our training, or the great YouTubers out there. Ideally run something like a 12-week get-up-to-speed, plus a Tuesday lunch-and-learn where everyone shares what they're building each week. That gives your team speed. **Level two — systematic brand intelligence.** Now everybody has gone from Claude chat to co-work, using projects, connectors, skills. The next phase is building a knowledge base, access to tools, API connections — with governance, shared and understood, so everyone can plug in. What some people get wrong at this stage: it's not on one AI builder to do all the work. It's on you to get your team inspired to manage their own scope. The CMO owns the marketing skills; the head of customer success owns theirs; each of them updates their own skills and connectors and owns that part of the organization. All of it rolls up into the **company brain**. If you do this well, you get the **AI slingshot effect**: you slow down for a second to build the company brain, and then everything else is so much easier and faster. The hardest part of getting AI employees going is HR-style onboarding week. Imagine a company that just raised a billion-dollar seed round and has to hire 100,000 employees — extreme on purpose — how does HR onboard that many? The company brain is your HR. It's how every new AI agent learns your company infrastructure, your org chart, who to talk to. Build it once, and hiring AI employees, clearing the constraints in your business, and scaling profits all get quicker. ### [00:49:22] What Agentic Society provides — Austin Distel `[H]` A few ways we've been helping our community: - **Training programs.** The Claude Crash Course is live — thank you, Darby, for recording such a great one. Over 7 days it takes someone from Claude chat to Claude Code, and gets the whole team using the same projects and skills — a cohesive AI environment. - **Pre-built AI agents** — skills and connectors you can start using immediately. - **The Agentic OS.** Think of talking to a recruiter: "the constraints of my business are that I need a high-ticket salesperson and a setter — I'm trying to go from Instagram DMs to appointments to closed deals." You talk to Harper, the HR recruiter, and she looks across all the AI agents in the community and finds the right one. We currently have **over 100 AI skills and nine AI employees** in the Agentic OS. Think of it this way: a CMO is responsible for all channels — SEO, paid traffic, social — and each is an individual skill. Don't over-fragment: you don't need a "LinkedIn skill," that rolls up into the social media manager. Name skills by **who** uses them, not what they touch: it's not "the QuickBooks skill," it's the **P&L Manager** skill. Who, not what. - **Weekly rhythm.** Tuesday calls from 2:00–3:30, and a Friday AI board meeting in person upstairs from 9:00–12:00. Three hours of sharing, showing, telling, and solving the real constraints of our businesses quickly, with a like-minded community that gets the hardship of being a CEO. If you're interested in being in that room, shoot me a text — that's my real phone number on the materials, and it rings this phone, so please don't call it right now. It's for CEOs and operators using AI every day — a 12-week to one-year engagement, meeting upstairs in person twice a month, with all the resources, courses, and AI agents. That's what Agentic Society is. We launched the founding member round only 45 days ago and we're a little over halfway through the 50 founding member spots. And I host this free event once a month — the next one is **August 7th**; I'll show the QR code at the end. ### [00:53:19] Member spotlight stories — Austin Distel `[H]` Stories from our community — people you might recognize in the room right now. **Justin Day** is a local SEO expert here in Austin — he helps businesses get to number one on Google, especially with the AI changes happening fast. Over the last year he's reduced his team from 30 heads to 9 while productivity went **10x**, running 12 agents with 6–8 skills per agent. He's got a high-volume dashboard pulling his daily goals and KPIs into one place, he built an AI SEO scanning tool his clients use to understand how their sites will rank, and he recently closed a six-figure client where much of the SEO work is done by his agents **in parallel, in one day** — compressing time and getting the client their outcome faster than ever. **Erin Kenny** runs The Celiac Space, a functional and integrative health clinic. She and her COO had been asking how to increase lifetime value by understanding retention — who's up for renewal and how to get ahead of it. She joined a call showing us her beautiful spreadsheet — she loves that spreadsheet — but wanted to visualize renewals and turn the data into proactive insights. By the end of the call she was in Claude Code and had built the dashboard straight from her database. Now that's what she presents her KPIs from. **Rohan** is our go-to LinkedIn growth advisor. In our boardroom we run a constraint-discovery exercise: what's the one domino that, if you knock it over, solves the rest — and how do we do it in two weeks with AI? His number one problem: taking long-form videos and clipping them for LinkedIn — finding the three most impactful moments takes forever. Now he's advising an AI editing tool called BetterClip where, from within Claude, you can have Claude edit long-form video and find the three most impactful clips for social. **AZ does taxes** — azdoestaxes.com — I love this guy. He's a CPA, and he told us: "I'm a financial guy, I don't know how to make things pretty." Over the course of one call we taught him Claude Design, and now every new client pitch gets a personalized, beautiful deck in a unified brand voice — and he just closed **$10K in the last two weeks**. All his new clients are impressed; the quality is on a whole new level. **Isaiah** is one of our AI builders — very intelligent at building autonomous agents, and he builds software specifically around building a **company brain**. If you're deep in the weeds on that, Isaiah is the go-to guy. And as a software developer he said something important: "I've been really lifted up by this community, because I've been in a silo talking to robots from 3:30 AM to 8 PM this season." That's what real life is about, y'all — we're physically here, in person, together. **Prashant** — what's up, bro — is a real estate investor and builder. He has an AI agent that texts landowners to see if they're open to deals, qualifying them and growing his pipeline. **Roman** is on our team here at the Society — he worked on Disney+ for eight years building infrastructure. For our business he's built **Slack agents** that listen to all our calls, watch all communication in Slack, and create and maintain the project boards in real time, constantly. We don't need a project manager, a chief of staff, or an orchestrator — it's literally updating the kanban boards in real time: work in progress, who's done what, who owes what. I wake up to a re-ranked feed of my priorities every morning. The common thread: these are mostly **non-technical business owners** — just like me — using these tools to scale a business. So now it's your turn. ### [01:00:29] The exercise — "My agent did X and got outcome" — Austin Distel `[H]` I'm passing around post-it notes. You're not required to participate if you don't have a personal AI agent story. Write it like this: "**My agent** did something specific **and got** an outcome" — an outcome, not just an activity. Shout-out to Oscar: at our last mastermind, his was the one that won. We'll do an awards ceremony later for the most helpful AI builder in the room — the award matters less than being recognized for being **helpful**. That's what the mastermind is: we come to share and to learn, so be open-handed and giving. All the brilliant ideas will be posted on the AI event page afterward, and everything is in the transcription — if you need anything redacted, shoot me a text and we'll take care of it. *(Five-minute brainstorm break while the room writes.)* ### [01:11:46] Twenty-four builds from one room — the reading & voting — Austin Distel `[H]` Wow — 24 outcomes that AI agents have brought to this collective. About 40 people in the room, about 24 brilliant AI agent builds — that's the proof that it works. Here's how this goes: I read each one out; if you want to hear more about it, raise your hand; Cristian counts; the top three most-voted share in depth. Even when one doesn't win the vote, write the builder's name down if you want to follow up — all of these, with names, will be on the event page tomorrow. **[01:12:59] Oscar `[H]`** — built an AI agent using Hermes that books sales calls, and it landed **two jobs worth $200,000 each**. **[01:13:32] Jay Douglas `[H]`** — a **CFO agent** that builds on-demand profit-and-loss statements and 13-week cash-flow reports with what-if scenarios. **[01:14:51] Geoff Woods `[H]`** — "My thought partner transformed how company offsites run — collapsing six months of alignment conversations into two hours." He does this with Fortune 500 companies on six- and seven-figure scopes. How do you align an entire workforce on what matters to solve the company's biggest problems? Like a war room. **[01:15:57] Ed Vincent `[H]`** — an AI that **checks hotel prices daily** after a booking; if the same room and itinerary shows up cheaper, it books the new reservation and cancels the more expensive one. A price-guarantee agent that saves money automatically. **[01:16:40] Rohan Karunakaran `[H]`** — "My content-writing agent identified the highest-value moments from the last mastermind and wrote a post that drove me **10+ inbound DMs**." ("It wasn't me," Austin laughed — it was Rohan, the room's LinkedIn growth advisor.) **[01:17:23] Lauren Goldstein `[H]`** — "My AI agent took hundreds of my notes-to-self and reminders, triaged and organized them, and added them into **ClickUp** to keep me on track" — everything from texts to notes, organized over 30 days. **[01:18:07] Prashant `[H]`** — an AI agent that underwrites and auto-records into the CRM — "mission control for my life." A kanban of tasks where the agent comes up with next steps on all of them, reprioritizes, and shows up with a plan or clarifying questions. **[01:18:52] Chris `[M]`** — an AI agent that reviews client websites, tells the client whether their site is **AI-agent-friendly**, and offers a solution — "we've seen **10x traffic** since." **[01:19:32] Blue Delgado `[H]`** — shared a story from a marketing coordinator candidate he'd just interviewed: the candidate built an AI agent that sourced hundreds of job applications and applied to all of them — six interviews yesterday, six more next week. **[01:20:34] Rick `[M]`** — "I used to create financial and sales merchant reports and sheets by hand. Now my agent API-links to all the tools and sends the weekly reports and emails for me." **[01:21:04] Erin Kenny `[H]`** — uses **Granola and Claude to audit sales and retention calls** and improve delivery and KPIs. How many of you do sales and retention calls? That's what this agent improves on the back end. **[01:21:38] Wilson Smith `[H]`** — "Leverage the strengths of different models by running a **live multi-round debate between Claude, Codex, and Gemini**" — in loops. "I say: I want to build X. They debate every round of planning to pick the best approach. You're not allowed to remove features — you only add — until you've built the whole thing." *(16 hands went up — and Wilson took a bow: "Wheels up. The token burner. Hi, my name is Wilson and I burn tokens.")* **[01:22:24] Connor Cullip `[H]`** — "My AI agent takes property listing links and creates dynamic custom websites displaying everything in one place — prices, maps — and creates a feedback loop to the Ops team, saving them **8 hours a day**." **[01:23:14] Sagun `[H]`** — "My AI agent generates content ideas daily so I can post 5–6 times a day, finds the most viral-worthy info, and creates a newsletter from it. Because of this I have millions of views." The AI rates which topics will generate the most engagement — that's how he's grown his page. *(About 80% of the room voted to hear this one.)* **[01:24:19] Darby Rollins `[H]`** — "Build an **AI HR director** that headhunts, hires, onboards, and trains your agentic workforce — one AI employee at a time." **[01:24:50] Justin Day `[H]`** — "I built **Zoom meetings with my AI agents** for real-time collaboration and execution with the AI team. Instead of a chat interface, avatars come to Zoom meetings — they give me updates on the company, I tell them what to go do. It's getting out of the terminal." **[01:25:59] Isaiah `[H]`** — a "**closer OS**": a 16-phase post-sales-call pipeline that captures, extracts, researches, and builds a website proposal based on the sales call — Stripe handling and contact information embedded — then loads the signal into a loop to watch how the pipeline and offer are being viewed. He's been on a lot of sales calls lately; it doubles as a post-call sales coach. *(Austin, counting hands: "What was the count? Nineteen.")* **[01:27:04] Adrian `[H]`** — "My AI agent, Promptmaster, helped me recreate and find an architecture for a coding repo — turning **three years of work into two hours**." It was for a government project; they'd been trying to find the architecture for three years, and his agent solved it in two hours. *(Details limited for confidentiality.)* **[01:28:15] Bryce `[H]`** — created a "**Connected Mind**" from 305,000 texts, 15 years of emails, and all his saved Instagram posts. It generates content, business ideas, and historical profiles on everyone in his life and network: "Every person I've ever talked to has a profile — all the topics we've discussed by year, how many texts each way, their interests, gift ideas, the last thing going on with them. I'm bad at following up with friends and family, so it tells me: check in with this person, you haven't talked in 68 days." **[01:29:32] Nikki Pepper `[H]`** — an **AI-powered search function for an NBA coach** who coaches other coaches on his philosophy: search across everything said in his video library — plays, drills, diagrams, tags, resources — far more accessible than a standard course library. Also building video shot analysis where players upload their shot for AI feedback. The translation for any business: if you have a course, let people search the transcripts for exactly what they need instead of scrubbing videos. **[01:31:30] Christin Raschke `[H]`** — a **Business Foundation Blueprint** that builds the foundation of a client's AI brain, reducing client onboarding time for business operating systems by **85%**. **[01:32:08] Christopher David `[H]`** — "My **promise-keeper agent** organizes all the concepts, feature ideas, and product promises from my 200-episode build-in-public video series into one web page — which has inspired multiple developers in our community to write code and accelerate our whole roadmap at no extra cost." **[01:32:48] Chrissy McDannell `[H]`** — "My AI agent grocery-shops for me on Amazon Fresh. The best part: she's triggered when I tap an **NFC tag on the side of my fridge**" — or by voice through WhisperFlow. **[01:33:54] Lynn Hayden `[H]`** — "My **digital twin agent** set up my digital business departments — marketing, ops, engineering, R&D — plus an AI HR department, which then built all the department lead agents." It lives in Obsidian in markdown: the office index, all the protocols and SOPs, and a department-head agent per department. **[01:34:51] Unattributed `[V]`** — an Instagram **DM-setting agent** that improves its scripts based on which ones are closing fastest. A human setter does the sending; the agent learns from the feedback loop. **$12,000 in deals closed in three weeks** from fitness coaches in the DMs. **[01:35:55] David Gonzalez `[H]`** — "My finance agent runs the daily process that took me from six figures of personal debt to accredited investor. By hand it was a 10-minute daily micro-reconciliation across my 12 bank accounts — an hour if I missed a day, because it compounds. I built an agent that uses **Plaid** to download all my transactions, and now I just go *ching* — it shows each transaction blurred, I click to clear it, reconcile, done. It's got streaks. Seeing my bank balance every day motivates my money behavior that day. It's under two minutes now — I could get it to fifteen seconds with a new Mac mini." *(From the room: "Is that a Tiller sheet?" David: "No — I have Plaid connections, but it's not a Tiller sheet. It's my own build.")* **[01:37:58] CT `[H]`** — "My agent did **not run out of tokens**, by using Fable 5 — because it wrote like a boss that's super efficient. I spent 5 million tokens of plan usage in a day before; working only with Fable it was 200,000 for the same work. It's figured out efficiency." If you're hitting max usage all the time, that matters. ### [01:39:04] The vote — three deep dives — Austin Distel `[H]` The winners: **Oscar**, **Sagun**, and **Justin Day** — and if we have extra time, CT will share. ### [01:39:38] Deep dive 1 — Oscar Hidalgo: the outreach agent that landed two $200K projects `[H]` **Austin:** Oscar, you built an agent organization that books sales calls and landed two jobs worth $200,000 each. Tell us how. **Oscar:** I have two companies — I'm the CEO of an HVAC company called CG Service Pros, and I have a consulting company where I implement these AI workflows in other small businesses. In the HVAC company, one of our departments does a project with the City of Austin: Austin Energy has a program where multifamily properties can get up to $300,000 of value to reduce the city's energy consumption — HVAC tune-ups, Wi-Fi thermostats, LED retrofits. Our job is to find property managers and owners of apartment complexes that can benefit, get them assigned to the program, and do the work. These are huge projects — 400–500 unit properties, every unit touched. For years we did manual outreach. So with Hermes I set up an agent — though it really starts with Claude. I signed up for a data platform (~$400) and downloaded lists of all the apartments that would qualify. I added my own data on every project we've completed. And to Austin's point — each agent is like training an employee, except better, because you can give them **skills other people have developed**. You don't have to know everything or train every employee yourself. I literally Googled for a cold-email skill that actually gets responses, found one, and gave it to the agent. Interesting details in there — like writing the subject line in all-lowercase, just two words, so it reads like an internal email. Open rates jumped. So the agent has the skill, my research on the properties, and our completed-projects history — and I just let it run. It's a cron job every night around 12:30 AM: it goes through the list, sees which ones we've already tried, researches the property — whether they own other properties — and prepares a customized email. It uses the **Apollo MCP** to enrich contacts so I'm reaching the right email address. It sends, copies me, tries to book a meeting, and follows up every other day. I've since landed **two of these projects — $200–300K each**, multi-month jobs. Pretty much all outreach for that rebate department is now on autopilot, every weekday. **Q: How are you sending these?** I created a new user in our Microsoft 365 — this is a full onboarded employee. Her name is **Ali** — that sounded natural — ali@ our domain, and she sends directly from it. I keep volume very low, 10–20 a day, so it's not spammy. And these are apartment managers — land one and they often have a portfolio of 15 properties. **Q: Any buying triggers you've noticed?** It's almost cheating — the customer doesn't pay a dollar; the city pays. Many are income-qualified properties without funds to improve, so as soon as they hear about the program it's "let's see if it qualifies." The offer is incredible — the city wins on energy, the customer gets improvements, and we get the projects. **Q (Nicole — possibly Nikki Pepper): For someone who hasn't gone down the rabbit hole — what's the biggest benefit versus what we're already using?** I was using OpenClaw before, and my problem was it breaks a lot — and I'm non-technical, so I'd lose a whole day figuring out what went wrong. Hermes has done amazing work making it stable: after you train it, it just runs. One thing — I'm on a MacBook Pro, so when it's off, the agent's off. So I host Hermes on a **VPS** — a virtual machine, about $8/month — so it's live 24/7 and doesn't rely on my computer. If I had a Mac mini I'd consider running it with Claude directly — probably cheaper — but if your computer might go down, use a VM. **Austin:** For reference — at the last mastermind Oscar shared how he grew his HVAC company's SEO from 800 to 15,000 visitors with an AI content play. Look up the last mastermind on the website for that deep dive. Why'd you switch from OpenClaw to Hermes? **Oscar:** Stability — plus the new Hermes desktop app is amazing. You can build different agent profiles — different names, different souls, so they speak differently — without going crazy in the terminal. **Q: What's your advice for leveling up from level 2 to 3?** Use Claude as your active working partner for the transition itself. Go to the Hermes website, give it to Claude, and say "I'm trying to set this up — walk me through it." Claude can even create the files and the soul — everything you need — and you transfer it to your virtual machine. When you're non-technical: just ask Claude everything. What does this mean? What's an MD file? And consume as much content as you can — stay learning. **Q: What's the value versus just using Claude?** 24/7 availability, and multiple profiles. Claude always feels like the one Claude to me. In Hermes I have **Ali**, who's now a customer service specialist, and **Kai**, connected to QuickBooks and Airtable doing forecasting. Way different personalities, different tools — like different employees. That's what these agent harnesses are for. **Q: Security concerns with Hermes?** Everything it does goes through official MCPs and APIs, so it stays pretty safe. There's always risk — that's why Ali has her own email, not mine; I'm the CEO after all. Think of it like hiring an intern: you wouldn't give an intern your personal email. Their own email, their own SOPs, a limited set of tools. Yes — Ali is a full Microsoft 365 user; I can log into her account and see everything. She doesn't need passwords to other websites — she needs Apollo for contact data and web search, which Hermes has by default. To translate this beyond HVAC: any B2B business with a prospect list matching your ICP can use Apollo to find real contact information, and your agent can run that motion around the clock. **Q: Average cost?** Apollo ~$100/month (I use it myself too). The VPS on Hostinger ~$8/month. I use OpenRouter as the brain — it switches between models — on a $20/month plan. Initial data was ~$500. So about **$500 setup, and under ~$30/month ongoing** for the agent itself. ### [01:46:43] Room poll — what level is the room at? `[H]` **Austin:** As we think about the three levels: who's at level one — using chat most of the time, haven't stepped into skills and projects? *(Two hands.)* Level two — projects, skills, connectors, maybe Claude Code in a chat interface, but nothing running autonomously in the background? *(About 15.)* Level three — OpenClaw or Hermes running constantly in the background, agents with names and likenesses, Zoom, email, APIs, MCPs? *(About half the room.)* ### [01:52:26] Deep dive 2 — Sagun Sedai: the Real Austin Report content engine `[H]` **Austin:** Sagun — the Austin Report. You have an AI agent building daily posts about Austin news, and a newsletter. How did you build it? **Sagun:** When I first started I did everything myself — and when you cover everything happening in Austin, the news, trending topics — I got overwhelmed. I wanted to be consistent but couldn't research and find all these items every single day. So I went to Claude, like a lot of people do, and fed it my data: all the things I talk about — trending topics, tech, data centers. It scrapes different news sites — Austin Business Journal, KXAN, anything around Austin — finds those keywords, especially the high-figure-number stories ("Nvidia is moving $200 million here"), and a cron job at the end **ranks 3–4 topics with a trending score** — no guarantee, but based on previous traction, what's most likely to go viral. Then it builds me a **script** for each video, programmed to 90 seconds or less, because those perform best in my data. I go say those topics on camera every single day — that consistency is how I got about **3 million views this month**. At the end of the week, it scrapes my own Instagram for the highest-engagement posts — I put CTAs linking to my newsletter — builds a newsletter ranked by engagement with the top five topics, and **auto-sends it Sunday at 9 AM. I've done that 66 weeks straight.** That's how I've amassed the following — one simple tool, fed with all that data. **Q: How does the handoff between jobs work?** The next step after the cron job is *me* — I'm the one who says the content on camera. Then I use **Metricool**: I post once and it distributes to Instagram, TikTok, Facebook, everywhere, so I'm not posting the same thing on every platform manually — I post four to five times a day, so that would be overwhelming. The weekly cron builds the newsletter and sends it through **beehiiv**. **Q: Where do the ideas live?** Right now in Claude and my notes — honestly still a work in progress. *(Austin: even for yourself — route all the outputs into a single database like Notion, so a team member can access the same tool and take that off your plate.)* Yeah, I think that's a good idea. **Q: Editing? Cover photos, captions, music?** It's just me talking — green screen with the website behind me. I use **Edits** — Instagram's own editing app — that's given me the most success, since Instagram pushes its own creative tools. Cover photos and captions are me or my VA — that's the current bottleneck, and probably my next AI layer: have it find the topics *and* generate the images so I don't spend that extra 10–15 minutes. **Q: How do you monetize?** Not the content directly — I work with businesses. I help local businesses grow, promote them, advise on ads. The newsletter has sponsors because of the traffic. Businesses see me being involved in the community and providing value first — that's why they trust me. Value first; a lot of what I do I give back. ### [02:03:00] Austin's thesis — do the most human thing — Austin Distel `[H]` A thesis I have about our role as business owners in the world of AI: **whatever the most human job is — that's yours.** In Sagun's journalism business, what's the most human part? He shows up in person; he builds relationships; the community notices him. He's on camera doing the news. What's the non-human stuff that eats time? Editing, researching topics, organizing, prioritizing. Those are all great AI tasks. We don't defer everything to AI — the point is to do *more* of the things that are most human, the things that let you shine. He's clearly shining — 40,000 followers and a big subscriber base. Maximize your time on those things: that's how we scale. **Sagun:** My other bottleneck is DMs. I get a lot, and I want to reply myself — I don't trust AI with the wrong reply to someone important. **Austin:** Same — that's why I put my personal phone number on all of this material. I'm not going to have an AI setter talk to you all; that isn't human to me. But I *will* have AI organize the CRM. If I could spend the most time on human things, it's the interpersonal relationships, not the DMs — unless it's clearly labeled as an AI setter, like the one we'll have on the *company* profile. If you DM me personally, you're only ever going to hit me. **Suggestion from the room `[V]`:** disqualify the unserious: "I have a hard time keeping up with DMs — send me an email here." That one extra step disqualifies 90% of the people who were never going to move anyway. ### [02:06:31] Deep dive 3 — Justin Day: real-time meetings with your AI team `[H]` **Austin:** Justin, you built the ability to have conversations with your AI agents via Zoom. **Justin:** Last time I talked about my hive mind — my second brain. We've got all this business information in one place, and I can talk to it through the terminal or whatever interface — but that's still one-way, text or dictation. I wanted **two-way conversations in real time** — the most futuristic, human thing I could do. Leveraging Fable, I duplicated Zoom internally — built my own internal video-meeting system — so I can have actual face-to-face chats with my agents. Same philosophy as the society teaches: role-based agents, each with their skills. Instead of outsourcing one task at a time, **I hold a meeting with all my AI employees in one room, in real time** — a brief coming to me as a conversation instead of a Slack message. They have personality, a face, everything. There's a bit of latency right now — roughly 3–10 seconds — and that's what I'm working on. I built the avatars with HeyGen — anywhere from $19 to a couple hundred bucks — though you don't need a talking head; a static picture works. I went extra. We've got hundreds of agents, but the meeting is with the role-based leads. **Q: Is it generating video clips per response?** No — face and mouth are **mapped in real time**; whatever text renders on the back end, they speak. And because it runs **local**, it doesn't have to round-trip the internet, so it's much faster. **Q: Why your own internal Zoom?** If I used real Zoom I'd have to create separate email accounts for every agent and invite them. This way I just call a meeting any time — and it's my own asset. (I still use Zoom for my human employees.) Now's the time — we're able to duplicate almost any platform. **Q: How long did it take?** With CT's EBI skills and Fable, maybe 6–7 hours to one-shot it — after about 3 hours of planning. That's the biggest thing: nail the plan and the goal, run "even better if" loops until it's absolutely clear, so when it goes into action you get it in one shot. **Q: Does your team use it?** No — for now it's just me. I've given these agents full rewrite capabilities to execute strategy and do real business. I don't want my employees in there yet. **Q: Where does the second brain live?** All local on my computer — I push some of it up, but every data source across the business is in one place, from bank data to social. Because it's all together I can assign **roles with specific data and actions**. It took a lot of work, but that's what lets you take it to the next level. And yes — the agents get **role-based access** to the second brain: they don't need everything, I want each one to be a genius in its own lane, so there are guardrails on what each role can see. **Q: Cost?** Still figuring it out — I don't care about cost right now; I'm maxing everything on Fable while it's out just to see what I can do. I might switch models depending on what makes sense for the harness later. **Q: What model are you using?** Claude for everything right now — being creative with Fable, maxing it out. **Austin `[H]`:** You can literally create any simulation now. I think we just broke the internet. ### [02:14:30] CT's EBI skill — ebiskill.com `[H]` **Austin:** For reference — at the last mastermind CT shared his agent exercise: "what would this look like if it were **even better**?" It keeps pushing — no, better; no, better — in a loop, rethinking on top of itself. He turned it into a skill. CT, what have you done since? **CT:** **ebiskill.com** — Even Better If. It installs from a small prompt, and when you run it, it finds your most recent piece of work and runs the process — I always run it three passes. It works on any model, on anything. **Austin:** All the resources and links will be included in tomorrow's website, so you'll get access to the EBI skill there. ### [02:16:29] Security roundtable — hardening your agents `[mixed]` **Q (from the room):** For the level-3 people — when your agents have that much access, how do you make sure they don't shut you out of your own company? **Justin `[H]`:** My second brain has access to a lot — but everything has security protocols on it, SOC 2-compliant tooling, permissions on every layer. And right now, with Fable, is a great time to **harden** — it's so good at that. It's a massive concern for me too; that's why almost everything I do begins local, so it can't go up before it's ready. **Wilson Smith `[M]`:** You can set **hard rules** — in your agents.md, set rules it cannot cross: anything involving money, finances, or sending emails requires your approval. You set your permission rules there. **Oscar `[H]`:** Treat them as employees. I'm the only one with admin access in my company — an agent is never going to have admin rights that could remove me from Microsoft. They get normal user access to exactly the tools I give them. And tools are getting good at this: when you connect an MCP, it asks — read access to this, write access to that. You pick and choose. **Unattributed `[V]`:** Most of the time, if you're getting locked out of something, it's not the agent — it's that you downloaded something malicious that got *into* the agent. Even a great skill like EBI — I'll download it, audit it, and probably recreate my own version so I know exactly what's in my system. I don't run anyone else's skills, ever — recreate them from the features; the AI is smart enough to do it. **Wilson Smith `[H]`:** My general rule: **if I don't want it on a Times Square billboard, it doesn't go in my LLM** and it doesn't get access to it. And I run my skills through a **red-team agent** that hunts for vulnerabilities, with a blue-team agent building defense measures — little personalities. **Austin `[H]`:** We have a **redaction skill**. We record every AI boardroom meeting, and before anything publishes to the site it runs a whole redaction rule set: if people discuss layoffs, their team members, or important financial matters, it's removed unless I get explicit permission. It even automatically recognized that a member mentioned their child's name — and redacted it. *(Q: Isn't redaction built into the transcription tools?)* No — Granola and Fathom print the exact transcription. The redaction skill triggers as part of the publishing workflow, before anything goes live. *(Q: HIPAA?)* For medical conversations you can run a small **local model** to redact first — only the redacted version ever reaches the cloud. Claude's enterprise plan also has specific security hardening. **Unattributed `[V]`:** One more: **stay away from public skill registries.** There are a bunch of them now and most have no listing criteria — anybody can list anything, and you have no way of knowing if someone put prompt injection in it. If you find something you like: "hey, build me my own version of this," audit it, and it becomes yours. Don't trust anybody. **Austin `[H]`:** My mental model: whenever I build a skill, I'm writing the first job description of an AI employee. I've built a **head of security** — one of their jobs is reviewing skills; another is reviewing GitHub repositories and anything else in the download path. `/security` — that's their role, that's their job title. ### [02:24:03] What does a leader actually need to know? — the Geoff Woods question `[H]` **Geoff Woods:** A philosophical question for the room. Listening to all of this — it's clear this group has taken the pill. Then I think about leaders across the board. How much of what you're doing falls into "I took the pill and I'm excited, so I'm leaning in myself" — versus **what are the skills every leader actually needs?** **Oscar `[H]`:** Understanding the language of communicating with AI. Even if you're not the one doing the work — I had to learn how to speak to developers before; now I've had to learn how to speak to Claude to get exactly what I want. That communication skill is essential for leaders now. **From the room `[V]`:** Understanding the possibilities at large — knowing what's actually possible and what will move the needle for the business versus just creating a gadget. And **time in**: whenever you'd normally do something yourself, ask "how could you do this instead of me?" — and let AI teach you about itself. That installs the language too. **Austin `[H]`:** I had the CTO of one of the country's largest financial institutions come to my last AI event with his lead engineer. It was fascinating to see how his brain works — he manages 80 architects who can use however many credits they want, build whatever they want in the sandbox. At that level the CEO probably shouldn't be the one using Claude Code, realistically. His job is visualizing the trend line, communicating where the company needs to be on it, casting that vision — and running the feedback loop with his engineers: what are we building, where are we on the trend line, how do we get there. The leaders of his company lead the architects; the architects do the building and the discovery. **Geoff Woods:** I have a different customer — upper mid-market, a billion-plus in revenue. When I'm talking to the CEO of a household-name brand, there's no way they're doing this themselves. So what's the value of a leader in an AI-driven world? For the people who've taken the pill, "how does a very small company get to $100 million" is a very different game. Established businesses need to know **what's a priority versus what's a distraction**. I personally think the vast majority of AI stuff is exactly that — it's the 80% of activity that drives 20% of the results. I'm interested in the 20% that drives 80%. **From the room (consultant to mid-market clients) `[M]`:** What I see with my clients: a lot of cute things being built — but *so what*? Is it real progress or perceived progress? Where I'm changing their mindset is: **how do I lead and delegate better?** A lot of the reason people get poor results from AI is the same reason they get poor results from their teams — lack of context, lack of clarity. So for every AI project: what is this solving on our bottom line? What is it solving for productivity? The best use I see: elevate the team out of the work they shouldn't be doing so they can apply the institutional knowledge, context, and problem-solving only they have. **Austin `[H]`:** The role of any CEO is to **build the company** — not the individual outcomes. If you're in tech, building the company sometimes looks like pushing code, and for the first time in a long time there are CEOs pushing code again. But the CEO has to know how to hire the right talent, because those are the people who build the company. What are the cues in a job description now? Does the ad buyer you're hiring to run $100K/month in spend talk about using Claude, custom dashboards, MCPs? You don't need to use it for that use case yourself — but you need to hire people who speak that language. CEOs are naturally great at this: they think **who, not how**. They hire not out of laziness but because their goals are too big to reach alone. That's exactly the right kind of person to orchestrate AI agents and AI employees — delegating budget, setting the culture: "we are an AI-led company; when you approach any project, you architect it with AI first, plan it with AI, get the team involved." That's the role of the CEO. They're not in the code. *(Austin: It's a really good question.)* ### [02:35:14] Awards — Austin Distel `[H]` As a thank-you to those who contributed — drum roll, please: trophies to **Oscar**, **Sagun**, and **Justin**. ### [02:36:00] Shout-outs — the room `[mixed]` **Austin:** We close with shout-outs — moments that stood out. Be present with the integration of what you learned; your takeaway might be exactly the mirror somebody else needs. **David Gonzalez `[H]`:** "I want to shout out Peter Hewlett for making fun of me for raising my hand and voting on pretty much all of your things — because they were all gangster. Every idea I had, I was thinking: why aren't people raising their hands for that? All of these things I want to implement." *(laughter — "nothing but love, baby.")* **Austin:** Thank you to everybody who contributed a brilliant idea we didn't get to hear live. I'm launching the **Agentic Show** — long-form discussions with the people here — so even if you didn't get time with me today, I look forward to a live conversation in the future. **From the room `[V]`:** "Shout-out to Oscar — I love how you're always asking *how can we make this a little bit better, maybe think about it this way*. It always expands my level of thinking." **From the room `[V]`:** "From somebody who's between level one and two: just being in the room hearing these ideas was incredibly inspiring, all over the board. Thank you all for sharing the process." **From the room `[V]`:** "Shout-out to Austin for holding this." **From the room `[V]`:** "For me it was an eye-opener about how I want to use AI more — and how not. I don't want to reduce my human voice at all. I want to use AI to give my people **more hours** — to create even more human jobs. I never thought about it that way before today." **From the room `[V]`:** "I'd like to shout out Geoff for that question. That's the question we need to be asking: are we doing the right thing, or are we wasting time on outreach that doesn't matter?" **Austin:** Geoff's book — *The AI-Driven Leader* — the update comes out next week. **Geoff:** "It hasn't changed the philosophy — I wrote it to be timeless. I only mention one AI model, and I focused on showing readers how to use AI for strategic thinking, which is timeless. I've updated stories and language." *(He also helped with* The ONE Thing *with Gary Keller.)* **From the room `[V]`:** "Thanks to everybody who shared not just what they built but *how* — the tools. What's really cool about a community like this: you get guys like Justin who go 'this tool, this tool, this tool' and essentially give away the playbook — because we all know there's so much business out there, and each of us will have a unique take on how to implement it." **From the room `[V]`:** "I think it was Oscar — when you bring a new agent online, start it **read-only first**: an email address, read access to what it would do in person, then write access to what it would do in person, and go from there." ### [02:41:16] Close — Austin Distel `[H]` That wraps this mastermind. The next free one is **August 7th** — scan the QR code for Luma, and subscribe to our Luma profile to get notified whenever a new mastermind is posted. There were **30 people on the waitlist today** — you all got in the room, but register early; we filled up a week out, so 7 days ahead is right on time. If you want to meet with us every Friday morning — my number's on the back; you can schedule time with me this week and we'd love to see if the Agentic Society is a good fit. With that, we're moving to the lobby to network — you're welcome to stick around and co-work. **Group photo in the lobby in about five minutes.** Thank you — and I hope you all have a great 4th of July.