# AI Boardroom — Agentic Society Members · Clean Transcript
**Date:** 2026-08-07 · **Type:** AI Boardroom
**Location:** Industrious East 7th, 924 E 7th St (Conference Room A), Austin, TX
**Duration captured:** ~2h 24m (in-room, one shared mic + a tabletop recorder)

> **Timestamp note:** times below are from the RAW recording (Zoom cloud
> timeline). Re-map them to the published/edited video before enabling any
> "watch at" links.
>
> **How this was built:** the room was captured on a single shared in-room
> mic, so "who said what" was reconstructed by triangulating a tabletop
> recorder's separately-tagged voices, the Zoom AI's chapter summary, direct
> address ("Okay, Jay?"), and self-description matched against the calendar
> invite.

## Who was in the room
| Name | Business | Notes |
|---|---|---|
| Austin Distel | The Agentic Society (host) | facilitator |
| Roman Bediner | The Agentic Society | AV / systems |
| Mike Couvillion | Alexa Guru | advisory CTO win |
| Chance Forman | Lillybug Home | dashboard demo |
| Justin Day | Day by Day Productions (Local SEO) | client win |
| Adam Weisberg | Restaurant operator | dashboard win |
| Darby Rollins | Gen AI University | content-distribution win |
| Jay Douglas | KeyFive | security-automation win |
| Oscar Hidalgo | CG Service Pros (HVAC) · Vindex Consulting | Hermes deep dive |
| Lauren Goldstein | Golden Key Partnership | self-management win |
| Cristian Aguilar | The Agentic Society | event content |

---

## Transcript

### [00:08:39] Welcome & setup — Austin Distel
The group tested microphones in its new, larger downstairs room at
Industrious — "that's all the microphones in the ceiling, which is sweet."
Austin confirmed the room is booked every Friday through the end of the
year, with one exception: "The only caveat is on the 21st, which is the next
boardroom. Darby will be leading that one, and I'll be out of the country
during that time."

He also flagged something several members had noticed about their meeting
recorders: "Have you noticed that it no longer joins calls? I see it's
recording, and it's transcribing, and it's collecting video, but it's not
actually physically joining the call." Nobody had seen an announcement about
the change. The room's consensus: "If it's still recording, then I guess who
cares."

### [00:12:34] Framing the day — Austin Distel
"Today's focus is AI for marketing," Austin said. "I like to keep that as
the primary topic — conversations about growing revenue, different channels —
as a theme for our content and what we're creating over the next sixty
days."

He previewed the member-led teaching coming down the pipe. Oscar has joined
the team and is creating content around three things: setting up a Hermes
harness for AI agents; "an SEO automation that builds high-quality,
localized pages to bring in traffic — that's how he grew his HVAC company";
and cold email prospecting — "how he's been able to cold-email apartment
buildings in Austin and collect revenue. I think we can all learn from that
setup: how do we find our target market, enrich those leads, and then create
an agent that books calls on your calendar." Justin is creating content
around agentic SEO and "how to make websites that are self-improving,
self-healing, and growing," coming in September. Meanwhile, Will and
Cristian are building SOPs around video editing and social content at scale
— Will has developed his own software for clipping long-form video with AI.
"All these playbooks you'll start to see show up over the next sixty to
ninety days. That's our big sprint — ninety days."

Then he opened the floor: "Go around the room, share what wins you've gotten
using AI over the last two weeks — or longer, whatever's showing up for you.
What's the one win that you've gotten recently using AI?"

### [00:15:16] Win — Mike Couvillion: an anthropomorphized agent team, and a one-day close
Mike went first, crediting the seed of the idea to Jay and his
project-manager agent. "I recreated my programming team to actually be way
too anthropomorphized — they actually have names and everything."

The design is deliberately tiered by model cost. "I ended up building a
five-person team, where one is kind of the architect — it's equated to
Fable. Two are workhorse programmers — it's equated to Opus. And then a
couple of juniors that are Sonnet, but at different thinking levels. Because
if I'm just updating documentation, I don't need to spend expensive tokens.
But if I need hardcore thinking, planning — not debugging, but architecture —
then I want to use Fable for that. It's helped me balance my token spend,
because I was doing everything with Opus at one point," and burning through
his allowance too fast. "Now I've been able to stretch that out — get the
same amount of work done in less tokens."

Austin connected it to a pattern Roman built for the Agentic Society's own
fleet: "Something Roman has added to our fleet of agents is we talk to the
orchestrator first. This is basically the boss of the bosses — the manager
that's saying, hey, what level of skill set do we need on this job? We don't
need to put our best guys, the Fable Five, out there for grunt work that can
run in parallel on Sonnet. The default is Opus, and for the big behemoth
tasks, it's Fable."

Mike runs the same way. "I always had the orchestrator in place — that
dashboard I showed you guys last week is tied to an orchestrator. Typically
I have tickets, issues, and it just goes through and assigns them according
to the scale of what the ticket is. The individual agents know — oh, you've
given me something that's simpler than what I can do, or more complex — and
it will reassign to one of the other agents. And they'll communicate through
a Slack channel, so it's actually interesting to see them talking to each
other, trading things back and forth. It's kind of scary at the same time.
But interesting."

Then the payoff. "Having that allowed me to meet with the client. The way
the lead came in was not through AI — it was through personal contact.
Meeting with the CEO, showing him what I had — and closing the deal took
less than a day. Normally that's a two or three month sales cycle." The
client was a sizable business with only a two-person programming shop, and
the one person who built everything had just resigned with two weeks'
notice. "He knows where all the skeletons are buried. So I said, I'll show
up and help you work on the project and support you until you get new hires
going."

The timeline underneath the story: "This was literally — I worked on this on
Saturday and Sunday, I got the call on Monday, and here it is Friday,
wrapping up the work now. We haven't even signed the contract fully yet,
because the lawyers are still reading things."

Asked whether this is retainer or project work: "It's short-term, because as
I've told others, having been a CTO for thirty years, I purposely don't take
long-term jobs. I come in, I tell you what's wrong." He covers the client
for a couple of weeks until they hire, and helps with the interviews. He
expects the system to keep paying off: "It'll make it easier to market what
I do, because I can now put it in a nice package. I do anticipate it'll
shorten the sales cycle by at least half. And it also just makes the work
much more efficient now."

### [00:20:31] Win — Chance Forman: the dashboard grows a content pipeline
"Man, I've been cranking on this dashboard for a good little while — I mean,
months. Everyone wanted a CEO dashboard. Well, I manage a portfolio of
companies, so I've built all my stuff going in there — KPIs, metrics, and
then it has a project board that I can work with."

Two team members are already inside it, each with their own agents. "It's
similar to what Mike's saying, but I have Codex, and Chat, and OpenClaw and
Hermes inside there. Whenever I sign a new C-level member or manager or
partner in — that's what I'm trying to figure out now — they can just
connect to the harness that they choose and create their own agent that
helps them inside the dashboard the moment they log in. It has the same
brain and thinking system, where you can pick between the models, and it'll
downgrade as well."

This week's win: "We created a content pipeline inside there, where I can
just take photos and videos throughout the week, or the day, batch upload
them inside there, and then I can either push them to AI to edit — and we've
been building that — or AI will give me a sub-edit, like a seventy percent,
and I can push that to Nick and get it a real full edit. I've been launching
a new company, so we're testing it there. We're able to put out content
daily pretty quick. Not only is it a dashboard — now I'm able to live my
whole day within one platform."

He gave the room a preview of the design idea he'd demo later: "Inside each
one, it's a pipeline, similar to a project pipeline. You can click on — I
call it Coach, but Coach is: if you talk to it, it's Codex, because the chat
interface is better, but if you text to it, it's Claude. When the project's
in there, you just click on Coach, and it has all the context of those
files, whether they were the raw uploads, a summary, a brief, or whatever.
It'll give me a script, a hook, and at the end it has performance — it's
going into Instagram, uploading anything I've loaded already, giving me my
performance, and then auditing that: hey, this is what's working, this is
what's not, let's switch it up this time." They're extending the same loop
to paid ads, starting with Meta.

Austin flagged it for a deep dive later in the session: "Social media
content in an agile, quick way — capture, edit, and then distribution. There
are three big phases that each have to be automated in their own unique
ways."

### [00:24:16] Win — Justin Day: a second brain as a retention moat
Justin is building a "second brain" for one of his whale clients — an
event-rental company with a massive inventory of over thirty thousand
products, whose websites he built and whose ads he also runs. "This week I
was able to leverage a WordPress MCP, go into the data within WooCommerce
and everything, see what products are actually selling, and provide them a
full breakdown of what products are moving in each market, where the
inventory should actually be pushed, where the warehouses need support —
giving them insights year over year on what the growth has been, the
seasonality of each of the products, so we can really help strategically
grow their company."

The context makes it a retention play, not a deliverable. "I'm in a shifting
sands scenario where they brought in a new CMO. I've been with them since
the beginning — we're doing seven figures a month for them on organic SEO
alone — but you can get these results for clients and they'll still bring in
someone else. So this week, a big win was just showcasing them things
they've never been able to even see on their end, because they're such a big
company with fragmented data. Now that I'm building the second brain for
them, being able to showcase this, give them ads that perform better, and
give them clarity on all of that data — that's really helped win over the
client and keep that retention there."

The client is up 20% year over year from his services alone. "Been in these
scenarios before — people still move on, even though you make millions for
them. So I'm just trying to dig in and create my own moat. It wasn't in
their scope of work by any means, but with the amount they're paying me
every month, and how much I've made them, I want to continue to showcase
what we can do. Going above and beyond with AI is easy. Because I've got all
the data and they don't, this is how I make sure they don't switch companies
— because who else is going to do that? Especially with the new CMO coming
in now, I'm like, okay, I've got to show this guy what's up. Big data guy.
All right — I dare you to switch."

Austin's read: "You're using AI to increase retention by over-delivering
value in ways beyond the scope of just SEO. This is the advantage this room
is going to have — being able to differentiate from everybody else that's
not using AI, going above and beyond for clients, and getting results faster
than ever."

### [00:27:43] Win — Adam Weisberg: a restaurant KPI dashboard on OpenTable + Toast
Adam picked up the dashboard thread directly. "I love his dashboard — it
really inspired me to just scrape mine and start over. So I started working
on my dashboard, brought in OpenTable, and Toast, and some of the things
that we talked about. That's really helping me get a better idea of my KPIs,
and I plan on sharing that with the team." To Chance: "Yours is baller, what
you just described. It's actually re-inspired me to go look at what I can do
in the next two weeks, based off of just what you described."

Chance answered with the origin story of his own build: "It started as
everyone doing the Jarvis on Instagram, and it had no purpose — but I
thought it was cool when I first started building eight months ago. And then
that turned to: okay, this really needs to work for me. It can't just look
cool." Adam: "Every time I've been in this room, I've been incredibly
inspired."

Austin noted the trajectory — Adam joined about a month ago just using the
chat, and is now building dashboards and standing up his first agents. Adam
credited the community: "Darby's really been a nice help — I've been pinging
him a little bit, so a shout-out to you, Darby. I've never done AI agents
before. I literally just started Wednesday evening trying to load in the
CMO, and I'm really glad, because I didn't know about the orchestrator
agent, so I'll make sure to get that. That was my sixty-day goal that I set
for myself about forty-five days ago — so I've got to get in there."

Austin reframed the pace: "You're not behind. Ninety days — that's twelve
weeks, six sprints, between the first and third Fridays. You can get so much
progress in these sprints; I think it takes about two sprints to level up.
Over the course of six sprints, anybody can go from chat — to co-work,
connectors, skills, artifacts — to dashboards — to actually setting up a
harness, agents, orchestration, and all of that. It's hard to jump and see
the whole path there, but within six sprints, I definitely can see that for
you."

Adam also shared where he is personally: twenty-six years running
restaurants — he owns several in Austin, plus one licensed location out of
state he was flying to open that weekend. "I'm the dumbest guy in the
building," he joked about being in this room. "I've been doing restaurants
for twenty-six years, and I'm going to continue doing that, because I still
love it. But this is kind of the first time where I'm like — no, I'm really
going to step out and away from it, on probably more than a part-time basis,
and just let them giddy up. These subject matters are literally the things
I'm most excited about learning."

### [00:32:36] The Toast API unlock — Adam & the room
Austin followed up on a blocker from their last call: OpenTable and Toast
data, which don't have easily public APIs. "No, we got it — so we're good
there," Adam reported. The unlock was being an existing customer: "Since I'm
with Toast, you can do a general one or you can do an authenticator. I
actually got the wrong one the first time, so it jammed me up for a day. And
with OpenTable, it's the same. If I wasn't with Toast, I'd have to be a
billion-dollar company to get an API with them, because they're so backed
up. If I wasn't a customer, I'd have no shot." Other restaurateurs can get
access to their own Toast API the same way.

The goal the data serves: growing specific days of the week and tracking
demand. "I plan on going a lot deeper into it — I just got done with that
piece probably forty-eight hours ago, where it's actually stable. It kept
breaking; I think I got it fixed. Now I want to do a much deeper dive into
the numbers. Hopefully I'm going to figure out the AI agent that can just
give me cron jobs every day, for some very specific numbers to share with my
team." His next project after that: "My hope is to redo all my websites and
build them myself. That'll be my next thing — doing the agentic websites and
trying to learn that."

### [00:34:52] Agentic websites — Justin's system, previewed
That teed up Justin's upcoming teaching module. "I've created an in-depth
system on how to actually create these websites, because what I've run into
— in this community and others — is: how do I even start with creating a
good website? You can go into Claude and say 'build me a website,' and the
output looks like every website out there. What does it actually do?"

"So I created an end-to-end system based off of my ten years of experience —
what I know converts, what actually matters for websites — piecing that with
design and resources. The best way to get websites is to run it in a
specific order of stages: onboarding yourself, getting all the questions
answered, understanding who your target audience is, what your brand
guidelines are. Then the architecture of the website — the sitemap, what
needs to be on each of those pages. Then the copy, and in that copy
methodology, go ahead and implement SEO. Now move into the design components.
And at the end, when we've got all these pieces, we put it together and you
start to iterate. Doing all of these things in the right order gets you a
phenomenal output. That system I'm just going to be giving away, and going
through how to use it — go to town."

Adam confessed the timing: "I met with Justin about two weeks ago, because I
was so interested in what he did. I said, man, I'd love to talk to you about
you doing our websites. And he's like, 'Well, I'm going to be teaching on
it.' And I was like — okay, I guess you just talked yourself out of our
websites." Justin is also presenting the same material at an outside
marketing event the following Thursday — "the shifts on Google, how to be
found and make more impact, what agentic websites look like, and the
methodology — and I'll be giving away that website system there too."

### [00:38:02] Win — Darby Rollins: solving distribution, not production
"My win this week — I've mentioned before, I've been working a lot on the
content side of things, and distribution has been a big bottleneck, from
content production to distributing," Darby said. "We finally got the first
full batch of shorts from our editor, and I've been working on a workflow
with Claude and Codex to create a content production dashboard — where we
take all these different shorts and create different variations of the cover
photos and images, so we can continue to cycle the same shorts five or six
posts apart, and have that consistent content drip out to our YouTube and
Instagram."

"I built that whole system, and now it's starting to work. Every day it's
posting across six of our channels — YouTube and Instagram are the main ones
I'm worried about doing well. Still relatively small in terms of viewership,
but the daily consistency is getting five hundred to a thousand views a day
on some of these, without me having to go in and post it. A lot of that
initial lift of getting the system built is done. Now I can get a new batch
of content in, tell Claude or Codex to look at this folder, run it through
the entire content-production pipeline — identify snapshots from the
long-form video, turn them into a ton of still-shot variations — and
automatically start to post it."

He's layered engagement automation on top: "On Instagram, I have it set up
to where anyone that comments a certain keyword automatically gets a link
back with more information — similar to what people use a separate chat tool
for, but on its own."

The tool underneath is Blotato. "They built an API and MCP that can post to
about nine different social media platforms — that was the initial reason
the tool got built — and they just recently introduced the comment-response
feature, so I don't need a separate tool for that now. What's interesting is
it tracks the analytics for the ones I post through their API — oh, these
videos are doing much better here, this is how much people are watching
them. It's still pretty early on, but just knowing that if I had fifty
shorts ready to go tomorrow, I could run it through this and have two
months' worth of content scheduled — that makes me happy. That's always been
the bottleneck: how does content actually get out there, once we spend all
this time making it?"

On what else it does: "Inside the tool you can scrape content, source
content, identify outlying trends — it can be used as a research tool too.
You can connect different image-generator tools through it, so it's sort of
an aggregator — you can create carousels. But when they came out with their
MCP, you could just connect it to Claude and do all this stuff from there —
I'm never inside their dashboard outside of connecting the accounts."

### [00:41:28] The Blotato founder story — Austin & Darby
Austin asked the room: "How many of y'all follow this AI influencer named
Sabrina? She's the founder of Blotato." Darby filled in the story: "She was
a software engineer, had some type of exit from her company, took a break,
and came back saying — I don't really need to make money from this stuff, I
just want to help people learn AI content. She created the tool because she
didn't want to spend her entire week posting and scheduling to social media.
Now she's a content creator who documents her whole process — she does
content one day a week, on Saturdays, and puts out tons of it." Darby has
had her on virtual summit interviews before, and the two teams are now
collaborating: "I'm doing a fifteen-minute video with their channel,
breaking down a lot of what we talk about with the company-brain stuff — how
to apply it to building a content engine that can work with you."

### [00:44:27] LinkedIn's broken API — the room
Lauren asked whether Blotato could pull analytics out of LinkedIn — "that's
a challenge we're having, trying to get more data on our posts, and LinkedIn
is being a little difficult." Darby wasn't sure how deep it goes there.

Austin relayed what he'd learned that week from the consultant who manages
his own LinkedIn: "LinkedIn has a pretty bad API. Even him, as a
professional LinkedIn consultant, can't get a working API from LinkedIn —
which is wild. You're supposed to be able to export reports, and it's
erroring out on that. And from what he was telling me, even the LinkedIn
softwares that are supposed to work on the API are not always functional,
because of LinkedIn. The professional social network of the world is
behind."

Jay confirmed the diagnosis: "On LinkedIn, I think it's by design. They make
it difficult to integrate with, for the very reason of the paid upsell
products — I give them a fair amount of money for those." He noted that
third-party workarounds exist that crawl the site outside the official API
in a daily cat-and-mouse game with LinkedIn's defenses — the room treated
those as a buyer-beware category rather than a recommendation. The practical
path Austin has landed on: "Rohan has figured out ways to do it with cron
jobs. He does weekly reporting for me — conversations, number of connections
added, number of impressions made. All of those metrics can be found, and
then you can have an agent pull that, outside of the API."

### [00:47:12] Deep dive — Jay Douglas: the security-fix pipeline
"My big win: we're constantly going through cybersecurity audits — right now
it's a SOC 2 — and these consultants are expensive. I come in and manage
that, from the findings to the triage to the implementation. So I created a
suite of agents that take the finding, do a triage, come up with the fix,
and then automatically create the ticket inside GitHub Projects — Roman
challenged me on that a couple weeks ago. Then there's a human review gate,
and then it passes off to the agents, and they do the work — implement it
and put it in our staging environment. I was basically able to conclude a
contract with one of the contractors that was managing that. It's all
automated now, from triaging the vulnerability, to coming up with potential
fixes, to actually implementing those fixes."

Austin: "It's solving itself — so, self-healing?" Jay: "It is self-healing,
and it learns. A lot of times you have to be pretty explicit about
infrastructure changes — it's a least-privilege model. It'll say, okay, we
need to implement this privilege; then after that's approved, it can spin
off another agent to implement that privilege and roll it through. And our
SLA on implementing fixes on these findings went from three days to minutes
— or an hour for the most complex things."

Asked what model runs it: "I have something that kind of picks between
Claude and Codex, or different things — I run it through a model router, so
it'll use whatever's the best agent for the job, like the orchestrator
you're talking about." Austin: "Fable Five for security?" Jay: "It rarely
picks Fable Five, for some reason — I don't see that come up often. Opus
Five is the primary, and it'll drop to Sonnet, and then some of the lower
Codex models we use quite a bit as well — especially for ingesting data,
manipulating data, those types of things."

He described the router itself in plumbing terms: it presents one
open-API-style interface, you can put the CLIs from any of the products into
it, it aggregates the models behind it, and it manages token spend — "a
great abstraction layer between the harnesses and the actual models
themselves. Now you have one interface you can integrate into almost any
harness." The room compared notes: Composio came up as the equivalent layer
for tools — "Composio is more for grabbing the tools, connecting multiple
MCPs. This one is more for routing the model — if you want to use Claude for
one task, OpenAI for another task, Gemini for another, it allows you to have
them all in one place."

### [00:50:47] Deep dive — Oscar Hidalgo: the zero-touch week
"Because we're working on a lot of Hermes stuff, I challenged myself this
week to do zero work, and let Hermes do absolutely all my work — across my
CEO job at the HVAC company, preparing the courses for the Agentic Society,
and doing client work; I do AI consulting on the side as well. My challenge
this week is I will not touch a single tool to do any work. Learned some
lessons — but basically, it's very possible. So now I have a crew of twelve
agents on Hermes, with one orchestrator."

The week forced a full stack audit: "My goal now is every tool that I use
needs to be agentic first. I've been cleaning and switching up a lot of
stuff. Instead of the email provider I was using before, now I'm using
Superhuman for email, because they have an amazing MCP that connects to all
of the agents — Claude, whatever. For the browser, I've been testing a new
browser built for agents, which allows your agents to open tabs but doesn't
take over your computer — you can still use it. Because agentic is the new
way of working, all these different apps are switching to being MCP-first
and agent-first, so I've been doing a lot of cleanup on that. I moved all my
agents to a remote server, so now they're accessible even when my computer
is off, and I can talk to them from any of the channels. And I connected my
second brain and fully synced it with that remote server as well, using
Syncthing. My goal is to find new tools and new things that improve these
workflows and bring them to you guys, ideally every time we meet."

### [00:52:38] Oscar's data-pipeline pattern: stop re-fetching, start hosting
"On the dashboard conversation — I'm currently working with a client that
has a big e-commerce company, and they're currently using their agent, just
Claude, to pull the data weekly from different sources. It's grabbing some
MCPs, it's grabbing some browser automation, and that's burning tokens like
crazy every time he does this one little task."

"So, a couple of tools I found for you guys. Composio is one, and there's an
alternative called Airbyte — these are data pipelines, so you can connect
all your data sources, all your logins, to bring in Google Ads, Facebook
Ads, all this data. And then there's a database provider called ClickHouse,
which is open source — you can download it onto your computer. So you can
set something up through Airbyte, pull all the data from all your different
sources, host it in ClickHouse or any other type of database, and then your
agent can just create the dashboard front end. That way, every time you're
pulling data, you're not burning tokens on an agent working the MCPs every
single time. These exercises of challenging myself — okay, this is not
something AI would normally be good at, how can we make it good at it? —
are making me find a lot of little tools on GitHub, open-source stuff people
are working on."

### [00:54:03] Austin anoints the group's Hermes advisor
"You guys know Oscar from several of the masterminds now, and he's
officially our go-to Hermes instructor," Austin said. "Courses are coming,
and we're taking on advisory clients now. We're always seeking to build what
you all want — I've said it a hundred times, we only build things people ask
for, so if you ask for it, it gets to the top of the list. Many of y'all
have asked for project management tools and dashboards, which is why that's
been the focus so far. Now that we've started to conquer those milestones,
we're upgrading to things that have been requested — the cold email
prospecting, the agentic SEO, and how do you go from a Mac Mini to an
online-hosted AI agent so you don't have to have your computer open. Who,
not how — remember? I think Oscar is the smartest user I've met on Hermes.
Over the next six weeks, expect three courses, and more on the way."

"Why I'm also impressed: Oscar's the CEO of an HVAC company. He's applying
AI directly to a real business — not metaphorical AI, theory. He's getting
tangible results for his companies and clients, and that makes him even more
qualified to speak on this."

### [00:56:13] Hermes Q&A — the room works the topic
**Justin — build a fit quiz first.** "What would be super valuable for
everybody as an entry point is a website, a short form, that describes where
you're at with the company, what your goals are — so an AI can tell you: is
Hermes a good fit for you at this time, or in the future? What would the
benefits be for your scenario? That's the biggest question for people — is
this just me shifting over here? Do I actually need this? What are the
capabilities?"

Oscar agreed — it's exactly where his course starts. "The course starts with
'why Hermes?' Why does it even make sense — we're all using Codex or Claude.
My point of view is that it's not either-or. Claude or Codex can still be
your active working partner. The benefit of Hermes is that it has individual
agents you can create — they call them profiles. You can name them, they can
have their own soul.md, which we reviewed last time. Each agent can operate
its own way, and you can use a different model for each. If you know a
particular image model is better for generating images, or you like it
better, you can have an agent that only does that really well, connected
through that model. We're targeting anybody in here that considers
themselves at level two — 'I'm already using Claude Code actively, and maybe
I have some automated tasks going.' Hermes is the next level of that. And
again, not either-or: you can set it up on a Mac Mini, on your computer, or
on a server, and do more automated stuff in the background that you don't
always have to prompt."

**Chance — the contrarian take.** "I'm using Hermes as well, and I started
with OpenClaw — I still use it, because I've finally gone through all the
struggles, and it's good now. But I'm hearing everybody — more agents and
more agents — and I've been going in reverse. I have a Hermes agent that's
my orchestrator under Hermes, and OpenClaw the same, but I haven't created
any more agents. I'm using MD file structures and letting it spin up the
sub-agents within Codex, and Claude has more of the skills, if you just have
it orchestrated. I think a lot of people are so heavy in agents-agents-agents
— I was, in the beginning. I wanted to build this whole infrastructure and
org chart, and I wanted them to talk and be in rooms and be able to see it —
and the more I studied it, the more that was overbuilding: complex, causing
more disruption than helping. So I've really been more on the MD files and
skills lately. Could you talk to why you're going twelve agents deep?"

**Oscar — token economy is the answer.** "Absolutely. More and more, even
big enterprises are dealing with token economy. The smarter the models get —
whenever we want to use Fable Five, that's burning through tokens hard. So I
don't want my agents trying to figure out which task I'm doing. I want one
agent that's really good at a specific task, and it only has the connectors
to do that one task really well. When I ask 'can you pull a payroll report,'
I don't want it doing browser automation, opening Chrome, taking a bunch of
screenshots, trying to log into my payroll software. That agent knows, in
its soul.md, that the only way to run a payroll report is through the MCP to
my payroll software."

"It is doable through Claude, or a project in Claude or Codex. To me the
benefit of Hermes is that in your brain it makes sense — oh, this is Harvey,
he's really good at payroll. It's a good way to separate them, and the
desktop app now lets you switch between profiles really easily. If I have
one agent that's very good at everything, at some point, when I don't give
it enough context on a prompt, it might do the task the wrong way — it might
go through the wrong connectors, it might open the browser and burn through
half my usage for the month, when all I needed was one specific MCP. And I
don't want to be saying, every time I need to do payroll, 'you do it this
way.' My orchestrator knows: anything related to payroll, send it to Harvey.
Harvey knows the only way he works is by pulling the MCP. And that agent I
have set up on a lightweight model, because it's only pulling reports — it
doesn't need to think. It lets you get granular per use case. This agent
only does outreach, and you need three skills and four connectors to do
that — and that's all I'm going to teach you; you don't need to know
anything else."

**Lauren — how long does setup actually take?** "I've had 'set up Hermes' on
my calendar for three straight weeks in a block, and I keep going — is this
really going to take this amount of time? This might be a dumb question, so
forgive me, but a good starting point to add to your course is: how long
should you budget to set it up? What's the MVP, where you can just test it
and see what else you need to add? Something I tell myself and my clients
is: just start with the MVP, then figure out what you're using it for. And
because you're using it all the time, I'd love to see use cases — here's
something you probably haven't thought of that you could use it for. I think
I have PTSD from setting up OpenClaw."

Chance jumped in with his own data point: "OpenClaw set up my Hermes in two
hours, with Telegram. I put them on the same Mac Studio and said, hey, I
want to try out Hermes for just this one company — so it learns the brand of
that company only. OpenClaw set it all up, and I just had to log in to
Telegram and a few things like that. It was less than a two-hour setup. And
I love Hermes compared to OpenClaw because it shows when it creates a new
skill for itself, what it's working on, when it needs an approval. I talk to
OpenClaw through iMessage, so I can't just approve with a button — but in
Telegram or Slack, they'll send you an 'allow' and you can just approve."

Oscar's MVP answer: "Honestly, they made it pretty easy now. It's a desktop
app that you download, just like Claude Desktop. You download it, it
installs everything in the background, and now you have Hermes. You just
need to connect it to one model — and that's your first agent. That should
be your MVP." Chance added a cost tip: "Give it Codex, because of
third-party token usage on the harnesses — there's a cap on how much you can
route through Claude that way, but it's unlimited with Codex. Then you still
give it the MCP or the API to Claude, where it can pull it in when it needs
it — but it's unlimited brain on Codex." Oscar: "If you have a ChatGPT
subscription, that's still the best brain for Hermes, because it's included
in your subscription."

### [01:04:39] Oscar's HVAC operations stack
"To get people thinking about what's possible: I have it running payroll
reports. I have it connected to my QuickBooks and running full P&L
forecasting for the next six months of the company. We use a separate app
for our field team — our technicians — to book their hours, and that one has
an MCP as well, so at the end of every day it's checking their hours against
what's in QuickBooks to make sure payroll is good."

"We use Housecall Pro for our normal service technicians, and we use
Connecteam for the projects we do for the city. On those city projects, they
need to submit multiple pictures daily — so at the end of a project I'm
dealing with three or four thousand pictures. The system creates multiple
folders, separates them by building, by apartment unit, before-and-afters,
and gets them all ready on one link to share with the city. That's the type
of stuff that's possible once you get through the hurdles."

"But there is a way to start very easily. Something like Composio is what I
recommend to anybody starting out — connect all your MCPs there, give it to
Hermes once, and you're done. It's one place to put all your MCPs." Chance:
"It has easy connectors for Google Drive, Notion, all that — one click, sign
in, connected. Composio handles the multiple-Gmail problem too. If you give
it super Google access, it can do everything — calendars, emails, Drive, all
of it." Austin asked Oscar to build a lesson on it: "It seems like it's part
of Hermes setup — where do you put all your MCPs and APIs?" Oscar: "That's
been the challenge — how do I make this replicable to everybody easily? By
the time we have the training set up, I'll have an easy way anybody can get
started." Chance's power move: "I plugged Composio into the back end of my
dashboard. Now when I bring on a new user inside my company and they want to
connect any of their stuff, they just go to Connectors in my dashboard and
connect. It looks like it's internal — but it's going through Composio."
Oscar: "That's genius."

**Lauren — installing this for clients, not just yourself.** She described
the reality of her fractional-COO work — "a lot of my day-to-day is herding
wet cats with my clients" — including one client whose owners could clearly
benefit from an orchestrator, if someone could get it installed for them.
"Something that would be helpful is not just installing it for yourself —
if you do consulting, what's the best pathway to get this installed for
someone else?"

Oscar: "That's another perfect use case. If I set up one Hermes on a server,
I can create separate sub-agents that have set permissions, and those are
the agents that I share. You can create Telegram or Slack or whatever
channels you want for each one — one agent that's only for this executive,
with only these permissions and this access to the second brain; we make him
a Slack bot, and it only interacts with that person."

### [01:09:28] From Hermes to company brains — Lauren's question
Lauren: "Should I make the company brain for the whole company on Hermes?
This is unlocking my brain now."

Oscar: "It connects to the company-brain training we already went through,
because a company brain is just files — and Hermes can access files. You
tell it which folder is the home folder. If you have a company brain, you
just point Hermes to that folder — and then each agent can have its own
little workspace and its own permission set. The design group can have their
little agent that only has access to the design folder, and it will never
cross that boundary. Or the executive can have one with access to the full
second brain."

Lauren: "I was going to do this in Claude, but this sounds better." She
suggested the comparison itself was worth a module: "Hey, you want to set up
this company brain that lets your team access different things and ask
questions, so we field fewer questions and they're more empowered — is it
better to do it with Claude, or with Hermes?"

**Chance's version — Obsidian as the hub.** "What I did — and I don't know
if it's right or wrong — on the second-brain thing, is I just did Obsidian.
Whether it's Claude, Codex, OpenClaw, Hermes, my dashboards, my iMessage
with OpenClaw, my Telegram with Hermes — every night at eleven fifty-nine it
sweeps any new action from that day up into the Obsidian account. And then
every morning at four thirty, it pulls a summary per platform — this is
what's updated, this is where we're at today, this is what I need to feed
OpenClaw, this is what I need to feed Codex. So it has fresh context every
morning by the time I wake up, from yesterday's action until midnight. The
orchestrator can go pull the whole brain, but every agent doesn't need the
whole brain — it just gets the small daily summaries. I do different things
in all the different platforms, but the context has gotten so much better.
If I'm over in Codex, it knows what I'm working with in Claude."

Oscar noted the enterprise end of the spectrum: there are database-backed
second-brain products where permissions are layered and people log in with
their company account to reach only their part of the brain. "Depending on
the size of the client, there's a way more customized way to do it. We can
probably target a few use cases — if I was doing it for myself, here's how I
set it up; for a small company, this way; for a bigger enterprise, this
way." Lauren's client is exactly that hard case: "They have a parent
company, but then a design-and-engineering firm, a property management
company, and a construction company. They've all been intermingled, but
we're separating entities — and that creates complexities, where some people
will still need access across." Oscar: "Exactly. So this could be a good
infrastructure."

### [01:13:20] The company brain, taught — Austin Distel
Austin walked the room through the framework the group teaches (the recorded
training is being edited into a polished course).

"What we're currently teaching is using Google Drive — because it's
agnostic, and because it already has permissions and governance in every
single big company. Most companies use Google Drive or OneDrive, some use
Dropbox — but as a general theme we recommend Google Drive. A shared drive,
not My Drive. Downloaded on the desktop app, so it lives locally and it's
connected. And then when you're using Claude or Codex, you point it to that
folder inside the shared drive."

"Darby has created that training and built basically a house blueprint of
this whole drive setup. All you have to do is import it through our MCP, or
go to the site and download the zip folder — it's a self-contained company
brain with all the instructions for the agents to onboard itself. A few of
the features built in: **dreaming** — Justin Day's idea. We all go to sleep,
and we dream — we learn, we think, we review ourselves, and we try to get
better. We digest what happened that day, and then we think about what
tomorrow holds. That's basically what your nightly cron job is; there are
directions inside the company brain that carry that instruction when it
onboards itself. Another is the **mailman** — like a librarian. You brain-dump
all the files you want into that folder, and the directions of that folder
rename every file and put it in the right department. We all don't have time
to be the filer in our business."

### [01:16:12] The bookkeeper story — a self-built AI CFO
"I recently did this, so maybe I'll share my win here for a second," Austin
said. "My bookkeeper has been asking me to review these expense reports and
get this last quarter all updated, for a while. And I was like — okay, I
could either just spend three hours doing this, or I could train an AI agent
to go in, review all of our emails across all of time, get learnings, and
build a skill for itself on how to be my CFO."

"It created an organized structure for every bank statement, every expense
it sees through my Mercury accounts — because Mercury has an MCP — and
itemized all of it into a folder structure my bookkeeper has access to. Then
it drafted her an email: 'Hey, pretty much all of what you asked for is
already done now. I went ahead and organized every folder by month, with the
expenses of everything you asked for.' It pulled the context of
conversations I've had — email chains with my team that said 'hey, I just
signed up for that new tool,' where my bookkeeper had asked 'what is this
expense for?' Because the AI had access to my email, it was able to
understand the context of each expense. And then it drafted an email for
her. I didn't need to change a thing. So I just sent it."

"Now my bookkeeper and I have a new process. Every Monday, I have a cron job
that looks at all emails related to her firm, drafts replies, and tells me —
hey, I've drafted a reply for you to your CFO team."

### [01:18:15] The job-description method
"The cool part of the brain was that it self-built its own department. And
one way I did that — y'all know how to do this, because you hire people. You
put out a job post: hey, I want to hire this person because I need to solve
this problem, and the qualities I'm looking for are the best of the best.
Do you go to the industry conferences? Do you read the right books? Do you
have the right certifications, the right degrees? And you just say: go and
research what that would be for your industry — head of HR, head of finance,
whatever — and build yourself to be that."

"After that: here are all the tools that kind of person would manage. So now
my AI CFO has access to QuickBooks, Gusto, Mercury, and the other financial
instruments that role would use. It's a self-contained thing — it organizes
its own workspace. I didn't even have to think about that, because I don't
know how to organize a financial workspace in a company brain. But the kind
of person that we built would."

"I've been doing this department by department as I've needed to. Three
weeks ago I built a head of HR — all of our company handbooks, policies,
every write-up, every one-on-one conversation, every team conflict — now
documented, organized, and filed into an HR department."

He clarified the altitude he runs it at: "The way I've set it up is level
two — I don't have Hermes doing it, it's inside Claude Code. I create the
schedule, I tell it: create yourself the skill. I just type slash-CFO and it
operates. I have different workspaces in Claude for Teams — I switch my
workspace between my businesses and my personal one, so all the skills for
that business are just there, and I have different shared drives for each
business. My team at one business only has access to their workspace — they
don't need to know anything about my other businesses, and I don't want that
AI getting mixed signals. They're each their own company brain, and it works
really seamlessly that way. With Claude Teams I also think about billing — I
only want each business to spend tokens on its own card."

### [01:21:04] One brain or several? — the dividing line
Adam raised his real case: "That was one of my struggles. I have three
different concepts, all under one company brain — one shared drive, three
folders. I wasn't sure if that was the right setup, or whether I really
needed three different company brains and three shared drives."

Austin's rule of thumb: "Here's how I think about it. If it's an aligned
industry with similar SOPs — where it would be good to have the visibility
of being an executive of a portfolio — then you want shared context. In my
example, I have a location in Austin and another location in Dallas. They
have different bookkeeping, they're different entities, different team
organizations, different payroll, different HR. But they run on the same
exact SOPs — the same branding, marketing, the same admin, the same calendar
cycles of everything. So I keep that in one brain. Your concepts have
different brand books — they're very different concepts," which pointed Adam
toward separate brains.

Chance offered the portfolio-owner variant: "I have the infrastructure for
all the different entities, but for my second brain, I've been thinking of
it less as a company brain and more as *my* second brain — I want it to help
my decision-making across all the businesses, like a CEO of the portfolio,
not a CEO of the individual businesses. Now I like the idea of a business
second brain for each one — but I'd still want that top-down one. Sometimes
I just want to talk to it about my time, and where I'm supposed to be best
allocating right now, jumping between the different businesses. I'm feeding
my management and my partners into there too, so I can best support them.
That thinking partner — like what chat used to be for me, but it knows so
much more context about me, how I actually run the businesses, the people I
manage, and then their agents."

And the line that stopped the room: "With your CFO agent — I just took my
CFO and got her her own OpenClaw. I didn't have to set it up. She just set
it up to already function the way she functions. It's an agent, so all my
stuff ships over and her agent takes care of it." Different people, different
build-outs: "A solopreneur doesn't need the same build-out as a small
business, or a mid-sized business, or enterprise-style businesses."

Mike made the same point through personas: "I've built different agents to
do different functions — I haven't used Hermes, I do it in Claude. But it's
not based on what I want. My wife — she's not a technology person, and she
thinks in personalities; she's a people person. So I built a named agent
called Sarah that is her project manager. She asks Sarah, 'where are we on
project X?' — because that's how she thinks in Slack. To her, it's almost an
individual." Chance: "That's how Taylor is — she calls hers Judy, and she
loves that Judy just takes care of it." Mike: "It's less about what the tool
can do, or the best way to do it — and more about who's it for, and how do
they want to use it."

Austin closed the loop on terminology: "Let's get aligned. A second brain is
your own — that's what you've done. An AI company brain is specific to the
organization; others use it. Both are valuable, and we use them in different
ways. This isn't Severance, where we separate work-life balance — but from
an architecture standpoint, it is separate folders." Oscar added the
owner's-eye consideration: "Are they different LLCs? Then think about it as
a business owner — this is an asset you're providing for the business. You
never know when you're going to exit one of them, and you might want to
package up that company brain with the assets you're handing over. So I'd
say both: a company brain on that company's own drive for each entity,
shared to yourself as the entrepreneur who owns all of them — and then a
second brain for you, that can manage all the entities and has access to all
the context from everyone."

Lauren added two practical notes from the trenches: in Google Workspace you
can add your other company domains as trusted domains, which makes sharing
between drives much easier. And on cross-entity billing: "One thing I'm
really struggling with at a client with different entities is the
commingling of finances — eventually, if you want to sell, or you're getting
financing, and it's fuzzy, that's going to impact it. So we're separating
billing — or you have the parent company invoice the other companies and get
reimbursed, so you have a cleaner paper trail. The question I'm always
asking is: what's the most simple way we can do this, so there's not a human
element of 'oh, I forgot to bill the property management company for Slack
this month, so now we're in arrears.'" The room's consensus: keep
subscriptions separate per entity where you can, so the company card stays
clean — and where something is truly cross-company, automate the bill-back.

Before moving on, Austin pointed everyone at the follow-up channel: Hermes
questions go in Slack or the group's learning platform, where Oscar is
notified on the keyword. "We're calling it the Agentic Academy — it's where
you go to learn. School is where we're going to focus all of our content,
and Slack stays as the group chat. I'm not going to close that Slack account
out."

### [01:31:14] Wins — Lauren Goldstein: the notes-to-self orchestrator
Lauren had arrived mid-session — she'd been waiting in the group's old
upstairs room ("We were upstairs going, where IS everybody? This
communication chasing is too much for my brain") — so Austin caught her up
on the new logistics: boardrooms on the first and third Fridays, downstairs
in the bigger room through December, with the mixer for new guests on first
Fridays.

Her opener: "One thing is clear — I've got to up my AI game, because I am
not doing as cool of stuff as all of you guys. There are actually two things
that were huge wins for me, both related to getting data and information and
synthesizing it."

**Win one — the task-audit helper.** "My client is on Microsoft, which is
the bane of my existence. Something I do when I come in as a fractional COO
is make everybody do a two-week time audit. And I realized — these aren't
Google links, so now I have to go look at this Excel sheet, and download
this, and put this here. I was like, I'm not going to do this — how can I
make an agent do it? So I have a little Task Audit Helper now: I put them
all in a Microsoft folder, and it downloaded all the things, uploaded them
into Google, and synthesized it. That probably saved me at least a day's
worth of work. I know that's not super sexy, but I was very happy."

**Win two — the notes-to-self orchestrator.** "I don't know if you guys have
notes to self everywhere — a Slack message to yourself, a text message to
yourself, a Google Calendar invite. When I'm falling asleep at night, I'm
like, 'hey Google, remind me to do this tomorrow.' And I just realized I had
hundreds of things — content ideas, things I needed my assistant to do,
things I needed to do. So now I have an agent that runs a couple of times a
day, and it pulls in my iMessage, my Apple Reminders, my Google Calendar
invites of notes-to-self, adds them to ClickUp, and triages each one — this
is a client thing, this is a content idea, this is an actual personal thing
you need to do. Then it goes back after about thirty-six hours — because I
clear stuff during the day — and says, oh, she cleared these, so we won't
add them; this one is still outstanding. And on Friday it says: hey, these
are all the things you added to your to-do list that you haven't done. It's
a really good reminder — oh, you're right, I forgot to email my attorney
about that."

The archaeology was the bonus: "Now I have a really good content repository,
because I went back three years, and it was like — you have seven hundred
and fifty content ideas. The next step is: which of these have I actually
posted, and how did they do, versus the ones I still need to do? So my
biggest win is my notes-to-self orchestrator. It's definitely not a cool
dashboard, but I'm like — this is great. And it was very easy to build. My
recovering perfectionist is why AI is good for me: you have to just get in
the water and get it working, to realize 'oh, I didn't think of that' or
'actually, we should change the timing on this' — instead of sitting there
thinking about how to perfect it."

**Chance — consolidating the PM tools.** Asked whether her team shares the
ClickUp (no — it's her self-management tool; the client team lives in
Monday), Chance described how he collapsed the same mess: "When I started
building my dashboard, I'd used ClickUp and Monday and all of them. Trello
was one I liked — the way it flowed — so I put Trello in as just a card; the
UI of Trello sat in my dashboard, because we already used it and I didn't
want to take my team away from it. Then I added to-dos on the front end of
my Today page, so I could handle all that agentically myself. And then
eventually Claude just rebuilt the whole Trello board my way. We ran it as a
mirror for a little while — anything done in one mirrored to the other — and
now the team is just using the dashboard, and we've eliminated Trello. It
can be a good way in: build a front-end page, have your ClickUp inside
there, and you have one login place."

Austin added the team's own version: "Roman just did a thing for us this
week. Our agents are all on Slack — that's where they communicate; they
listen to every transcription, all communications in Slack, email, et
cetera, and that's how they update the boards. Slack actually has a
board feature now, but it gets tucked away — so Roman had the AI build a
mirror of it on an actual dashboard, and they're linked: whenever one
updates, so does the other."

### [01:39:33] "Can my Claude agents live in two Slacks?" — and the Apps Script PSA
Lauren: "Speaking of Slack — has anybody solved this? My Claude agents are
all on my Slack, but it would be nice if I could also get them into another
Slack, and you can only have one connector in Claude."

Oscar: "In a complicated way, yes — but it's not as hard as it sounds. You
can make a bot in Slack — you actually go to the Slack developer page and
make a bot. And that bot can use whatever you want as the brain — obviously
I have mine on Hermes. That bot can be installed in different workspaces, if
the admin of that workspace allows it. The agent kind of walks you through
it. And maybe look at Composio — because it lets you do multiple accounts,
you might be able to do multiple Slacks in there and just connect the
Composio MCP." Austin: "We set up our Slack agents the same way — we built
our own custom Slack bot and gave it the MCP access. Talk to Roman as a
follow-up and we can walk through how to set up a Slack bot." Mike: "My
developers are Slack bots I built myself — and Claude will actually walk you
through the whole thing: go here, click this, hit save, copy this code, put
it in this thing."

Lauren then delivered a public-service announcement: "If you guys aren't
using Google Apps Script, here's your PSA to do that. I'm using it in a lot
of different ways to run cron jobs. A very funny example: one of the team
members was spending a lot of time on — celebrating birthdays in Slack, and
work anniversaries in Slack. There are apps for that, but we didn't need
ninety percent of what they do. So we have a roster, and I created a Google
Script that reads it every day. On the first Monday of the month it says,
here's who we're celebrating this month — and on the day of, it just posts
to the Water Cooler channel: Happy Birthday, whoever. She was like, 'I saw
you took that away from me.' And I was like, 'I didn't take it away from you
— I need you to go do other stuff that's really important.' That was a
ten-minute build with a Google Script."

Darby seconded it: "I tied one into my daily brief — scanning my Gmail,
helping with tagging, flagging certain things. Have Claude walk you through
any sort of setup. It removed the need for anything I would have needed
Zapier for." Lauren's other Apps Script build: "We use Airtable as our CRM,
and I created a custom pop-out on the Gmail sidebar, so I don't have to
leave my inbox to add somebody to the CRM — and every day it logs our email
interactions back and forth, so it has a log of when you last communicated."

### [01:45:58] Full dashboard walkthrough — Chance Forman (screen share)
Austin set the frame: "As a general theme we're talking about marketing, and
what I've heard from this community is that social content is a huge problem
we're all trying to solve at the same time — capture, ideation, research,
posting, team collaboration, handoff, distribution on all channels at once."
Chance joined the Zoom room and shared his screen.

**The Today page.** "This is my dashboard — it pulls in all my companies
every day. Anything I'm working on is based out of a project. The Today page
gives me what's active — what I need to work on right now. The stuff at the
top that says 'needs your attention' — that's blocked by you. Inside my
project pipeline, that means AI has been working agentically in there, and
it needs the human interaction to get it across the next step." He clicked
into one: "It takes me into that exact card and tells me what's blocked by
me. There's nothing I can do here — I'm waiting on a website to go live from
my developer, so it's just in a waiting period."

**Coach, on every card.** "Inside this project card, I can click on Coach —
and now Coach has all the context of this card specifically. I don't have to
go feed it to it. It knows this project; I don't have to say 'hey, we're
talking about this.' If I type at the bottom, I'm working with Claude — and
I can click between and change the models there." He tapped the microphone
icon and talked to it live — telling it about a meeting he had lined up
right after the session to review a website draft — and it acknowledged and
logged the note. "When I talk, this is Codex now. When I'm done with the AI
voice, it still drops it into the transcription there — so Codex and Claude
are working in the same spot."

**Four numbers per company.** "I can drop into my companies. The big things
I track in a trades company: how many leads, my booking rate, my closing
rate, and my average ticket. By those four numbers, I can run an entire
business without knowing any other numbers. Those are my four metrics — each
with a target, and right now we're running a new campaign on a lower ticket
to get our foot in the door with some clients. It pulls in all my
QuickBooks, all my APIs, so I can run real numbers."

**The master project template.** "Going into projects — this is where the
real grunt of the thing was built. This file is pinned to the top as the
master project template, for a reason: all of the AIs, no matter what —
harness or direct LLM — will not progress any project forward out of the
idea stage unless these questions are answered. And I can be driving down
the road, open up Coach on a brand-new project with not a lot of context,
and just say: 'ask me one question at a time to get the master project
template filled out.' We go back and forth while I'm driving. Now it has all
the information — the definition of done, the goals, who's involved, risk,
success metrics, approval gates — and it can move itself down the stages.
It'll create checklists for itself: a human checklist and an AI checklist.
Nothing goes into 'blocked by human' until something actually becomes
blocked — and that's what goes on the front end of the Today page."

**Team members with their own agents.** "An idea comes in, and each stage
gets it better — strategic, deeper, priority — and when it hits In Progress,
it's an active project. I can look at just mine, or see what other people in
the company are actively working on. I'm in Taylor's now — she has her
dashboard agent that helps her push things through, tied directly to Codex.
And she has her CFO agent, on OpenClaw, that gets in here and does the
financials for all the companies in the portfolio. It moves itself to
waiting or blocked; once I solve that, it goes to review, and if I approve
the review, it moves to complete."

**Phone-first by design.** "Coach is literally just like using Claude or
Codex or ChatGPT — but on my phone, when I open the dashboard, it opens
directly to Coach automatically. On my computer it opens to the Today page,
because I'm here to work. When I'm on the move, I want to be able to just
talk, and it'll take action inside the dashboard, whatever I tell it to do."

**Journal, Ideas, and the nightly sweep.** "Journal I just built — if
something cool happened, I can drop in entries: it was a really good day, it
was a really bad day, and I can offload into here. Anything I enter, or
anything my team enters, gets swept up into Obsidian at eleven fifty-nine
tonight — documentation into the second brain. Ideas is for whenever the
team or I are out and about and something is broken, or I wish it could do
something else — we jump in and rate it on impact and effort, and that's
what we work on first. Codex actually watches the Ideas section, and
anything it can take care of itself, it fixes. If something's broken, it
tags the Today page in red at the very top — 'this is broken' — and I can
just put 'retry' and it'll try to fix itself. If it can't, it tells me what
I need to fix."

**Connectors — Composio under the hood.** "Connectors is where I took
Composio and linked it into the back side. When my team comes in, they can
connect ElevenLabs, OpenAI, any of their things, right inside there. You can
see Nick's in here with his agents, and Taylor's in here — those are the
only two employees I've brought in so far; we're on a sprint right now to
bring everyone in. For the trades, if I took this enterprise, it'd connect
to the field-service platforms — Jobber, Housecall Pro, ServiceTitan — and
I'm trying to connect the AI voice-agent services too. Making it easy for
the team to come in, click connect, log in themselves — they're not going to
be as technical."

**The content pipeline, end to end.** "You get ideas scrolling on Instagram
— someone else has done something cool, and I want to put my own spin on it.
I just take the link and text it to my OpenClaw agent, Lily, and Lily drops
the files in. Codex is my creative agent inside this pipeline — it takes
that reel and says: here's a brief, here's three viral hooks. You could
record these hooks individually and test them. Here's your script. If I
approve it, that becomes the script and we move forward into strategy and
actually recording — it gives me a shot list, talking points, a caption. And
then, because I'm trying to build all of this so I can be out in the field
and everything keeps moving: I just click record, and now it's a
teleprompter. I read the script really quick, and it drops the audio file in
there for either AI or Nick to lay over the B-roll."

He's testing an AI video model for the finish work, connected over MCP to
his generative-asset library so it can pull extra B-roll as filler, and
working on the judgment problem with Nick: "You give it the context of the
video, and that's the hardest thing — if I give you a ten-second clip, but
only three seconds of it is what I'm trying to take out of there, can it
tell what *we* can tell is a good clip? The right framing, the impact. We're
working on the back-end prompt to get the sub-clips clipped better. And I
have Coach inside here too — if I didn't like it, I can click on Coach, it
has all the context of this video, and I can work with it on fleshing the
idea out before pushing it forward."

**The TikTok-first testing strategy.** Asked what he meant about platforms
punishing you: "On Instagram, if you put the same reel out, it doesn't want
the same style of content — so a lot of people on social will put the same
reel with minor changes on TikTok, do ten of them, and whatever the best
performer is, is what they pull over to Instagram. Usually they're just
changing the hook at the front end. Even though it's a small tweak,
Instagram picks up on that more than TikTok — TikTok, you could post the
same video over and over and they don't care. Blast it over there, then move
what works to Instagram, run a few things on trial reels, and whatever works
there, do more of it public-facing." Justin added the ads parallel: "A good
framework when you're beginning ads is to test the hooks. But Meta is
cracking down — if the body and the CTA of your ads are the same and you're
just testing hooks, it treats it like duplicate content. It'll punish you
for optimizing. They want unique content, not a testing platform — whereas
TikTok, blast away, then move over what works."

Oscar mentioned a brand-new model from Meta that's reportedly strong at
image and video recognition — worth A/B testing as the "eyes" reviewing your
clips. Austin suggested one more capture path: "Rather than sending a link,
you could have your agent scrape your Instagram saved folder — whenever you
save, it goes through." Chance: "Stop it — that's why I used to save so
much. You can do it through Composio, because Composio has Instagram with
full access — that's what it's doing on the back end anyway." Austin
described the trending version of the workflow: saved posts land in a
database table with the script and the hook extracted, the AI writes your
own version of each, "and now that becomes — I'm turning saved ideas into
personalized content that I can produce," with statuses moving from review
to ready-to-record to ready-to-produce.

**The physical workflow.** Austin pushed for the concrete mechanics: "Walk
me through physically — what are you actually doing?" Chance: "Right now I'm
documenting me launching this new company. Every day, as I'm working, I'm
just getting some B-roll of everything I'm doing. At the end of the day I
put it into Capture — batch upload from my photo library, super fast — and I
give a summary of what the day consisted of. It writes the script with those
videos and my summary, and then I just go record the video. And then Nick is
taking that and turning it into a finished product — and I have it at the
end of the day." He created a new card live — day five of the launch — and
batch-loaded the day's clips into it while the room watched, then showed the
two finishing routes: "I can give it to AI to make me a video, or I can give
it to my video editor, or I can give it to AI to get me to eighty percent
and then Nick to — we call it — polish it. They upload the cut right there,
I review or suggest edits, and it feeds back to either AI or Nick."

Lauren named what she loved about it: "I was having a conversation with a
friend about this — now I really don't go into the user interface; I'm
chatting through someone. I think that's the unlock for a lot of people with
AI. Not necessarily going into Claude or chat, but — how do you talk to the
interface and let it go do all the stuff, instead of you typing it out?
'Hey, good morning, Addison' — that's my AI assistant — 'I need your help
with this,' and then it goes and does it. People who can do that are going
to move so much faster with AI. Having some interface on top of all the
complexity." Oscar: "Which could be Hermes, right? Knowing how to
communicate what you need from AI is such a big part."

**Under the hood.** Oscar asked the infrastructure questions: where's it
hosted, what's the database? Chance: Supabase on the back end, handling the
OAuth and the per-user sign-ins. "The sign-in for the team — they don't have
to do any of that. When they sign in on the front end, it all activates for
them. What I'm trying to get to now is one click to set up their Hermes
agent, so they don't have the complexities and the fear — 'I don't know how
to set up an agent, I'm busy working, I do construction, what do you want?'"

**The one-house philosophy.** "When you get into the trades — and you start
stacking more software and more apps onto their plate — less gets done. They
start hating you: 'I hate all this, dude, I work with my hands for a
reason.' The big thing for me, why this keeps evolving, is I want things to
be in one house. One app. We already have to-dos on the front end — we can
give each other to-dos and check them off on the Today page. If I want to
tell Taylor to do something, I just put it in here and she gets it done — I
don't have to text her or call her, and I can see it got done because she
checks it off. And people can give tasks back to me the same way."

### [02:09:03] Invisible AI — the call-coaching reveal
Oscar picked up the thread: "There's an element of this — your whole team
doesn't need AI. Mike, you were mentioning this: a lot of times you can do
it for your team where they don't even know it's AI. A background automation
where something gets pulled, gets processed by AI, and lands back in front
of them — and they don't even know what happened."

Chance's confession got the biggest laugh of the morning: "My team literally
thought that I was listening to the calls every day — that I was taking
their three worst calls and their three best calls, auditing them, and
giving them feedback. Because it was coming from my email. It goes through
GoHighLevel every morning, and then it sends a summary to each one of my
CSRs: their three worst calls and what they could have done better, their
three best calls and how to go deeper into those things. And they thought it
was me." When he finally told them: "They were like, 'oh man, it's AI.' But
I'm like — yeah, but I'm watching and listening to everything AI is telling
you. There's definitely oversight there."

Oscar generalized it into the design question of the morning: "Techs might
take a picture of the nameplate of an HVAC unit — you don't want them
dealing with 'okay, let me give it to my agent so it can figure out which
part I need.' That can all be done by the system in the background. As
agentic CEOs, we're also deciding: at what point am I making my team worse
by adding all these extra layers of complexity — and when can I make it so I
just control the whole system, and it's doing it for them, enhancing them
without them having to do it manually?"

Mike had the cleanest example: "For us it's inventory management — we have
thousands of samples of different things, and they used to have to go
through and fill out the spreadsheet. Because I didn't want them to know it
was AI, I just put a front end on ChatGPT's voice feature. They literally
just launch it and say, 'we've got seven of these, we've got ten of these.'
They have no idea it fills out a spreadsheet — they just talk to it. 'Okay,
I got rid of five of these' — and it does all the math for them. Now
inventory is a twenty-minute exercise." It started as a spreadsheet, "like I
think everything does," and the voice layer just fills the same spreadsheet
the rest of the system already reads. Chance, who uses a dedicated inventory
app, immediately wanted the same trick: "I was wondering if I could do the
same thing, where I could audio-talk to it. That would be pretty cool."

Mike's second one: "I built another one for Christina, with a whole website.
It's less about updating quantities — it's when a new line of stock comes
out. She physically wants to be the one to say 'this is how I want this
presented,' because of brand and marketing — she won't let that go. So I get
out of the way and just made it easy for her. It's a Slack bot: she goes, 'I
want to add column five, and I want to put this picture in there.' She has
no idea it rebuilds a whole inventory system — it just sends her a link and
says, 'look at this, is that right?' And it's learned her vocabulary,
because she doesn't say 'approve, deploy the system.' She'll just say 'that
looks amazing' or 'this sucks' — and it's learned her language."

### [02:12:33] Justin's parting tip — jobs into SEO pages, automatically
"One quick note for both of you guys on the content side — something we're
doing for our clients that helps SEO and social," Justin offered. "With
Housecall Pro or ServiceTitan, when you have a job that's completed, upload
as much content as possible — before-and-after images, any relative content,
potentially a review. When a job is marked done, have it automatically fill
into a template page and post to your website. Now you have automated
content going into local SEO, with keywords based off of real things you're
doing. Then you can take that and formulate social media content from it as
well — blog post to social, with real jobs that are done. That'll help scale
SEO and social together."

Chance runs a cousin of the same play already: "Similar but different — the
guys upload before-and-afters into Housecall Pro, but they also send them to
a text message thread. Nick's little AI bot pulls them from the text thread,
edits them to Nick's specifications, and sends them to the dev team to add
to the website — before and after, meta-tagged with location and all that."
Justin: "You can skip all of that and have it done with one click."

### [02:14:00] Content distribution, part two — Darby Rollins
Darby was up next to demo the distribution dashboard live — but his laptop
had other plans. "My Mac is just overloaded, I think. My Zoom is crapping
out. I can talk through it a little bit, but I'm not going to be able to
really show much — I'm putting that video together next week anyway." He
talked through the highlights: the pricing is genuinely affordable for
scaling it up, inbox and comment management live inside it, rescheduling and
flipping posts around is one click — and he committed to recording a fuller
walkthrough video for the group the following week.

Meanwhile lunch had arrived for the afternoon mixer — "tacos, chicken or
steak, coffee, water, sodas, all out there" — with a few dozen new guests
expected within the hour. "You're also not required to stay — I get it, it's
a long day. I think it's fun to have a mix of members in the room. Y'all are
the success of this."

### [02:16:25] The October hackathon — Austin Distel
One announcement for every member's calendar: a two-day, hands-on "AI
Advisory" hackathon, October 15th and 16th. "We start in the morning on a
Thursday, we end on a Friday afternoon, and everybody here is invited — it's
part of what you get access to. You'll get a menu of items to install in
those two days, and the definition of done is clear. We're still exploring
what the two to three AI agents you'll build in two days will be — I'm going
to source ideas. One of those: a project manager. One: a chief of staff.
Maybe agentic SEO. Maybe set up your Hermes. All in alignment with the
Academy. Just get two weeks' worth of work — maybe even two months' worth of
work — done in two days." Calendar holds were requested on the spot, and
granted.

### [02:22:22] Close — Austin Distel
"Okay guys, we're going to wrap up this section here. For the next
forty-five minutes: socializing, networking, having lunch. And then part of
the mastermind, which is kind of fun — we do a voting game where the best AI
outcome wins. You guys will get to see that. What's cool is it surfaces what
the best builders in Austin are doing, and gives an award."

One last exchange before the room broke: Chance mentioned he'd wired
Austin's prioritization skill — the one shared with the group — into his
content inbox: "I plugged in the skill you shared with us. I can dump a
bunch of ideas in there, and it takes the pain and impact off of all the
data and says which piece of content I should push forward with first."
Austin summed up the method behind it for the room: "It's a problem-solving
technique for what you should solve first. You want to solve the most
painful problem that happens the most frequently — always start with the
highest pain, then work your way back on frequency."

And with that, the official boardroom wrapped, and the room went to lunch.

---

*Attendee-only transcript. Please keep it within the Agentic Society.*