Marketing Against The Grain · 2026-07-01 · 39 min
Key moments - from our scoring
Substance score
40 / 100
Five dimensions, 20 points each
This episode demonstrates how to build a fully autonomous AI employee capable of executing end-to-end workflows rather than isolated tasks. Nick walks through a concrete use case: re-engaging past customers and leads through personalized email sequences - typically the highest-leverage activity in any business. The build uses Hermes (positioned as more reliable than OpenClaw) as the agent framework, paired with either GLM 5.2 (a cost-effective open-source model from Deepseek) or Claude Opus depending on speed requirements. Composio handles integration with all existing tools (HubSpot, email, etc.) through a single connector, while Orgo provides the virtual computer infrastructure that allows agents to actually operate browsers and manipulate files. The episode explains the shift from explicit step-by-step workflows toward "loops" - where you define an outcome and let the model navigate toward it - and how this skill will eventually become less necessary as models improve. The conversation also touches on Obsidian vaults as knowledge systems for agents, allowing them to access and reference institutional intelligence.
Hermes is more reliable in production environments with fewer infrastructure failures like gateway downtime and connector issues, while OpenClaw (now owned by OpenAI) had reliability problems that made maintaining AI employees difficult.
GLM 5.2 costs approximately $0.98 per 1M tokens compared to Claude Opus at around $5 per 1M tokens, making it roughly 80% cheaper while performing equivalently in most knowledge work use cases.
No - Composio lets you connect all your tools (HubSpot, email, databases, etc.) once, then install a single Composio connector into any new agent to give it access to all those integrations automatically.
Re-engaging past customers and leads through automated email sequences using an AI agent, which typically generates the fastest ROI because it targets warm audiences that have already purchased before.
Orgo provides virtual computer infrastructure that agents live in, allowing them to actually control browsers, manipulate files, and do real autonomous work - not just call APIs - with a 3-day free trial available.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode delivers a handful of genuinely useful, specific tool recommendations (Composio as a universal connector, GLM 5.2 as a cheap Opus alternative, Agent Mail for agent inboxes, Obsidian vaults as agent knowledge bases) but the density is badly diluted by tutorial narration, mutual enthusiasm, a long host tangent about his own Obsidian setup, and repeated promotional segments for Orgo. The actionable insight-per-minute rate is low for a 39-minute runtime.
GLM 5.2. It's built by a company called, uh, Zai or Zai. Uh, it's an open source model and it's like a deep seq moment because, uh, again, because it's an open source, it is a Chinese model, but it's a fraction of the cost of something like Cloud Opus and it even outperforms it in some use cases
Give it a computer on Orgo and lastly give it an Obsidian vault. Give it a knowledge base and a brain so it has an even deeper understanding of what your company does, who you are, who are these people you work with
The episode follows the exact AI-agent-zeitgeist playbook of 2025 with no contrarian or first-principles thinking; the 'army of AI employees' framing is ubiquitous across social media. The only flicker of genuine originality is a brief mention of goal-based 'loops' as superior to scripted step sequences, but this is immediately dropped without development.
rather than describing a sequence of steps for an agent to take, just describe a goal and give it the tools and context to nudge it towards that goal. And it does it and it does it often better system wise than what I might design.
There's only a few forms of leverage in a business. There's code, there's media, there's, you know, talent. Today I think agents is like the combination of all of that.
Nick is a genuine practitioner and co-founder of a real agent-infrastructure product, which gives him credibility above a pure thought-leader; however, the episode is primarily a product demo for Orgo rather than a distillation of hard-won operational lessons, and claims of scale are asserted rather than evidenced.
I had the first viral video on OpenClaw on all platforms. It did over like a million views on every platform. And I was talking about the Mac mini
Peter Steinberger was at, uh, Orgo. He was our first paid customer.
The episode offers a handful of concrete specifics - tool URLs, a direct price comparison between GLM 5.2 and Anthropic, and a Day 0/3/6 email cadence - but there is zero outcome data: no open rates, reply rates, pipeline generated, or customer case studies with numbers to validate any of the workflow's effectiveness.
98 cents. As opposed to, uh, what's anthropics? It's, uh, $5
Day zero, day three, day six
The host asks several genuinely sharp conceptual questions - particularly about workflow vs. full-autonomy, the skill-vs-simplification trade-off, and model fine-tuning rationale - but never pushes back on any promotional claims, never questions the '5 minutes to set up' assertion, and allows a long self-indulgent tangent about his own Obsidian system to consume several minutes of interview time.
should you think about it as an end to end workflow or like a real autonomous employee that can do the role of a marketer or the role of a salesperson or the role of whatever it may be
do you think that this is a skill that all kind of marketers and growth people need to learn, or do you think over time it's going to just become way more simplistic
Computed from the transcript - who did the talking, and the words that came up most.
Build Your First AI Employee (Free Workflow + Prompt): Ep. 433 What if you could have an AI employee working right alongside you, automating key business workflows? Kieran and guest Nick Vasilescu (Co-founder of Orgo) dive into how agents are the next big form of business leverage. Learn more on what it actually takes to get a functional AI agent up and running, how new connector tools like Composio make integrations simple, and why giving your agent a second brain with an Obsidian Vault is a game-changer for business productivity. Mentions Nick Vasilescu Orgo Hermes Agent Composio OpenClaw GLM 5.2 Obsidian Get our guide to build your own Custom GPT: Resource [Free] Steal our favorite AI Prompts featured on the show! Grab them here: We’re on Social Media!
Transcribed and scored by The B2B Podcast Index.
Speaker A: Okay, I'm coming to you with one of the best episodes we have ever done. We did this episode with Nick Gillescu. He just blew our mind on how easy it is to create your very first AI employee. We created one that was able to do a bunch of prospecting and create pipeline for your sales team. And the best thing is it's doing it all the time. You can message it, you can text it, you can talk to it like an employee. You can create an army of these AI employees in mere minutes. I've uh, replicated Nick's system and started to use all of the things that he shows on this very episode. And it is incredibly powerful. All of that and more on this episode of Marketing against the Grain. Okay, welcome Nick, to the show. Nick, you are the co founder and head of growth at Orgo. It is a cool product. Actually, uh, we were talking off Mike that will give your agents a full computer in the cloud and you are going to like take us through how to build your first AI employee. And so there's been so much talk online about AI being able to do fully autonomous work and AI being able to do full end to end roles. And I still think there's not been a lot of great examples of that. Right. I think there's been examples where it will do like a kind of small task, not an end to end kind of job or a role. And I think you're going to show us how to do that through a system called Hermes. And so we're excited to get started and kind of go through what you're going to build.
Speaker B: Yeah. Today we're going to walk through implementing your first AI employee for your business. Everyone talks about things like OpenClaw, Hermes, and they talk about all of the things that they're automating with it. But I'm going to give you like a concrete first AI employee that any business can benefit from. I'm just going to walk through the whole build out process. It's super fast, super simple to get started. Six months ago it took a couple hours to get set up. Now it's like insanely quick. So yeah, we can just dive right in.
Speaker A: We should maybe make this kind of really obvious for folks. But if I was listening to this and I was thinking, wow, you're going to create a first AI employee, I would think of it as, okay, well you can do everything, like every single task I give you, no matter what it is, you're going to do that task. But I think in this case, should you think about it as an end to end workflow or like a real autonomous employee that can do the role of a marketer or the role of a salesperson or the role of whatever it may be.
Speaker B: It can literally be all of these things. I know that's a little bit of annoying answer, but it's true that, you know, starting with a workflow is how I like to start because it's a, it's a single use case. Makes things clearer, simpler, less failure points. You get one nailed use case. Start there. But then after you have that, it's also just a very competent. Yeah. Employee that can do other things too adjacent to that task that it's already really good at or other tasks. And then finally yeah, just being able to just message it just like you would a normal employee of like, hey, like what's the progress on this? What's the update on that email sequence we sent out last week? Did we get any feedback from that? It can truly be an employee and I think it's the biggest arbitrage for businesses.
Speaker A: That's awesome, like great advice. So we'll get into how that looks. But start with one workflow and then over time it really just does feel like you have an employee sitting in your slack, sitting on your messaging imessage, whatever it may be that can do work and help your business grow.
Speaker B: Exactly. There's only a few forms of leverage in a business. There's code, there's media, there's, you know, talent. Today I think agents is like the combination of all of that. You know, to be able to have 10 AI agents, 10 AI employees, I mean, it's insane.
Speaker A: Okay, so you've got a pretty cool graphic up. Can you maybe show us how we get started here to create this awesome AI employee?
Speaker B: So the first step is with the use case. So with our AI employee that we're going to build today. Every business, they have thousands of customers, hundreds, tens of thousands, hundreds of thousands maybe. And they maybe haven't engaged on these leads or customers in a while. Maybe it's been a few months or years even. And you're just sitting on potential revenue that you could just generate overnight. If you re engage these leads. This is a warm audience. You already sold to them before. So it's like the highest lever to pull in any business usually. So what I want to build today with you is an AI employee that engages these past customers. Past leads, makes an email sequence, warms them up, crafts an offer to get them re engaged on your service or product. And this is something that you could implement Today and by tomorrow already be making money from it. It's very fast and easy to get started.
Speaker A: I think this is an obvious workflow that all business marketeers would want to actually have within their motion.
Speaker B: We want to give it a couple things, so we want to give it its own email. I use this company called Agent Mail. I've actually met with them, the founders, in person. They're great guys. What Agent Mail does is you can give an email inbox to an agent, uh, in this case our AI employee. So I use agentmail, I use Hermes. Hermes is our harness. And so a lot of people, they're familiar with openclaw, they maybe have heard or haven't heard of Hermes. Hermes is a competitor to OpenClaw and I find that it's a more reliable version. It's better. A lot of people today are switching to Hermes because it's more purpose built for kind of like the sort of use case of, uh, the scale at which people are using agents today. Lastly, we have the model that powers the harness. So the most popular one today is actually this model called GLM 5.2. It's built by a company called, uh, Zai or Zai. Uh, it's an open source model and it's like a deep seq moment because, uh, again, because it's an open source, it is a Chinese model, but it's a fraction of the cost of something like Cloud Opus and it even outperforms it in some use cases, which is just insane. It used to be the case that you'd have to spend thousands of dollars a month to really get a lot of great use cases out of something like Openclaw. But now with GLM 5.2, it's super affordable, everyone can get started. Um, and it works really, really well.
Speaker A: If you're enjoying this episode and you want all of the resources, you can click the link in the description or scan the QR code to get them all delivered to your inbox. Just a quick question. On that model, is it fine tuned for like knowledge work or there's some reason that it's better at this type of work versus the core LLMs or it's just a great model?
Speaker B: I think it's a kind of a mixture of things. I have my own speculation. I don't know if they did something super fast to like train their models on the outputs of Fable 5 or something and then they just like caught up super quick. I don't know. It's an open source, you can open weight, uh, it's affordable, it's fast. And yeah, I think more and more labs like OpenAI, they even said explicitly on when they made GPT 5.5, they kind of made it with this idea of OpenClaw in mind. So this is definitely the trend is that we're trying to make LLMs that are better at using tools.
Speaker A: If I'm watching this and we're doing this through Hermes, it says that this is all wired through Composio. So do I need to sign up for all of these and I'm going to have to wire them all together or is that done automatically through. It's like really easy through Hermes to do that.
Speaker B: Yeah. So Composio is really one of my favorite tools. So you just go to composio.dev. and it allows you to connect to all your favorite platforms, tools, applications, even HubSpot. And in one click, you just click authorize and it connect, connect, connect, connect. And once you connect to everything, you can then install just one connector, which is a Composio connector, into your Hermes agent.
Speaker A: That's cool.
Speaker B: And it has access to all the tools that you connected to with just one connector.
Speaker A: Okay, now, uh, you're giving me a ton of tools to go use. I need that tool. Um, so it basically allows you to have all of your connectors in one spot and then you have one link and you can connect that link to any AI assistant or tool and it will auto have access to all of those tools that you've signed up for.
Speaker B: That's awesome.
Speaker A: That's so cool.
Speaker B: And how many times have you, you know, switched to a new, uh, okay, I'm going to try Perplexity Computer. I'm going to try Cloud Desktop and you have to connect everything again. Well, with Composio, you don't have to do that. Anytime you set up a new agent, a new platform, a new tool, I mean, you just connect one connector Composio and you have access to all the same tools that you have on your other agents.
Speaker A: Very cool. That is a. I will be signing up for that today.
Speaker B: It's. It is awesome. It is, it is truly. And we're working with them closely to more integrate into and obviously into our product or go. And I can talk just real quick. Look, every AI employee, every AI, uh, agent, it needs its own computer. At Orgo, we provide our own computer. You can use others as well. Like there's companies like Hostinger, Herzner or DigitalOcean. You just need a place for your agent to live and an environment for it to store its files. And be able to do real work. Uh, and that's what we do.
Speaker A: Yeah. So Oracle, your company, allows the agent to actually have a virtual computer, so it can actually go do things on the browser. It can go sort files just to make sure our listeners really get that. So now you have agents who actually, just like you have a laptop, they can actually start to do work on their own little laptop. And so that's what your company allows these agents to do. And that is like the power of starting to set up these, like, uh, autonomous agents, these kind of AI employees. And the cool thing is, like, I spill water all over my laptop. Right. And I had to go get a replacement and they couldn't do work for like several hours. So the agents don't have any of those problems. They don't have to go ask it to, like, deploy their virtual computer or they don't have to, like, get a new one when they actually spill water over it. So even just that is a pretty powerful concept for people to understand.
Speaker B: Yeah.
Speaker C: When a buyer asks AI for a solution like yours, do you know if you're showing up or not? And HubSpot, we've got a great product that actually helps you get discovered in AI search. HubSpot AEO helps you show up in these moments with the right answers buyers are looking for before the first click, before the first fill. That's the moment HubSpot AEO is built for. Go check out HubSpot AEO, you can trial it for free, and it's going to help you understand how you're showing up in AI Search and how you can show up even more.
Speaker B: And in that time, uh, just now, in the last like five seconds, I spun up a whole computer. I just asked it to go do some research on HubSpot. And you can see, like in our playground here, our Orgo agent, or kind of own version of Manus, so to say. It's showing you what it looks like to have a. AI control a computer. Click around like, that's it, we're here. Agents can now operate a computer just like us. And so, yeah, uh, I'll be able to kind of walk through how to get everything set up on Orgo once. Uh, again, you can use whatever tools you'd like, but I'm a little biased. I think we have the best easy.
Speaker A: Uh, you should feel proud about your tool. Uh, give everyone free time to, like, deploy something. So you should feel free to, like, talk about how great your company is for sure.
Speaker B: And I want everyone to give it a shot. We have three day free trial, uh, 20% off your first three months if you use code HubSpot. We'll also have a link in the show notes. Um, and yeah, it's no risk to get started. You can try it out and give it a shot. So as far as getting set up with something with our AI employee, should we just dive right in?
Speaker A: Yeah, uh, yeah, let's do that.
Speaker B: So you can see here I have this workspace in Orgo. I created a HubSpot workspace, added a logo and I have all of my workspaces here, uh, that I manage our customers like businesses. I manage their AI agents inside of these workspaces and I could just easily spin up a computer and install an agent inside of that computer and get going. Now we launched a new feature and I'm really excited to show this and it's called Templates. Um, so it used to be the case that you'd have to create a computer, go inside of it, manually install like Hermes or openclaw. Well now with Templates you don't even have to think about that. You just click templates. Click what you want. I want a cloud code template, an openclaw template, a Hermes agent template. Let's do the Hermes Agent 1 and then I'll call this employee. Uh, uh, let's give it a name. Should we give it a name? What's a good name? We should give our.
Speaker A: Give it a Kip. We need to get. Kip isn't here, he's in cans and so he needs to start doing some work here.
Speaker B: Okay, Tip Agent. So type in the name. I click Launch computer and you can see off to the left here that it's creating a computer with Hermes pre installed and everything. And it takes a little longer since it's installing all these packages and so forth, but it takes about 20 seconds.
Speaker A: And so Openclaw was the kind of autonomous agent harness of choice for a little bit of time. They got bought by OpenAI. It was a great deal for Peter. I think they're still doing things. They just haven't got as much of the kind of zeitgeist online. Could you maybe just break it down? Why do you like Hermes? What are some of the things they do better for the full autonomous AI employee kind of task?
Speaker B: Yeah, I think that Hermes by default it just for whatever reason the team behind it, it's much more reliable. I spent days at a time fixing some of the issues I had with open cloud gateways going down, things breaking cron jobs, connectors just failing. There's a lot of, lot of chaos. I switched over to Hermes. I really haven't had any issues on Hermes with reliability, and that's the biggest thing. And if you're going to have AI employees, the minute they break, you kind of lose trust. Exactly. And look, I have nothing bad to say about. I think OpenCL, maybe it's much better now and I should give it a fair shot again. I have to give Peter his credit. I had the first viral video on OpenClaw on all platforms. It did over like a million views on every platform. And I was talking about the Mac mini and it was cool. And Peter can like repost it put us on the website and everything. And he was. Peter Steinberger was at, uh, Orgo. He was our first paid customer. Crazy.
Speaker A: Very cool.
Speaker B: So I love the guy, but, um, but it just happens to be the case. Herbies is a little better. Yeah.
Speaker A: And these tools, are you kind of interchangeable at times? Right? Like maybe they built some features that you love and you kind of move over. All right, so Hermes, you basically set up a computer through Orgo. You give it a computer, then you give it the tool Hermes. So the agent has access to that, I guess, like the computer has that tool pre installed.
Speaker B: Exactly, yeah. So now the computer has Hermes as like the harness. It has everything it needs for that pre installed since we launched it off the template. And so you can see here like it spun up. The computer even has a little background, you see. Oh, okay, Hermes agent. And here there's a terminal that's beneath this computer in our interface here. And now I just type in Hermes model. And I'm going to walk through this as though no one's ever done this before. Just because it's easy, it's fast, and, uh, it's simple enough that I can do it live. And once I type Hermes model, it's going to take a second here and it's going to spin up all of the different available models that you can connect to for the agent. GLM 5.2 is on that list. Claude is on that list. GPT 5.5 is on that list. Um, so we'll just give it a second here to spin up.
Speaker A: Well, while we're giving that a second to spin up, one thing. If you look at the graphic you created, uh, to kind of give the overview here, you know what's interesting is I'm interested to get your take because you're obviously an extremely technical person. You have a technical company, you think in systems, see this, how it works. Can you scroll down to show the full flow. Okay, this, this to you is probably just like, hey, this is just how I think, um, the average person actually doesn't think in terms of workflows and systems. And so what you're showing right now I think is a great use case to follow this video and try it out and build your first AI employee. But do you think that this is a skill that all kind of marketers and growth people need to learn, or do you think over time it's going to just become way more simplistic, that you're going to just tell an agent to say, I want you to like improve my paid media and that agent will spin out little AI employees who can in the background. It's like figuring out what they should be doing. I guess what I'm asking is not everyone has a skill set. And so do you think over time people need to learn this skill set of understanding how to build agents to do specific workflows and think in terms of like workflows and systems, or do you think the systems are going to get so much better that you'll just have to give them an outcome and it will build all these things for you?
Speaker B: I think it is both, I really do. Um, because when it comes to thinking in terms of systems and having an AI agent that can automate a workflow, like for some people, like you said, it's more, it's more natural. For others it's more of a conscious effort of like, what do I do when I launch a paid ad campaign? Like, okay, let me walk through step by step. And if you can just kind of like be conscious of it and like write down what does it look like step by step. Actually on, um, my formal background I was a pre med, I studied biology in college. But I, uh, think it helped me in the sense of I definitely learned everything in school in terms of like systems and step by step. So that is useful. Now in the current state of AI today, you're probably seeing a lot of stuff about loops. Have you heard of like loops?
Speaker A: Yeah, yeah, yeah, yeah, yeah. That is all my reading list right now.
Speaker B: Yeah, it's fascinating because rather than describing a sequence of steps for an agent to take, just describe a goal and give it the tools and context to nudge it towards that goal. And it does it and it does it often better system wise than what I might design.
Speaker A: Yeah, I have a whole bunch of things to read on loops, but you give it an outcome and you tell it to loop until you finish that outcome and it's able to start and you give it a bunch of tools. After ChatGPT came out, um, I was like, deep into prompting. And prompting was my leverage against marketers and growth folks, not against true AI, uh, engineers or anything like that. And then over time, if you just tell Claude what you're trying to do, it would give you a better prompt than I could do. And so the models just get better. I think this is an example where if you're off put by this and you're like, wow, this is going to be really hard. I suspect at some point the model will build the loop workflow for you based upon your outcome.
Speaker B: Yeah, exactly. It's almost like if you just can think of what you want, it'd be able to describe it. It's like we're getting to that point, which is crazy.
Speaker A: Uh, you still need to have the ideas and you have to have the prioritization of what is worth doing. But, yeah, get into the point where the agent will just figure it out.
Speaker B: It just figures it out. And so here, I'll show you. So we type in Hermes model in this terminal. It spins up here. You can see all these different providers. I just click. I use the news portal. It's the company that makes the, uh, Hermes agent. I have a subscription with them and they give you a link. So you just hit enter when you select your model provider. I copy this link, I go over here, I paste it, and I just click connect. And once I do that, I come back here. Boom, login successful. And now I can select whatever model I want to use. And you can see all of them here. You can see there's Deepseek, there's GLM 5.2. And you can see how, oh, my gosh, it's so, so affordable. 98 cents. As opposed to, uh, what's anthropics? It's, uh, $5. Um, so now I will say this. Um, I'm going to use Anthropic for the sake of this video because it is faster. GLM 5.2 is almost equivalent in the end outcome, but for whatever reason, Opus is faster. So for the sake of this video, I'll use it for that. So enter, enter. We're all set up. We're ready to go. So now when I spin up a terminal, I just type in Hermes M. I have the harness, of course, the Hermes harness. And now I have it connected to Opus 4.8 in this case, or GLM 5.2, whatever you want to use. If I say hi to just test it, the Agent's going to reply to me. Now we can just get started with getting it connected to all of our tools. When you want to use something like, um, once again, any kind of connector, like HubSpot or Supabase, uh, wherever your data lives, use Composio. Make sure you're on the, for you on the top left here, click 4U, connect everything, boom, boom, boom. Click Install. And when you click over here, you can just click the Open Claw install. It's the same thing. Click that. I'll go ahead, I'll do it, and I'll actually give the instruction here. I'll just copy this and I'll rotate my key after this. I'll copy this instruction under MCP and then actually I'll even click API key over here and I'll copy this instruction and I just paste it to the agent. And the agent is going to install the Composio connector so that it can access all of those other connectors that we set up with Composio takes just a couple seconds. Yeah.
Speaker A: So now it has access to all of the tools.
Speaker B: Now it has access to all the tools. And then the next thing you would want to do, and we'll include in the show notes as well. I have a prompt that builds, uh, Obsidian vaults.
Speaker A: Yes, now you're talking my language. I'm pretty obsessed with Obsidian. Graphify all these different things.
Speaker B: Yeah, yeah, tell me, tell me.
Speaker A: So like, I, yeah, like pretty much following the standard AI geek, uh, read, you know, the Andre Karpathi all on wiki. So Obsidian allows you to turn your local file system into a wiki, uh, for your agent. And so mine does pretty much somewhat standard things where if I run a command called ingest, it will go off and find all of the raw intelligence that lives across my docs, my comms channels. I have a staging folder called raw. And so I download all of my docs that I get. I have a script that will basically go in if someone sends me a loom. It will take the transcript and put it into the staging folder. And then I have logic that will say, these are the types of intelligence I'm looking for. And it will route it to different places. And so I have a bunch of like, files that will tell me how different teams are performing, how different bets are performing, and it routes it into there. And so the beauty of AI, uh, Second Brain is you have an LLM wiki for your agent. It has all of the intelligence across everything you care about. It's continually updating it. And for me, it's amazing because I send my little, uh, meeting note taker to like meetings that I don't go to, but then it has that context. It ingests all of that context, extrapolates all the intelligence. And so then anytime you work with, I do mine in Claude code, but anytime you talk to your agent, it's preloaded with your MD file that tells it where to go to find the intelligence and any question you're asking it. And so I'm like, hey, what's the update? What blockers does this team has? And it's like, oh, I found. This is what's happened across Slack. You haven't come back yet. They're stuck here. There's a meeting thread on that, and there was a meeting that you weren't in that said this about that. And this is how I think you can solve it. So you are continually talking to an agent that is preloaded with all of the intelligence you need and it's mapped to what you care about. This is like the things that I care about. And then actually has my core outcomes. It has all of my goals across my different teams. So that's what I built. And the thing I'm building, um, so it's portable. So mine basically syncs with GitHub so I can use it with any kind of cloud web, Claude cowork, cloud code. So I can use it on my phone. But I have so many different ideas. I was in the, I was talking about, it was a clay conference and I presented it for the first time. And it's the thing that's changed the way I work the most ever. Like, I, like, I could never go back to not working that way.
Speaker B: Yeah. And do you have it have context over everything? Like people projects as well?
Speaker A: Yeah, yeah. It builds basically. Uh, so it has a file for each of my direct reports. But then what it will do is if someone comes up two or three times, it has a wiki folder for intelligence. So it will say, hey, here's something you're learning. It's not associated. I have mine broken into bets and people. So bets are the core outcomes. Uh, I'm trying to drive and people are the people who report to me. But it will find a concept that is not routed to either of those, but say, hey, this is really important. And it will build a wiki page for it so it has that intelligence. And then if I've met a person twice and that's a person that I'm learning about, it will just auto build a file for that person. And so it's capturing and it's capturing insights about them.
Speaker B: Do you have it set up so like it's running on a schedule to pull in all this context?
Speaker A: No, I have a command called ingest and whenever I run an ingest it will look in the log file and say the last time you run this command was two days ago. And so it only looks for the last two days of updates.
Speaker B: Nice.
Speaker A: Yeah. And then I have a command called lint that will look across and see and try to clean it up. So we'll say, hey, this thing is in contrast to this thing, uh, which one do you want to use? And so it tries to keep the whole intelligence layer. It's running the Obsidian plugin to sync with GitHub and that syncs every five minutes.
Speaker B: Nice. Yeah, that's like. And I love Obsidian because you can sync it to GitHub but also they also have like for 4 bucks a month you can sync it to their own managed cloud. Yeah, it syncs live. Like it's like it's all instant.
Speaker A: You can sync across laptops. So if you wanted to have a team based one, you could have it in GitHub and across obsidian on your laptops.
Speaker B: Yeah, I love Obsidian. And uh, so that is a prompt. I have a prompt and it'll be in the show notes. You can give this prompt and what it does is it actually will go through all of your connectors via Composio or however you have your agent connected via MCPS or what have you and it'll build a vault for you. It takes some time, it takes honestly several hours. But it's like a Carpathy styled wiki vault. Obsidian Vault knowledge base. And it's super important because with an AI employee you want it to have context, you want it to know who's doing what, who is this, what's this project, why is this important? And you don't want it to have to always fetch it through the tools and the connectors. If it could just do it through Obsidian, it's faster, it's all there. Um, and it's very agent friendly. So now we have Composio installed. I'll just do a quick reset uh, here, uh, in the terminal just to make sure and I can ask my agent, hey, is Composio working? And let's see what it says. Also while it's replying here, I'll just kind of like talk through a couple other uh, cool features. You can see this plug icon up here. If I click connect an agent, if I want to just run this terminal, rather than running it on the web, I can just copy, uh, this Orgo ssh Kip agent. And then I can open up my own local terminal and I could just type orgo and copy that paste command Orgo ssh kip agent. And now I have it running on my computer. I type in Hermes.
Speaker A: That's cool.
Speaker B: And this is actually the Hermes agent. And I can ask it. Hi, is Composio working? And let's see what it says. So now I could actually just operate the cloud computer from my own local computer. And you can see here, Composure, it's finding the tool, it's searching that, and it seems that that's working. Yep. So now Composure is working. I'll jump back here into Orgo. And the next thing we need to do is give our, um, agent an email for this purpose because we want it to be able to obviously send out these email sequences to our, to our customers. So I like agent mail. It's super easy. Go to agentmail to. It's free to get started. And I created an inbox. I called it the orgo claw@agentmail to. And I created the inbox and I could actually connect agent mail via, uh, Composio as well. You can see here just find the agent mail connector, connect it over here. Boom. And now once you have that connected, your agent will have access to its own email inbox. And now you literally just would tell it, which I'll also include this prompt. You will just tell it, hey, I want you to build out this customer re engagement sequence. And because it's connected to all of your tools on Composio. For instance, for me, my customers, they live on stripe. They live in Supabase. They live in like these connectors. I have connected in Composio. So uh, it found all the ones that are high candidates to reach out to using this prompt. Built out everything, literally took like five minutes. I'll include this. Let's go ahead. Let's paste it here and tell our Kip agent to, uh, build this out.
Speaker A: So this prompt is that graphic you showed. It's just that workflow in a prompt telling us to do these things in sequence. Is it triggered like it runs once? Does it run every day? So how is the frequency set up?
Speaker B: Yeah, it sets up these scheduled tasks through cron jobs. This was literally me asking the agent, like this morning, and it took like 5 minutes, 10 minutes max. I was like, hey, I want to build a customer re engagement agent. Um, and I was like, you know, can you help me do this? I've connected to Composio, so you don't even have to paste this prompt. You could just tell it what you want and it will do it all for you. Like, that's genuinely what I did. And yeah, it says here, email one email to. It has like a whole sequence. Yeah, Sequence three email templates. Here's the personalization it's going to do, here's the timing of it, here's the offer that it's going to create to re engage these customers. Uh, yeah, Day zero, day three, day six. So, yeah, it came up with this idea and I'm just approving it. And what you might want to do is say like, hey, look, um, use your agent mail email to send these emails. And look, when the customer replies, I want to be pulled into the loop. And what's cool is the minute you set this up and the agent has this up and running and it's like it's on automation, it's running. You could text it, you could text it via telegram or however you want to connect, uh, M to the AI agent or Slack or whatever and you say, hey, how's it going? How's that email nurture sequence working? Did we get any replies on that? Oh, we did. Okay. Like, it's kind of cool. It's truly like an employee.
Speaker A: Yeah. I think the cool thing here is so your, your agent has its own email inbox. And so if, if on the off chance it did go a little rye, because I think that's what people worry about it, it basically said something, you know, that wasn't great or said something that was low quality. It has its own inbox. And so people do know that that is not you. Like, I think there was the early days, people were like emailing on their behalf, I think agent having its own inbox. And then the cool thing is what you probably set up here is if there's a reply, the agent then will just reply with you in the thread. Oh, I'll hand over to like Nick now. And so it loops you in. So it really is like a little, um, bdr, a little sdr, kind of doing all your qualification and prospecting.
Speaker B: Yeah. And I, like, I'll even show you here. Um, when I was doing it this morning, like I called it Hermes employee number one. Here we go. And I have the terminal here. And you can actually see like when I told it about, hey, I want to build this out, I literally just said you can send all your emails from oraclawgentmail to here's your email or you could even draft it. It's literally me just talking in a, uh, kind of exploratory way to this agent that's like smarter than me. It's like, okay, cool. And then it uses all the connectors, builds everything out. He's like, look, here's what I'm gonna do. I just detected that you have this many customers in Stripe. I just detected, I'm going to draft these emails. I'm going to make a review doc for you. It gave me a link. You can see here it has a link. I open the link, it's like, here's the customers we should reach out to. Here's what to say. And I'm like, okay, cool, let's do it. So you can just tell it what you want and it'll build it out and it's amazing. It's like so simple to get started. Literally took me five minutes to get this set up, um, this morning before the show.
Speaker A: So, you know, when you look at your graphic, I think it's really well done, the graphic, but it's actually even easier than this graphic, which is.
Speaker B: It is.
Speaker A: Hey, you went through, you used Orgo. You basically set up your first virtual computer. You pre installed with Hermes. You went into Hermes and you said, hey, here's kind of what I'm trying to do. Um, here's the composio. So you have all of the different tools. I've set you up an agent mail. So you have inbox. Here's what I'm kind of trying to do. Like I'm trying to build this employee that will do this task for me on, you know, every three days we'll do this task. And how do I do that? And then you've connected it to like a really smart model. Uh, I think in that case it was cloud. But you, you were kind of recommending GLM is much cheaper if you want to do this on scale. And they would say, okay, well here's how you should do everything. Set it up. You can like text a little guy and you can say, okay, how's it going? Are you getting Hm, many replies? I suspect you could probably say, hey, um, what's the, what's the open rate? Open rate's like 10%. Okay, what can we do better?
Speaker B: Right, right.
Speaker A: You can start to like work with it as an employee. Okay, like adjust that. That's a good suggestion. That's not a good suggestion. Try this update yourself. Try those things for the next 50 mills. Tell me what the results are.
Speaker B: Yeah, it's, it's Insane.
Speaker A: That is insane. Yeah, I think like, why wouldn't you do this? It's like you can let, you can literally have an army of autonomous agents doing. I think the best place to start, would you agree, is like find some like pretty well controlled workflows to get yourself familiar and trusted. And I think if you're like, well, I don't want my first one to be emailing folks, then like is you had it in there? Well, it can do the, it can do the draft, it can set up all the drafts and you can send the emails until you get comfortable because it has its own email inbox. I probably would just let it go for like, I would say, hey, email the first thousand and then we'll review the results after a thousand sends to like get going and check in how it's performing.
Speaker B: Yeah, and even that right there of like, hey, like, hey, how about this? Do the first thousand and then send me an email about the performance of that first thousand and how it went and what we could do better and then we'll, we'll go from there. You know, it's like, it's truly that simple. So if a quick recap like it comes down to get composio set up, it's free. Connect all your tools to it. Okay, now it has access to all these tools and context. Cool. Give it its own email so it can be able to do work, real emails, sequences, so on and so forth. Connect your model glm, uh, 5.2, give it a computer on Orgo and lastly give it an Obsidian vault. Give it a knowledge base and a brain so it has an even deeper understanding of what your company does, who you are, who are these people you work with, so on and so forth and then just describe and nudge it in the direction of your goals. Yeah, it's genuinely like that.
Speaker A: That bit's killer actually. That I have an AI brain that gives me core intelligence about my company people, my outcomes. And in this model you give the agent the brain and the agent has all of that context to go do its own work so I don't have to like it's using the second brain to do the work. I think that is like killer actually I am going to do this tonight. That's how powerful I think especially the part given at the Obsidian Vault. Um, I think this is killer. I think this is truly speaks to one of your core skills is to build your own team. Like everyone is a manager. Because now I can create a whole team to execute that domain expertise across such a broad set of things. And I feel like that's a core competency you're going to need to have is like I, I can be a quote unquote individual person working on things, but I have a team of agents that are actually executing all of this work.
Speaker B: Yeah, I think one of the coolest things about when you get set up on Orgo is obviously here we set up the KIP agent and what we did thus far on the KIP agent, we gave it the connectors to Composio, uh, and I guess we initiated its prompt so it could start building up these sequences. But what you could do is you could clone it. And so I click clone here and I actually just created a complete clone of this computer and the Hermes that we installed in all of the system prompt and everything. And if I open this computer now, this is a clone of that AI employee we just made. So now we have two. Wow. And now I could tell this one, I spin up Hermes and I'm going to say, hey, okay, we have KIP agent one. He's going to do the email sequencing and he's going to do that. But I want you to maybe check his work and maybe do the follow ups or so I don't know. Yeah, yeah, yeah. Uh, or do a sale. Do something sales related like.
Speaker A: Yeah.
Speaker B: And then you could just quickly see how like, okay, like we're going to have a lot of these agents. Shoot. If you want to spin up a Windows computer in here so that you can install OpenAI's Codecs desktop app into Windows. And then maybe Codex related. Codex might be better at uh, the goal mode might be better. Whatever the next new agent is, whether it's Hermes, OpenCloud, Codex, Claude Code, have it here in a workspace, have them working together, talk to it from your phone and manage your whole fleet of AI, uh, employees. This is clearly where the future's going.
Speaker A: Yeah, clearly 100%. I couldn't agree more. I think, Nick, this is being a awesome overview. I love it when I do one of these and I just get a ton of new tools to go try. I've got um, three brothers, we're all on WhatsApp talking about this stuff all the time. And I've already texted them a bunch of things that you have shared and so I really appreciate this. I think that this is incredibly actionable. I think it's something that everyone could go try out and build their first AI employee and get a ton of value from.
Speaker B: Likewise. I can't wait to see what everyone does. I'm sure people will have even better ideas than me. And it's really, it's the people who have their domain expertise. They're in an industry, like some niche manufacturing industry and they're like, oh my God. I just rebuilt a software with my agent that costed us $6,000 a month and I just told her what we need and I built it out in a couple days. It's like the things you can do today with, with AI and, and these agents is incredible. So I'm excited to see what people do. Cool.
Speaker A: Thanks a lot, Nick. Appreciate you coming on.
Speaker B: Cheers. Talk to you soon.
Speaker C: Let me tell you about a show I love. Hosted by my friend Ross Simmons. Each episode hosts an in depth analysis of some of the greatest creations and creators of all time with deep dive conversations on the creative process that went into building companies, brands, stories and more. If you like learning about history, learning about the creative process, you'll love this podcast. This podcast is perfect for anyone who wants to learn how to systematize their creativity but still be organized. Ross just did a fantastic episode about Reddit and AI distribution in the playbook that everyone is missing. Listen, wherever you get your podcasts, this
Speaker B: data is wrong every freaking time.
Speaker A: Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated. Whoa.
Speaker B: I can see the client's whole history. Calls, support tickets, emails. And here's a task from three days ago I totally missed.
Speaker A: HubSpot grow better.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
How Top 1% Marketers Are Shifting Their Strategy in Q3 2026Automate Boring Tasks With Codex & Claude Code in X MinutesHow Beehiiv's Founder Turned a Newsletter Into a $1M PipelineThis AI Second Brain Remembers Everything I Save (Codex)Google is overhauling Search - what you need to know
