The argument in one line.
Grok 4.5 makes AI agents dramatically faster and cheaper than the prior model generation, but the real unlock is giving an agent its own computer, email, phone number, and payment method so it operates like a genuine co-founder instead of a scoped-down automation.
Read if. Skip if.
READ IF YOU ARE…
- You run (or want to run) an always-on AI agent in a harness like Hermes or OpenClaw and are deciding which model to power it with.
- You're a solo builder or small team considering an AI-agency or managed-AI-employee business model for other companies.
- You already use MCP-connected tools (idea research, social trends, video analytics) and want to see how they chain together in a real session.
- You want a concrete sense of what an agent can produce unsupervised in under an hour when given full tool access.
SKIP IF…
- You want a no-code explainer — this assumes familiarity with terminals, SSH, and connecting MCP tools.
- You're looking for an independent benchmark — the Grok 4.5 vs. GPT 5.6 Sol comparison here is one live demo, not a controlled test.
The full version, fast.
Nick Vasilescu, co-founder of Orgo, argues Grok 4.5 is the release that makes AI agents feel like a genuine co-founder rather than scripted automation. Running inside a Hermes agent harness on an Orgo cloud computer, Grok 4.5 delivers roughly Opus-4.8-level intelligence at about a tenth of the cost and 10-15x the speed of the prior generation. Live, it builds a landing page in about 40 seconds, edging out GPT 5.6 Sol on design and copy. Given tools — email, a phone number, a debit card, and connectors like Composio, Idea Browser, X MCP, and vidIQ — the same agent generates ten startup ideas, then builds a full landing page, thumbnail, market-insight one-pager, and cold-email sequence in one continuous session. The takeaway: give an agent more tools, not less, and falling cost plus rising speed compounds every time a new model ships.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Who's talking.
00:43guestNick Vasilescu
Where the time goes.
00:00 – 01:40
01 · Intro
Greg cold-opens on Nick's conversion to Grok 4.5 and frames the episode as a hands-on walkthrough of using it as an AI co-founder.
01:40 – 03:39
02 · Why Grok 4.5 release matters
Nick explains Grok 4.5 as roughly Opus-4.8-level intelligence at a fraction of the cost and time, in the same week as Meta's Spark 1.1 and GPT 5.6 Sol.
03:39 – 05:16
03 · Automation versus a co-founder
Nick reframes the goal away from hands-off automation toward an agent with its own computer, email, phone, debit card, and notes vault.
05:16 – 09:01
04 · Setting up Hermes and Grok 4.5 on Orgo
Nick spins up a fresh Hermes instance powered by Grok 4.5 on a new Orgo cloud computer, live, by texting his existing agent Dewey.
09:01 – 11:23
05 · Why Orgo to manage Agents
Nick contrasts running agents locally versus on cloud computers, and shows his own template stack and fleet of roughly 200 agent machines.
11:23 – 14:02
06 · Grok 4.5 Cost discussion
Nick lays out the cost paradox: Grok 4.5 is a tenth of the cost and far faster than Fable, but total spend rises because the freed-up time gets used.
14:02 – 16:20
07 · Grok 4.5 Fast Execution and Unlock
Both hosts note how Grok 4.5's speed shifts the interaction pattern from send-and-wait to a tight, conversational back-and-forth.
16:20 – 19:13
08 · The Agent tool belt
Nick tours his connector stack: Agent Mail, Agent Phone, Agent Card, Composio, Latitude for observability, and Linear.
19:13 – 20:37
09 · X MCP for trends
Nick demonstrates using an X (Twitter) MCP to surface trending topics and content ideas, including a story about building an agent from a DM thread.
20:37 – 22:11
10 · vidIQ for outliers and thumbnails
Nick shows vidIQ surfacing outlier YouTube videos as a source of content ideas and thumbnail inspiration.
22:11 – 26:15
11 · Finding new startup ideas
Dewey is asked to generate ten startup ideas using Idea Browser, X MCP, and web search tools together.
26:15 – 30:56
12 · Grok 4.5 versus GPT 5.6 Sol
Both models are given the same landing-page prompt; Grok 4.5 finishes in about 40 seconds and Greg prefers its design and copy.
30:56 – 34:06
13 · Ranking and Reviewing the startup ideas
Dewey ranks its own ideas and consolidates on a bootstrappable managed-AI-employee agency as the strongest option.
34:06 – 38:46
14 · The AI agency opportunity
Greg compares the AI-agency moment to the early Facebook/Google-ads agency wave and references MoltBook's fast acquisition as a sign of agent-native demand.
38:46 – 40:10
15 · Thumbnails over Telegram
Nick messages Dewey over Telegram to generate a YouTube thumbnail using his face and vidIQ-sourced outlier patterns.
40:10 – 41:58
16 · Reviewing AI Agency Landing Page
Dewey's landing page for the managed-AI-employee business is reviewed, complete with three-tier pricing and a lead-qualification flow.
41:58 – 43:36
17 · Vertical MCPs and agent startups
Greg and Nick discuss how vertical-specific MCPs fill the context gap general LLMs lack, framing it as its own startup opportunity.
43:36 – 45:25
18 · Skill graph and the offer
Idea Browser's skill graph recommends Dewey narrow the offer to one weekly client-reporting deliverable instead of selling AI employees broadly.
45:25 – 46:42
19 · Reviewing the Thumbnail Generated
The Telegram-requested thumbnail comes back, styled to Greg's channel and informed by outlier-video research.
46:42 – 48:07
20 · Email Outreach Campaign
Dewey is asked to build a full cold-email outreach sequence for the managed-AI-employee offer in a Google Doc.
48:07 – 50:45
21 · Reviewing Market Insight 1-Pager
Dewey produces a market-insight one-pager with segment-level revenue and meeting figures to support the sales pitch.
50:45 – 52:12
22 · From a Camry to a Ferrari
Nick frames the model jump as going from an expensive, slow Camry to a cheap, fast Ferrari in a few months, with more gains compounding ahead.
52:12 – 53:22
23 · Reviewing Cold Email Outreach Sequence
The finished cold-email sequence — ICP, offer framing, and full sequence architecture — is reviewed in the shared Google Doc.
53:22 – 56:00
24 · Closing thoughts
Nick closes on the shrinking gap between idea and implementation; Greg sends listeners off to build.
Lines worth screenshotting.
- Grok 4.5 was described as delivering roughly Opus-4.8-level intelligence at about a tenth of the cost and 10-15x the speed of the prior-generation model it replaced.
- The real unlock of an agent harness isn't automation — it's giving the agent its own computer, email, phone number, debit card, and memory so it acts like a co-founder.
- Faster, cheaper agents don't reduce total spend — they get used more, so usage and total cost often rise even as the per-task price falls.
- In a live test, Grok 4.5 built a full one-page landing site in about 40 seconds, finishing well before a competing model attempting the same task.
- An agent connected to enough tools generated a startup idea, a landing page, a thumbnail, market research, and a cold-email sequence in one continuous session.
- Startup-idea tools trained on thousands of past landing pages and launches give a model 'taste' that raw web search doesn't provide.
- A vertical-specific, template-based AI service (an agent pre-configured for one industry, like HVAC) can charge far more than a horizontal AI subscription because it removes setup work the buyer would otherwise face.
- Running multiple parallel agent sessions lets one operator work several workstreams at once instead of waiting on a single agent to finish sequentially.
- An observability layer that detects user frustration in agent conversations can flag a missing capability before it becomes a support complaint.
- Advice to sell one specific, visible weekly deliverable — not a general 'AI employee' — mirrors why productized, scoped services close faster than open-ended platform pitches.
- Cost of frontier intelligence keeps falling while speed and capability keep climbing, so agent infrastructure built today gets free upgrades with every new model release.
- The gap between having an idea and shipping a working prototype — landing page, thumbnail, and outreach sequence — has shrunk to a single working session.
Tool access, not automation, is what makes an agent useful.
WHAT TO LEARN
A faster, cheaper model matters less on its own than what happens when you hand an agent real tools — its own computer, inbox, phone number, and payment method — and let it chain them together.
02Why Grok 4.5 release matters
- Grok 4.5 pairs roughly Opus-4.8-level intelligence with a fraction of the cost and processing time of the model generation it replaces.
- The jump matters less for raw capability than for the user experience it unlocks inside agent harnesses like Hermes and OpenClaw — tasks that used to feel sluggish now feel instant.
03Automation versus a co-founder
- Treating an agent purely as automation caps its usefulness; treating it as a co-founder with full tool access is what actually changes outcomes.
- The demo agent had its own computer, email, phone number, debit card, and notes vault — the tools a real co-founder would need, not a scoped-down bot.
04Setting up Hermes and Grok 4.5 on Orgo
- An agent can be instructed to provision its own new cloud computer, install its own harness, and configure its own model, turning setup into a self-service step.
- Running an agent on a dedicated cloud computer keeps it reachable by text at any hour, since a personal machine going to sleep would take it offline.
05Why Orgo to manage Agents
- Fleet management matters once you're running many agents — one operator in this demo was managing roughly 200 concurrent agent computers.
- One-click templates bundling a harness, model, and connector stack remove the setup step for anyone copying a working configuration.
06Grok 4.5 Cost discussion
- Grok 4.5 was described as roughly a tenth of the cost and 10-15x the speed of the prior-generation model at completing the same task.
- Cheaper, faster models don't reduce total spend — freed-up time gets filled with more tasks, so usage and total cost tend to rise even as unit cost falls.
07Grok 4.5 Fast Execution and Unlock
- A model fast enough to respond in seconds changes the interaction pattern from send-a-task-and-check-back-later to a tight, conversational back-and-forth.
- Speed becomes a feature in itself: operators described needing to work faster themselves to keep pace with a model with no dead time between turns.
08The Agent tool belt
- A working agent stack in this demo included a cloud computer, an agent email inbox, an agent phone number, an agent payment card, app connectors, an observability layer, and project tracking.
- An observability tool that detects user frustration in agent conversations can flag a missing capability before it becomes a support complaint.
09X MCP for trends
- A connector to a social platform's search API lets an agent surface currently-viral topics and content angles without manual browsing.
- Giving an agent read access to direct messages let it build a customized agent for someone else's use case purely from the conversation, with no separate briefing.
10vidIQ for outliers and thumbnails
- A YouTube analytics connector that flags outlier videos gives an agent data-backed content ideas instead of guesses, and doubles as thumbnail research.
11Finding new startup ideas
- Startup-idea tools trained on thousands of past launches give a model 'taste' that raw web search doesn't provide.
- An agent given several research tools at once will pull from all of them and cross-reference — the resourcefulness comes from tool access, not prompting skill.
12Grok 4.5 versus GPT 5.6 Sol
- In a same-prompt, same-task landing-page build, one model finished in about 40 seconds while the competing model took noticeably longer for an equivalent result.
- Both models produced usable output — the practical difference observed was speed and token efficiency, not a capability gap.
13Ranking and Reviewing the startup ideas
- When ranking its own generated ideas, the agent optimized for buildable-and-monetizable-this-week over ideas requiring outside funding first.
- A vertical-specific version of a horizontal AI product can justify a much higher price than a generic subscription, because it removes the buyer's setup work.
14The AI agency opportunity
- The AI-agency model was compared to the early-2010s wave of paid-ads agencies: the same window of being early to a new distribution channel.
- A social platform built specifically for autonomous agents was acquired within about a week of launch, illustrating how fast agent-native infrastructure can find demand.
15Thumbnails over Telegram
- Messaging apps work as an alternate interface to the same agent, useful specifically for exchanging images and video rather than typing in a terminal.
- A thumbnail request combined a face reference, channel style, and outlier-thumbnail research pulled automatically from connected tools, with no manual asset upload.
16Reviewing AI Agency Landing Page
- An agent-built landing page included positioning copy, a tiered pricing table, and a lead-qualification flow — a full offer, not just a design mockup.
17Vertical MCPs and agent startups
- General-purpose models lack vertical-specific context, which is the gap niche MCPs are built to fill — treated here as its own startup opportunity.
18Skill graph and the offer
- A structured skill-graph tool recommended narrowing a broad AI-employee pitch to one specific, visible weekly deliverable rather than selling the platform generally.
- Selling one weekly artifact instead of a general offer mirrors why productized, scoped services tend to close faster than open-ended platform pitches.
19Reviewing the Thumbnail Generated
- Multiple thumbnail options were generated in a single request, giving the operator a choice instead of one locked output.
20Email Outreach Campaign
- Agent value extends beyond production (landing page, thumbnail) into distribution — the same session drafted a full cold-outreach sequence.
- Getting customers was framed as equally or more important than product polish, and treated as something the same agent could own end-to-end.
21Reviewing Market Insight 1-Pager
- A generated market-insight one-pager included segment-level revenue and meeting estimates formatted to support a sales call, not just internal research.
- Proof documents a prospect can see before committing were framed as part of what turns a cold pitch into a booked pilot.
22From a Camry to a Ferrari
- The model-generation jump was framed as moving from an expensive, slow product to a cheap, fast one in the space of a few months.
- Because cost and speed both keep improving, infrastructure set up today gets better automatically every time the underlying model is swapped.
23Reviewing Cold Email Outreach Sequence
- A completed outreach sequence delivered as a shared document included the ideal customer profile, offer framing, and a full multi-email architecture.
- One agent handled the full loop — website, offer, thumbnail, and outreach copy — inside a single continuous session with no separate tools opened manually.
24Closing thoughts
- The gap between having an idea and having a working implementation was described as shrinking toward zero, removing technical execution as the bottleneck.
- The closing advice was to actually use the tools rather than keep consuming content about them — the barrier described is action, not access.
Terms worth knowing.
- Hermes
- An AI agent harness/runtime that gives a language model persistent tools, memory, and the ability to act autonomously on a computer.
- OpenClaw
- An open agent-harness ecosystem, referenced alongside Hermes as software that turns a raw model into an autonomous, tool-using agent.
- Orgo
- A cloud-computer provisioning service built for running always-on AI agents, offering one-click templates and SSH access so an agent stays reachable even when a personal machine is off.
- MCP (Model Context Protocol)
- A standard that lets an AI agent connect to external tools and data sources, such as social search or startup research, as callable functions.
- Composio
- A connector service that links an AI agent to third-party apps like Gmail, YouTube, and Google Docs with one-click authentication.
- Idea Browser
- A startup-idea research tool and MCP that gives an agent access to patterns from thousands of past startups, landing pages, and launches so its output reflects market 'taste.'
- vidIQ
- A YouTube analytics tool that surfaces outlier (over-performing) videos and thumbnail patterns within a content niche.
- Skill graph
- A structured map, in this case from Idea Browser, of the steps needed to take a raw idea to a sellable offer — used here to recommend narrowing to one weekly deliverable.
Things they pointed at.
00:43toolOrgo
05:16toolHermes
16:46toolComposio
16:55toolLatitude
19:13toolX MCP
20:37toolvidIQ
26:50toolGPT 5.6 Sol
Lines you could clip.
00:01
“I'll say it. I'm a Grok convert.”
04:43
“Dewey has his own computer, his own email, his own phone number, his own debit card, uh, of course, Obsidian Vault.”
11:44
“This is like a tenth of the cost of something like Fable. And it's like, oh, this is like 10 to 15x faster.”
50:29
“It feels like a Ferrari.”
50:41
“Would you say we went from, like, a Toyota Camry to a Ferrari?”
52:00
“You didn't have to do anything. You're just riding the wave.”
55:10
“The gap between idea to implementation is shrinking down to nothing.”
Where the conversation goes.
00:00 – 05:16denseGrok 4.5 release and positioning
05:16 – 14:02denseSetting up the Hermes/Orgo agent stack
11:23 – 16:20steadyCost and speed tradeoffs
16:20 – 22:11denseTool and connector ecosystem
22:11 – 34:06denseModel comparison and startup-idea generation
34:06 – 48:07denseBuilding the AI agency: landing page, thumbnail, offer
48:07 – 56:00steadyCost/speed trajectory and closing thoughts
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00I'll say it. I'm a Grok convert. And I don't say it lightly.
00:04Grok 4.5 is a huge deal. By the end of this episode, you're gonna understand how to use Grok 4.5 with Hermes to act as almost this AI cofounder for you.
00:15We break down everything for you, every step of the way. And the people that stick around to the end of this episode, I believe, will be able to outperform 99% of people on this planet.
00:26Grok 4.5, I didn't think it was a big deal, but it is a big deal. And by the end of this episode, I think you'll be a convert too.
00:33Enjoy the episode.
00:42Nick from Orgo back on the pod. He's my agent guy. I had to bring you back on, Nick, because I'm I'm seeing a lot of buzz hype around Grok 4.5.
00:52Nick, by the end of the episode,
00:54what are people going to learn? I'm gonna teach you everything there is to know about why Grok 4.5 is the best model to be running inside of something like Hermes or Open Claw today. And the real big unlock here is to treat it like a genuine cofounder, something that has access to all your tools, capabilities, so on and so forth, and and stop, you know, being so shy about what AI can do.
01:18When you give a tool like, uh, Hermes and you you run it with Grok and it's so you know, it just wants to reach for every tool and actually go out and do things for you. That's the big unlock.
01:29And a lot of people, they're gonna experience for the first time what that feels like with Grok 4.5 just because of how fast, cheap, and accessible it is to everybody. So, yeah, I'm excited to get this in the hands of everyone.
01:40Okay. So, I mean, walk me through that a little bit. Like, you know, what why is this Grok 4.5 release so big,
01:48um, and why should why should people care? So the the biggest thing with Grok, um, the reason why it's such a kind of viral moment right now is because we've had intelligent models since, you know, Opus 4.5 came out. And in a lot of ways, the release of something like Opus 4.5 from Anthropic back in December and January, that was what led to the Clawdbot, the OpenClaw moment.
02:13That model capability unlocked something with these harnesses to be able to use tools, uh, for these AI agents that before was, like, you know, really not possible.
02:25And now with Grok 4.5, you have that insane level of intelligence of something like OPUS 4.8 today, the one of the most advanced models from Anthropic, But, also, you have it at a fraction of the cost and then at a at a fraction of the time that it would usually take for it to complete a task.
02:44I mean, this model is insanely, insanely fast, and it just unlocks, like, a new user experience for something like OpenClar or Hermes. Like, it really makes it feel like it's a brand new fresh product just because of how capable it is.
02:58So, yeah, I I would say, like, that's the biggest breakthrough with with Grok 4.5. And then, of course, like, all in the same week, have now, uh, meta just came out with their spark 1.1 model, and, uh, GPT 5.6 soul came out.
03:14And I've tested all of them, And I think that they're all great. Like, they have their benefits. For instance, GPT is a lot better at, like, computer use tasks.
03:23But still today, like, to this day, uh, and I guess the last forty eight hours, I haven't been sleeping much. Grok 4.5 is just my workhorse.
03:32Um, and so I just think that everyone should just be plugging this into their, uh, current stack,
03:36connecting it to their tools, and seeing what it could do. So on this FigJam, like, what am I looking at exactly?
03:42Yeah. So here, I'm just kinda trying to demonstrate. A lot of people when they think of something like Hermes or Open Claw, the first place that their mind goes to is automation and just like, I could be hands off, and things are just gonna be solved for me.
03:59And I think that's great, and that's fine and all, but I actually think it's, like, the wrong frame of mind. And let me tell you why. I think with automations we've had automations for a while now, like, even pre AI.
04:10We've been able to automate tasks and and and create kinda workflows and deterministic pipelines of, like, things that could be automated, quote unquote. But to me, like, that's not the big unlock of something like Hermes and OpenClaw. To me, the unlock is I genuinely feel like I have an AI cofounder who, a, has access to all the different tools that matter to me and all the different context and all the just everything that would make a a cofounder great, like, in terms of capability.
04:43That's what Hermes today, my Hermes, has access to. His name is Dewey. And Dewey has his own computer, his own email, his own phone number, his own debit card, uh, of course, Obsidian Vault.
04:55And I think people need to stop being so shy about the tools that they're giving their their agent because the minute you give it more of these tools and connectors and capabilities, that's when the magic really comes out. And then a model release like this, 4.5 Grok, it really just takes everything to the next level.
05:13Uh, and overnight, your whole stack just got 10x. So
05:17So okay. So may you know, someone listening to this, maybe they they buy this.
05:22They're like, okay. Grok 4.5. It's it's it's .
05:25This is how I should be using it. Can you give some examples? Like, can you show Yeah.
05:30You know, how someone could actually be utilizing Grok 4.5 and use it in this particular way?
05:38For sure. Yeah. Let's we can just, like, walk through, um, how you might be able to use this as, like, yeah, your side your side by side companion.
05:47So, uh, first and foremost, when you're setting up Hermes, like, obviously, I'm the cofounder of Orgo. We give you a cloud computer to to install Hermes or OpenCLI inside of, and we just make it really simple to get, uh, up and going, um, with building out these, you know, AI agents.
06:03Um, and so this is Dewey's computer here in Orgo, And this is entirely operated by my Hermes agent, and he lives inside of here. And what I'm gonna do, for instance, is I have my terminal here on my computer that I can message Dewey.
06:19And first things first, I'll tell him I'll tell him to spin up a a new dashboard so we can kinda see all the configurations and and tools that he has access to. So give me one sec.
06:33Let me just connect to him. I'm gonna say spin up your Hermes dashboard on your computer, and I can share this over here.
06:43So this is my terminal on my local computer. I'll zoom in real quick. This is my terminal on my local computer.
06:50I'm just telling my Hermes agent Dewey to spin up his dashboard on his computer for the Hermes agent local dashboard, and we'll be able to see it on his computer here.
07:03And so one of the big unlocks, one of the things I wanna demonstrate in this video is when you give your Hermes agent access to all these tools and connectors, and you have a model like Grok 4.5 with its insane speed running underneath, um, and powering it, Like, I'm gonna show you all of the cool different demos and use cases.
07:23For instance, content idea generation, thumbnail creation, finding viral startup ideas or or or your content ideas.
07:32You can see here it pulled up its dashboard here, and I'll I'll maximize this and make this a big screen here. But, yeah, all of the different cool use cases that you can do when you really give it access to all these tools.
07:47You can see here, I have a bunch of different, uh, MCPs, plug ins, uh, everything installed.
07:54This is like my stack. So I can kinda walk through first and foremost what that looks like and then dive into, yeah, like, a kinda side by side me and my my Hermes Grok 4.5 cofounder.
08:07Let's do that. Let's go let's go deeper.
08:10So the first thing that I do with every, obviously, with every agent, I set it up on Orgo. I give it an Orgo computer so it can do this.
08:18It can pull things up for me, and I can see it. And I connect it to all my tools. So the first tool that I lost the way
08:25Yeah. I know I know you're a cofounder of Orgo, but if, you know, there's Orgo and if they didn't you know, what's the alternative to Orgo? Just setting it up on a local machine, or how should people thinking about it?
08:37Yeah. You can set it up on your local machine. I I I mean, I don't
08:42I don't like doing that with Hermes or OpenClock because the minute you turn your computer off, all of a sudden, your your agent is off. You can't text it. So you can set it up on other cloud providers.
08:52There's companies like Hostinger or Hertzner. Um, I'm obviously biased, but we generally I think we really do make the simplest easiest way to spin it up. You can see here we have a template option, and we just have a template out of the box to spin up a Hermes, OpenClaw, Cloud Code, and then build your own templates.
09:09I have my own template like my NIC stack, which has all my connectors. And by the way, I'll include this in the yeah. I'll I'll give this to your everyone to have access to, um, so they can spin up an Orgo template using my stack and everything that I'm gonna talk about today.
09:23You don't have to configure it yourself. You'll just be able to use my stack and spin up a template in Orgo. So, yeah, bunch of options out there for kinda where these agents live.
09:33Um, and the biggest thing is I just love the ability to tell my my Hermes agent, my Dewey, hey.
09:41Spin up a new computer with a Hermes agent inside of it configured with Grok 4.5, and it'll do that in the same workspace. Should I should I show you that?
09:49Should I show you that? For sure. Let me show you.
09:52So if I tell Dewey here, hey. Spin up in your minions workspace in Orgo, a new computer with Hermes installed and Grok 4.5 powering it as the model.
10:13And because my agent lives inside of an Orgo computer, it'll be able to spin up a new Orgo computer with Hermes already installed and everything configured. So, yeah, like, when you're using something like Orgo and you're setting up agents and AI employees for other businesses, for instance, like, I've been on a call with a customer who wants a new agent spun up for a certain use case, and I'll just text my agent to do it right then and there, and it'll build it out for me.
10:41So and you're gonna see here, this is a great use case of, like, the the sheer grittiness of Grok 4.5 and just, like, a, how it just it it it's like that's one thing for a model to be fast, but it's another thing for it to be fast and thorough.
10:59And there it goes. So it spun up a new computer, Hermes Grok 4.5. Boom.
11:04This is a fresh computer, and it installed Hermes inside of it. And we're gonna see we're gonna wait and see what it's what it's saying.
11:13Okay. Box is up, installing Hermes, injecting the the API key, pinning Grok as the model.
11:20Like, it's insane. It's so cool.
11:24The other thing is, from what I understand, it's also relatively
11:28cheap. Am I right? Oh my gosh.
11:31Yeah. It so it it is. Here's the paradox of this, Greg.
11:34This is what's fascinating. Truly. Like, I I it's it's not entirely intuitive or obvious that this might be the case.
11:41But when you have a model like Grok 4.5, and it's like, woah. This is like .
11:48And it's like, oh, this this is like 10 x, like, literally 10 to 15 x faster than something like Fable at completing a task. And then you think, like, that's gonna save a lot of money, and it and it does. But then what ends up happening is you're able to do so much more work, and it's not like you're just gonna stop at what you would have accomplished with Fable.
12:08If what you were to accomplish with Fable would have taken you eight hours and you do it in one hour with Grok, it's not like for the next seven hours, you're just not gonna do anything. You're just gonna do more work in that next seven hours, so then you're gonna spend more money. So for me, I got the 300 a month super Grok plan, and I think I'm at like let me see.
12:29I can see it right here. I'm at 24% left.
12:32I literally got this the minute Grok 4.5 came out. I'm at 24% left already. So I mean, I'm burning through a lot of tokens, but I think it's worth it.
12:42You're a self proclaimed token max er. I am a self proclaimed but I'm and I look. Look.
12:47I I I show no loyalty to any lab. Okay? I have them all.
12:51I have the codex. I have the claw. I have perplexity.
12:54I have grok. And and so I when I tell you that this model is great and that I recommend it, it's genuinely it's an unbiased position. Like, I just think that it's such a big unlock and I want everyone to have access to that.
13:07And so, uh, I'm gonna advocate for that. You know? Mhmm.
13:12So let's see here. Okay. So the new Minions Hermes box is live on Grok 4.5.
13:18It spun it all up. And so now when I'm when I'm here inside of Orgo, inside of this computer, I can actually connect to this the same way I connected to my Dewey.
13:27I'm just gonna SSH into this computer. I can spin up a new terminal. Oh, let me I can spin up a new terminal, connect to that, run Hermes, and now here we are.
13:41So now this is the new Hermes agent that Dewey just spun up, and I can say, hi, testing. And let's see. Hi.
13:50I'm here. You see how fast that is? And I'll say just open Chrome just to show that, like, it has its own computer here.
13:56It's gonna open Chrome, and it's a Chrome is installed, launching it.
14:01So it should open it up. Let's see. I will say what I love a model that's fast.
14:06Like, you because you get into the flow. Right? You're, like, back, forth, back, forth, back, forth, and it just feels so nice.
14:12Oh my gosh. Yeah. That is like for me, like, with Grok,
14:17the big unlock exactly was that. Like, I got so used to, uh, like, GPD 5.5 to, like, sending off a task and letting it run-in the like, asynchronously just like, okay.
14:28I come back to it ten minutes later. But with Grok, I'm like, I have to be so hands on, And I'm just like going back and forth back and forth.
14:35I'm like, I I need to I need to up my aptitude. You know? So there it goes.
14:42Let's blend that up, and it's all everything's ready to go. So that's really cool. And you know what?
14:49Okay. I'm gonna do this too. I'm gonna tell it.
14:51Let's also so I don't know if you saw this, Greg, with the GPT 5.6 Sol. Did you did you get a chance to play with that?
14:58I did. So I wanna spin up another Hermes agent with GPT 5.6 Sol as the model running it. Uh, and that way we can, um, we can compare the two maybe, uh, on some on some tasks in this video as well.
15:17So I'll have I'll have Dewey do that as well. Oh, I I sent that into the wrong computer. Let me let me stop.
15:22This is the Grok. This is not Dewey. Let me tell it to the the Dewey.
15:25You have so many computers. You know? You're it's what a what a privileged life you live.
15:30Oh my gosh. I I I think I'm managing over at this point on Orgo. I think there's around, like, 200 computers that I have in my fleet of just agents.
15:40And so when I tell you, like, a stack that I'm using, it's it's me in the trenches. I'm deploying these things into businesses.
15:50There it goes. There's Hermes' soul. You can see it over here on the on the left over here.
15:54And so I I've I've broken a lot of things. I've gotten things to work. So when I when I recommend something, it's it's it's from the genuine of my heart.
16:03I don't want you guys to fail, and, uh, I'm looking out for you. So, um, so yeah.
16:10So we'll have we'll have a a Hermes spin up here for 5.6 Sola so we can kinda compare a couple of the use cases here. But to get back to kind of that stack, that's the computer element.
16:25And the next thing is obviously giving it an email, giving it a phone number, giving it a a debit card, and and the knowledge base and memory layer. Having it connect to all these tools, super valuable, super important, and and then all of your apps.
16:42And there's a lot of great tools for that. I use for connecting it to our apps, I use this company called Composeo, and they make it really easy to just one click connect to all these different tools, and then you give just the Composio connector to your Hermes agent, and it has access to all of it.
17:00So let's see here. So now we have the Hermes Soul.
17:05I'm gonna connect to this computer, copy this, spin up a new terminal here, SSH into that, run Hermes, and here's GPT 5.6 sol.
17:19I'll tell it to open up Chrome to make sure it's working, and we'll let it cook on that.
17:27Mhmm. So back here to Dewey, we have it connected to all these different tools and connectors and you can see here, I'll just kinda talk real quick to what it's connected to.
17:40I won't go through how to set it all up because, um, honestly, could just tell your agent, I wanna set up with all this stuff and it'll it'll be really easy. It'll it'll help you do that.
17:48But I use agent card, agent mail, agent phone, and Composeo. And now I've met with all these people.
17:55I I live here, uh, in the bay, and we all get together. We have dinner, you know, come by offices.
18:02And so these are all incredible startups, incredible incredible people. So I really like the teams, but I also just really love their products. Um, and I like to think that we're the agent mafia.
18:15And the the tools the tools are very, um, you know, the same way we use Gmail for us, our our agents need their own email. Same way I might use a Twilio for spinning up a phone number for somebody on our our team, I'm gonna use agent phone to give my agent its phone, so on and so forth. I have my Idea Browser connector here.
18:36That's gonna be super useful for coming up with startup ideas and and building them out with taste. I use this tool called Latitude. This is an observability kind of tool that allows me to see when my agent is failing, or it it'll it'll detect when I'm frustrated talking to my agent.
18:55And then it'll give me, uh, like, a kind of a signal to say, hey. We should add this capability or fix this use case for your agent. And Latitude is is great for that.
19:06It's like an observability thing. And then I also have it connected to, like, things like Linear, Orgo. Oh, have you used the x API, the XMCP?
19:18I I haven't but heard good things. Tell tell me tell me about it. The the XMCP
19:23is I think it's pretty recent. I think they came out with it a couple weeks ago maybe. Yeah.
19:28I used it before jumping on the show just now with you. I was like, hey. What should what should me and Greg what's a good viral Twitter?
19:37What's going, you know, viral on Twitter right now? What's some good content ideas that we can jam on? And so it's really good at picking up trends.
19:45And then also just looking at your bookmarks, looking at if you wanted to give it access to your direct messages. Fun story. I I had this person reach out to me, and he had a big following in kind of, like, the personal finance space.
19:57And we were just DMing back and forth on on x. And he's like, yeah. I would like to jump on a call.
20:02I'd like to build out an agent for, like, personal finance. And I just told Dewey to create a Hermes agent for his use case based off of the conversation I was having with him on on on Twitter, and I didn't have to paste anything. It just read my DMs with him, read the context, built the computer, built the agent, and he's been texting it now and and playing with it the last two days with the Grok 4.5, which he's spoiled because that's his first experience of Hermes is is Grok 4.5.
20:29And, uh, he's, like, texting me like, oh my god, dude. Like, this is insane. I've never had anything like this is like, wow.
20:35And it's true. It's insane. And then lastly, vidIQ.
20:39Have you used vidIQ?
20:41I'm actually an investor in vidIQ. Wow. Yeah.
20:45I invested in VidIQ in 2013. Wow. Yeah.
20:52Yeah. Fun story for a separate pod the ups and downs and and and of that business.
20:59Almost ran out of money, ended up, you know, kind of pivoting. But I do I use vidIQ. If I even wasn't an investor, like I'm a daily active user of vidIQ just because I create content on YouTube sometimes, as you know.
21:11Yep. And it's helpful to to to see tags,
21:16to see how things are performing, stuff like that. I love it. Yeah.
21:20It's it's a great product. I I discovered it for it it tells you what videos are outliers. So for people that don't know, vidIQ, you can download, like, the extension.
21:29You can see I have it here on the top right corner. Um, and when you're on YouTube, it'll tell you what videos are like outlier videos. And so for my Hermes agent Dewey, that's useful because I can ask him like, hey, what are the top outliers a for my channel, but also for other, like, people in my niche?
21:47And then I can kinda, like, spin off of that as a inspiration to make just infinite content ideas. And once again, 4.5, really, really good at this.
21:59So, yeah, that's that's pretty much my stack right here. Long story short, give it access to every tool that you can imagine being useful for you, and don't be shy about it.
22:11And then, yeah, now we can kinda, like, get into sort of the the use cases and and and build out process of what it's capable of of doing now.
22:21Let's do it.
22:22So here I have a a few terminals. So this first terminal is Dewey, and obviously he spun up these computers here and he he's telling that for us.
22:34But now what I wanna do is I just wanna start a new session here. I'll do slash new, and I'll say, I want to build let's treat them like a cofounder, like, genuinely.
22:45Let's say I wanna build a new startup idea. I have the I have the, uh, idea browser and MCP.
22:54I have I have the XMCP, and I have other search tools for grounding your response in the right context.
23:06Come up with, like, maybe 10 startup ideas using all these tools that we could build out today.
23:17And so I'm just gonna send this off to Grok, and and I I I will even show just like here, like, I know sometimes maybe watching me type and and and enter off a prompt into my agent, you know, maybe that's boring for some people, but look look at how the the use of the tools, how explorative it is.
23:36This is really my point, and it's that it's gonna use every tool available to it. And so when you're when you're wanting to experience the powerfulness, the the the potential of a of a model, this is really how you you let it shine is is giving you access to everything and seeing how it explores that and how resourceful it is and how it ties it all together.
24:03So I haven't talked about Idea Browser or MCP on this podcast, really. But what it is you know, people know I'm a cofounder in Idea Browser. I haven't about it, but it's probably one of the most slept on MCPs that exist.
24:18It basically just levels up your LLM so that whatever idea you build, it has what people call taste. It has because it's trained on, you know, thousands of startup ideas, data points, you know, landing pages that work, landing pages that don't work, apps that work, and stuff like that.
24:38So that was the idea around the idea of Browser MCP for people interested.
24:42I love it. It's like it's actually it's honestly, like, my main way of how I interface with Idea Browser just because with every tool that I use, like, the interfaces are for for me, all of the interfaces are kinda disappearing even for, like Yeah. Using Orgo.
24:57Like, I just text my agent, and my agent's connected to all the tools. And and so if if you guys haven't heard, like, that's this big saying going around that, like, SaaS is when people say SaaS is dead, software as a service is dead, or they say interfaces are dead, They're not dead. Like, I'm paying for more software than I've ever paid for in my life because of agents.
25:18It's just that the way in which we interact with them and the way in which we interface with them is just changing. And, yeah, it's kind of coming down to just texting your agent.
25:30And I think in the next couple months, it's going to be a lot of advancements with voice use cases. I think we've yet to have the Clawdbot, OpenClaw moment for voice.
25:41And when we do, think it'll be beautiful because we're gonna be able to just go out on a walk all day and talk to our agents, and they're gonna be doing stuff for us in the cloud on Orgo or whatever.
25:52And, I mean, I think that's just like that's so cool to me. So so, yeah, we have I we have IdeaBrowser being used here.
26:02And and, yeah, Grok is just going crazy. He's just going through all these tools.
26:08So while he's doing that, he this is this is Dewey that's doing all of this for kind of searching for these startup ideas. While Dewey's doing that, I can come over here to the the Hermes that's running the GPT 5.6 Sol, and I'll show that that terminal here.
26:30You can see that we had him open up Chrome just as a test to see if it's working. And so what I wanna do is we have Hermes running Sol here, and we have Grok 4.5 running Hermes here, and these are both fresh Hermes instances.
26:44Uh, and I would like to compare them in terms of how they could perform at, like, maybe a similar task so people can see for themselves, yeah, just how they stack up against each other.
26:57What's a good use case that we should do, Greg, of, like, just a simple task that might be simple enough for both of them to give a give a shot at?
27:06Maybe doing a a one pager like a you know, say we're we're building a startup idea, but we we we want a a one pager, you know, business.
27:18Almost like a mini deck, right, in in a one page.
27:21Yeah. Yeah. Let's just do then let's say, like, let's build a simple one pager landing page for a startup idea.
27:28Yeah. And just use use and I'll I'll say open it up in your browser so I can see it.
27:38And I'm gonna copy this prompt, and we're gonna send it off to the Grok 4.5 and the Sol GPT 5.6 Sol.
27:48We're gonna send it off to them at the same time. So I'm gonna paste that here.
27:53I'm gonna hit enter. I'm gonna go back here, hit enter. And so now let's see in Orgo.
27:59I'll go from a kind of a more project view here, and let's see what they build. And who knows?
28:06Maybe maybe Sol will be faster. You know? OpenAI, they did improve the speed at which GPT 5.6 Sol and Terra and Luna, those are the new names of their models.
28:19They they really improved the speed a lot, but I still think that Grok just the the edge is in the fact that it's yes.
28:28It's faster, but it's more cost efficient. Like, you'll you'll run out pretty quickly now with with the new codex models. They're very capable, but for whatever reason, they're a little more token hungry.
28:40And that's one thing about Grok is it's it's very efficient with its tokens. Mhmm. Okay.
28:47Yeah. This one yeah.
28:49Grok Grok finished already. You can see that. Wow.
28:53And, uh, soul is is not done yet. But
28:57That was crazy. That was, like, what? Like, forty seconds?
29:00Yeah.
29:01For a full website. And look. It's, good design.
29:04Meetings that write themselves into momentum.
29:06Yeah. Let's see. That is beautiful.
29:09I mean, part of that is the idea of Razer MCP and stuff like that. But the fact that it was able to do that so quickly is remarkable.
29:18So insane. Yeah.
29:20And Seoul is still cooking, but let's is it done? Built and opened. Okay.
29:26It says it's up. See? This is what I mean.
29:28I don't know. It says it's open. I don't see it open.
29:31Okay. Here it is. Okay.
29:33That's pretty nice. Wow. That's pretty nice.
29:36That's pretty nice. Yeah. So you you can't go wrong with these two new models.
29:40I mean, they're they're getting so good. But
29:44I I you know, based on these two landing pages, I I will say I I do prefer the Grok 4.5 landing page from a design and copy perspective, but they're both they're both pretty excellent.
29:59Yeah. I mean, how crazy is that that in under a minute, we just built like, two years ago, this would be like people would be, like, scared.
30:11Like, this is like magic.
30:13Yeah. And it's funny because we're we're, like, we're all getting a little numb to it. Right?
30:20Like so I think, for me, this moment is like, I I I've I'm I'm I'm not numb anymore.
30:30Like, I I'm seeing you know, I was numb for a while, but, like, looking at this, I'm like, this isn't it's it's crazy how quick it is because if you can create these loops for your business, you could really, really be making a lot of progress daily.
30:44Yeah. Oh my gosh. It's what a time to be alive.
30:48What a time to be alive. That's for sure. Yeah.
30:52So let's see. The the idea browser okay.
30:56So let's see. So Dewey said, alright. He checked XAI.
31:00He checked Idea Browser. He checked Composeo using the XSearch tool and the web search and Perplexity Search.
31:07These are all the tools I have connected in Composeo. And so then using all of these market signals, like, truly scraping everything, uh, here's top 10 startup ideas that you can start building today.
31:19Trade call, a voice receptionist that books failed service jobs. So it's like a voice agent for plumbers, HVAC, electricians. Okay.
31:27I like that. Next one, MCP gateway strictures, access control.
31:33Audit for agent MCP servers. It scans MCP servers and is like a security scanner. Interesting.
31:41Uh, a managed AI employee operating system for agencies using Orgo and Hermes but productizing it.
31:48Okay. This is actually a it was a really good idea. This is this you could you could build you could build, like, a template for a specific vertical, like like HVAC, for instance, on Orgo of okay.
32:01Here's the Hermes agent with the right prompts and skills and tools and connectors out of the box, spins up with one click with a template that you can build, and you can go after a whole vertical and just, like, instead of being trying to be like a perplexity computer for 200 a month who's trying to service a whole prosumer market, you can go to HVAC companies for 5 k a month and serve them really, really well, And they'll happily pay that money because you've solved all the problems for them upfront, whereas Perplexity Computer doesn't do that for them upfront.
32:36And, yeah, you could do that with a managed AI employee kind of operating system. Genius.
32:42By the way, if anyone's interested in that idea, I'm doing a whole sixty minute free podcast episodes coming out in the next two weeks explaining how to actually build that idea.
32:53And so, you know, like, comment, and subscribe for more of this in your feed.
32:58Love that. Yeah. Yeah.
33:00That's gonna be good. I mean, the I think that this is like the AI agency space for, like, creating these managed employees.
33:08It reminds me so much of marketing agencies in, like, 2013, 2014 who learned Facebook ads, Google paid ads. And, I mean, it was just, like, such a great opportunity to be early on something like that.
33:21This is happening all over again, 2026, with managing AI employees for businesses and making, like, an an agency out of it.
33:30So a lot of good ideas here. Let's see if it kind of consolidated onto one. Okay.
33:35Oh, it did. It said if if the goal is cash and a demo and a and to stack leverage and not just a pure venture kind of theater, like go raise money, something that if you wanted to start tomorrow, bootstrapped and start making money, it said the best one is a managed AI employee on Orgo and Hermes, and that's that's the first one I made.
34:01Okay. So very good.
34:03Let's trust it. Should we trust it?
34:05Let's.
34:06So then let's say for a next step, let's pick one of the top three. Okay. Let's do the let's do the managed AI employee top option.
34:20What's the next step? Perhaps a landing page.
34:26Can you open it up in your browser after you build it? So we obviously had these quick recap.
34:37We had Hermes running sole GPT 5.6, kinda create a website for us just to demonstrate.
34:44We had the same on a fresh instance of Hermes Grok 4.5 to demonstrate its landing page capabilities. Now I'm gonna have Dewey who has all my context and more kind of understanding of all my tools and me and everything.
34:57We're gonna have him do the same for this AI employee business, and let's see kinda the website that he spins up. And then and then I can show some really cool things too of, like, okay. You build a website for this AI employee managed agency business.
35:14What do you what else do you need? Well, you need customers. You need to find a way to get customers.
35:18How do you do that? I recommend content. Okay.
35:21How do I make content? Well, let's ask Dewey to do what he just did for finding a viral startup idea.
35:29Let's do that for viral YouTube content ideas. Can you make the transcript? Can you make the thumbnail?
35:35Can you make the can you publish it for me? Can I just text you on Telegram and have you publish it?
35:41And, yes, by the way, it can do all these things. So I just think like, wow.
35:46What what a time. So let's see here. Okay.
35:51So he's using some skills here. You see, he's using the popular web design skills, clog design skills. Oh my gosh.
36:01It's calling it desk crew. Very cool. Okay.
36:04We'll hang tight on this. Mhmm. So while he's cooking on that, I kinda talked through some of the tools here, some of the stack in terms of trusting your AI cofounder running Grok 4.5.
36:20So when you're working with these agents, you see how I'm kind of, like, on one session right now.
36:27I'm just using a terminal that's connected to this Orgo computer, and I'm going back and forth with it.
36:33But like you and I were talking about, you don't have to just stop there.
36:39You can spin up multiple of these and have multiple sessions going on at the same time with Dewey or your Hermes agent.
36:48And and and and when you do that, you can obviously, like, do so much more at the same time, and you're not restricted to just one one session. And so the I kinda just like using the terminal because of that.
37:01I can spin up, like, five different terminals. So I could have one of one terminal of Dewey working on the landing page, the next terminal of Dewey working on finding those content ideas we talked about, the next terminal of Dewey creating the thumbnail, so on and so forth.
37:17So he's cooking here. I think he's just, like, doing something with his display. Let's see.
37:25I'm gonna say, Dewey, man. I don't see the computer display.
37:33Did you break it? Please fix it. And he will.
37:38Trust me. This happens sometimes. These agents, they'll try they'll they'll almost break themselves sometimes, but they also know how to fix themselves too, which is kind of fascinating.
37:47So I'll tell them that.
37:49Dewey, man. Dewey, man. I'm disappointed in you.
37:52Dewey, man. Yeah.
37:54So let's let him cook on that. I wanna show too, like let me let me pull up here.
38:03I'm gonna pull up my telegram Dewey, and maybe I'll show like this. Okay.
38:11So here I have also Telegram just to show you, like, obviously, when you talk to your Hermes agent, yeah, you could talk to it through your terminal. You could talk to it on Telegram.
38:21I even have Dewey on iMessage, and so I can text Dewey. And so here is, like, a term I mean, a a Telegram session with Dewey.
38:30I can hit slash new. And I like Telegram for, like, a lot of exchanging of videos or images with Dewey.
38:40This is super easy to, like, drag and drop in an image and send it to Dewey and have him send me something back. But I just wanna say, like, alright.
38:48Can you create a viral thumbnail for YouTube for a managed AI employee business that I'm starting?
39:01Use my face based on my YouTube content and make sure to use inspiration from vidIQ connector
39:12Mhmm. For
39:15the thumbnail.
39:16I'll just send this off. So what that's doing is it's looking at outlier YouTube videos thumbnails, and it's basically saying, okay, this is a nice thumbnail.
39:29I think it can work in this niche with this type of content, and hopefully it creates a good thumbnail. Right? Exactly.
39:35Exactly. And finds the outliers, maps everything together, by the way, like, once again, emphasizing the point of just connecting all your tools because it has access to vidIQ, it can find the outliers.
39:51And then because it has access to Composeo and Composeo is connected to my YouTube, it knows to look at my YouTube, and it's just like context is king.
40:01Give the agent all the context it needs. It's gonna piece it together beautifully, and it'll be able to give you something tailored just for you.
40:11So it's cooking there. Let's go back to Dewey's screen. You can see he fixed himself, and he made a website here.
40:18And let's see. Let's go ahead. Let's take a look.
40:20Hire an AI employee with a real desk, not another chatbot. And so he built this whole website.
40:30Okay. Okay. Okay.
40:34And, yeah, here we go. Wow. Here he goes.
40:36He even included the pricing. So, like, a pilot for fourteen days, $1,500 one time, 2,500 a month for a managed seat, 6,500 a month for three seats, book a pilot, even has, like, the the lead qualification process here.
40:55I mean Super smart. Yeah.
40:58It's insane. So, like, when I say, you know, have Grok four point five and Hermes be your cofounder, like, I I do mean it.
41:07You know? This is this is awesome. Okay.
41:10So let's go back. Let's see in Telegram. It's cooking here.
41:14It's saying, okay. It found it found some some good content ideas it's gonna recommend.
41:19Um, so we'll let it cook on that. But let's go back here to Dewey. He built the the landing page.
41:26What and now I'm gonna ask him, use the Idea Browser MCP to tell me what to do next. Because the Idea Browser MCP, it also not only does it tell you ideas, but people don't realize you could use it actually to help you build beyond that. Like, okay.
41:43Yeah. I have the idea for a start up, but what what happens after I build a website?
41:47Well, I need to build the the messaging or the lead magnet or the offer, and it has access to be able to do all of that.
41:55So I'll just ask it to do that.
41:59And by the way, that's, like, sort of the bigger trend that we're seeing now with the Internet is these LLMs are are obviously amazing and but they they do lack, like, the vertical specific context and stuff like that.
42:15So you're seeing MCP's, you know, sort of you you have the mob in MCP for designs.
42:21You're seeing the Idea Browser for business building MCP. You're seeing just MCP after MCP.
42:27And that's, by the way, a whole opportunity for startup ideas. Like, what's what what do what do we need context in that you could be building an MCP in?
42:38Exactly. Yeah. I think you made, like, a awesome post about, like, building startups for agents.
42:46Obviously, like, something we're trying to do is build these computers for agents. And then, like, you see all these tools that we're using for our agent, like agent mail, agent phone. Like Yeah.
42:55This is just a such a cool opportunity of if you're someone who wants to build something like, yeah, something maybe more kinda like like maybe venture backable if that's if that's your kind of if that's your thing.
43:12I mean, you can go after some really ambitious ideas and just build them for agents. I think MaltBook was a good example. When Claude Bot came out, someone created MoltBook, which is like a social media for these Open Claw agents, and they got acquired by Meta within, like, a week.
43:28So, yeah, anything that we find valuable as humans, these there's probably a startup idea for it, uh, for agents. So okay.
43:37So Dewey says, I use the Idea Browser MCP, and it says, um, it did this inside of Idea Browser.
43:45Oh, so it created a project URL inside of Idea Browser, and it submitted the idea.
43:53It did a full idea research. Oh, wow. So it's, oh, it's actually telling me the different things that it can do inside of Idea Browser.
44:01Um, it's telling me my fit for this idea based off of my profile, profile, which I love. And so it's like, okay. Look.
44:09You already have a landing page, and the idea of a browser's skill graph still says that you need to crystallize your offer and prove delivery and then scale the funnel.
44:19So it's like, first things first, let's pick one employee role for the first pilot today. Do not sell AI employees in general. Sell one weekly artifact.
44:30I love this. This is this is really cool. So best first role for Nick, my call, is to do a client reporting employee for agencies.
44:42So it's saying, like, weekly visible deliverable, live demo demoable on Orgo, maps to agency pane, so on and so forth.
44:52So I guess, like, let's just copy that. Mhmm. And let's tell it okay.
45:00Let's tell it. I'll paste that here. Let's do this.
45:05Let's build the reporting demo on Orgo headed so that we could see it visually and proceed with everything we need to do.
45:19I'll just tell that. So we'll be able to kinda, like, watch it cook here. And then let's go back to Telegram.
45:28Look at this. That's fine.
45:31I I didn't tell Dewey. I didn't give him an image or anything.
45:35Yeah.
45:37Like, this is what I mean. Like, this is all Grok 4.5, by the way.
45:41Like, give give it access to all the different tools you're using, and it will be so resourceful in how it pieces that together. And this is a good this is a really good thumbnail.
45:55And he gave me a few of them. He gave me a few options. Okay.
45:59Let's see what it says here. Here's a viral style thumbnail pack for your managed AI employee offer. It's my face, my existing channel style, and then it looked at outlier videos in my niche, And then it used some design rules from vidIQ for, like, how to make a good thumbnail.
46:20I mean, this is so cool. This is awesome. This is really good.
46:24Okay. So then what should we ask you next?
46:27We have a we have a a video thumbnail. We have it building our website, building out the offer, building out the agent that we're gonna sell to businesses.
46:38Maybe we should ask it let's just get crazy. Let's start a new session, and let's ask it just to show to everybody, like, how far you could take this.
46:48Let's come up with a email outreach campaign.
46:56I want to do a managed AI employee business, you know, and I need to do a cold email outreach to customers.
47:10Can you come up with the entire outreach email sequence and put it together in a Google Doc for me to take a look at and approve.
47:29And that's the thing that people miss is a lot of people are using these models just for the building piece. But using these models, like 4.5 for as like a marketing cofounder too.
47:41Right? As someone who's gonna help you actually get customers to the thing you're building is equally, if not more important, than if your landing page is gonna look really nice.
47:53Right. Like the per imagine the person who's arbitraging this right now for like, okay, yeah, posting content, but also gathering leads, reaching out to them relentlessly.
48:05Yes. Like, giving them like, what did what did it build here?
48:09Let's see let's see what what Dewey said here. He's already done with this. Uh, market insight is ready.
48:15Pulling it, attaching the research, and focusing Chrome on the report. So he made a market insight.
48:24Wow. Oh, wow. This is crazy.
48:30He's just he's just he's just building oh my gosh. Like, relentlessly building all of these different things.
48:36So this is a market insight on home services, and it's kind of giving insights, I'm assuming, just to demonstrate to your clients of, you know, kind of the return that they would get with the services.
48:50So very cool. And, yeah, he's still he's just cooking on this.
48:56And let's see what he says. So landing landing updated to pilot locked copy with c live report demo call to action.
49:04So then it gives me market signals that reinforce this pilot.
49:11It gives me the whole status, next steps, market insights, full idea research is in progress. It's gonna attach that when it's done. And here's what it wants me to do next.
49:21Watch the report. Confirm I see it. Book three design partner calls this week using outreach.md, which is something that Idea Browser helped it make in a whole Outreach system.
49:33And then on the call, run the four minute script in pilot.md, and then do three to eight pilots. And then when the Idea Browser research finishes, I'll attach it, and I'll run a competitive analysis and money model.
49:49And so sure enough, Dewey here is like, do you want me to personalize the five outreach emails, which we have it we have it doing here in in Telegram, but I'll just tell it I'll tell it here as well.
49:59Yeah. Let's do it. Proceed.
50:03Do you see how fast this is, by the way? Like, this is a genuinely different experience with the Grok 4.5 because, like, with GPT 5.5, literally just a month ago, it would it would take, like, sometimes, like, thirty minutes for a task.
50:14It was very frustrating for a lot of even our clients of, like, why is it taking so long? And so, like, this is just, like, so much faster.
50:23Mhmm. Yeah. It's like bang, bang, bang, bang, bang.
50:26Like, things are happening. Things are moving.
50:29. It's like yeah. It's just it's just and and when you use these different models, you do get a feel of how they drive.
50:39And Yeah.
50:40I I think that GROQ is really just a a great balance of of all of that. So would you say would you say we went from, like, a bicycle to a Ferrari, or would you say we went from, like, ?
50:52Oh, we definitely went from a toy yeah. I would say, like, a a Toyota Camry to a Ferrari because, like, you know, we've had okay.
51:02For a few months now, these models have been pretty damn good at, like, Hermes and Open Claw and just, okay, using tools, but they would just take forever, eat a lot of tokens, be super expensive. So it'd be like an expensive Toyota, and now it's a cheap Ferrari.
51:17Mhmm. Like, what a what a dream outcome. So and, guys, also, this is another thing too is, like, this is only gonna get cheaper and cheaper, faster and faster, smarter and smarter.
51:28So what that means for you is, like, the sooner you can start taking advantage of it and just implementing everything you do like, just putting in the work right now to build your agent and get that stack and just configure it to your, you know, use case and needs means that tomorrow overnight, if Grok comes out with 4.7 or Grok five point o or whatever, you get an instant immediate value add overnight to your business, to your agent.
52:00You didn't have to do anything. You're just . So I just, like, encourage everyone, like, get set up with these agents and just ride the wave.
52:09There's a huge tailwind. Alright. So we have Dewey here.
52:14He made a full cold sequence cold email outreach sequence. He gives me the Google Doc.
52:20I open it up. Cold email outreach sequence. ICP offer, the whole sequence architecture, full email sequence.
52:34So this looks great. It even wow. This looks awesome.
52:39So when I come here, I come back to the Dewey that I'm in the terminal with, it is making this email outreach campaign.
52:50It has all this stuff here. I can just give it this link to the to the dock it sent me from Telegram and tell it I would like to do this.
52:58Let's do let's do this one. I'll just give it that link and let it take a look at that.
53:04But, yeah, this is this is insane. And so it was able to give me a link in my Google Docs.
53:11This is in my it created this in my Google Docs because why I connected it to Google Docs using something like Composeo, and it had access to that. So awesome.
53:21Awesome. Awesome. But I think that's like a good enough example here for for most people to kinda get the gist of like, guys, you can literally have it build your website, build your offer, build your email outreach, give it an email via agent mail and have it do the email outreach.
53:38Give it a a phone number and have it call on a lead. Like, it's insane.
53:44Um, and you could do it all today. I have how much how much my my Grok subscription is?
53:50I still have 19% left. I started at 25% at the beginning of this call, um, and we're doing so many different things. It's like intelligence is available on the tap for very cheap.
54:02Uh, for $200 a month, you could have an AI employee cofounder workhorse just like doing everything for you.
54:10So I'm excited to see, you know, what everyone is able to do with that. There
54:17you have it. 4.5, it's a big deal.
54:20Nick, thank you so much for coming, sharing how to use 4.5 like an AI cofounder.
54:28We'll include links in the show notes, in the description for some of the some of the things we spoke about. I know we talked about a lot of things.
54:35We'll include that there. Nick, is there anything you wanna leave people with?
54:40I think just, guys, we we live in a world that is literally changing every single day with the new tools that are available to us and the things that we're able to accomplish and the things we're able to do. And it's more clear to me than ever that the people who can just be the idea guy, have the ambition and the dream that you want to have this outcome, you want to build this thing, you want to do something, and you can do it with AI, and that gap between idea to implementation is shrinking down to nothing, and you can take so much advantage of that today.
55:17There's literally like, there's nothing there's nothing that's stopping you from there's nothing that's stopping you from taking advantage of that and and and putting that to use for you and your and your business and and what you wanna accomplish.
55:30So, yeah, just use these tools, learn them, be creative, come up with cool ideas, and just bring them into reality because it's possible today to do that instantly.
55:44So
55:45Amen, brother Nick. Thank you so much. Everyone, go build some stuff, turn off YouTube, turn off this podcast, and just get get your hands dirty.
55:56I'll see you on the next one. Thank you very much. Thank you, guys.
The Hook
The bait, then the rug-pull.
Nick Vasilescu opens with a confession: he's converted to Grok 4.5. What follows is a live demonstration of why — an agent that goes from idea to landing page to cold outreach in a single, unscripted session.
How they asked for the click.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
