A working developer walks through why two popular self-hosted agent platforms kept breaking, then rebuilds his personal agent on a file-based framework and lands on a bigger lesson about tools versus frameworks.
Posted
2 days ago
Duration
Format
Review
educational
Views
18.9K
427 likes
57 · 43
Big Idea
The argument in one line.
Self-hosted agent platforms like OpenClaw and Hermes break on nearly every update, so this creator switched to the Eve framework, and the real lesson is that an agent's tool set, not its underlying framework, is what makes it powerful.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You're actively running a self-hosted personal AI agent (OpenClaw, Hermes, or similar) and it keeps breaking on updates.
You're deciding between a bare-bones SDK, a framework, and a fully-hosted platform for building your own agent and want a concrete way to place each option.
You want to see a working example of a file-based agent architecture (agent.ts, instructions.md, channels, connections) before committing to one.
You're weighing whether to build a reusable tool layer (memory, email, computer access) instead of rebuilding it per agent framework.
SKIP IF…
You've never built or run a personal AI agent and want a beginner explainer of what an agent even is; this assumes that context.
You want a step-by-step coding tutorial; this is a comparative walkthrough and demo, not a line-by-line build guide.
TL;DR
The full version, fast.
OpenClaw and Hermes kept breaking on nearly every update, so the creator rebuilt his personal AI agent on Eve, a framework that sits between the bare-bones AI SDK and fully-hosted platforms like Grokbot. Eve defines an agent entirely through files, an agent.ts config, an instructions.md system prompt, and folders for channels and connections, which made it easy to extend and cut his rebuild to about a week. He proved Eve isn't locked to one host by running it fully on Convex and fully on Cloudflare. His real conclusion: the framework matters less than the tool set behind it, so he built one reusable MCP server for memory, email, and computer access instead of resetting up tools for every new agent.
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.
States the video's roadmap: why he dropped OpenClaw and Hermes, why he rebuilt on Eve, his Eve experiments, problems with Grokbot, the tools that make an agent powerful, and why the best personal agent hasn't shipped yet.
00:49 – 02:45
02 · OpenClaw and Hermes keep breaking
Walks through a viral OpenClaw 2.0 complaint thread, then argues both platforms are clunky and unstable, with Hermes only marginally better.
02:45 – 04:51
03 · The agent-building spectrum: AI SDK to Grokbot
Lays out a four-tier spectrum from bare-bones AI SDK, to the Eve framework, to opinionated self-hosted platforms, to fully hosted Grokbot.
04:51 – 06:55
04 · Sponsor break: Mobbin MCP
Sponsored segment showing how feeding an agent real UI/UX reference examples through Mobbin's MCP fixed a poorly designed onboarding flow.
06:55 – 08:16
05 · Placing OpenClaw, Hermes, and Grokbot on the spectrum
Details the tradeoffs of self-hosted opinionated platforms versus fully hosted Grokbot, and confirms Eve is his pick for full control without low-level plumbing.
08:16 – 10:56
06 · How Eve's file-based agent architecture works
Shows Eve's file-based structure: agent.ts, instructions.md, and folders for channels, connections, and sub-agents, and why that structure sped up his rebuild.
10:56 – 13:11
07 · Pluto v2 demo: Percy handles a storm scare
Demos his Eve-built Pluto agent and its sub-agent Percy, which used computer access to research storm-preparedness products on Amazon during a real power outage.
13:11 – 16:09
08 · Running Eve on Convex and Cloudflare
Shows two experiments proving Eve isn't locked to one host: Adam, an open-source Eve agent running fully on Convex, and Ezekiel, an unreleased build running fully on Cloudflare.
16:09 – 18:21
09 · Bezalel: the tool layer that makes any agent powerful
Argues the tool set matters more than the framework, and introduces Bezalel, a single MCP server bundling memory, email, payments, a computer, and connectors.
18:21 – 20:41
10 · The best personal agent hasn't shipped yet
Closes with a live voice demo of his agent Ruth, and argues chat may not be the ideal interface since the field is still discovering how people want to work with agents.
Atomic Insights
Lines worth screenshotting.
OpenClaw and Hermes are self-hosted, opinionated agent platforms that reliably break with almost every update.
AI SDK is the bare-bones library layer for building agents; Eve is a framework built on top of it, the way Next.js sits on top of React.
Frameworks like Eve trade some flexibility for built-in plumbing: sub-agents, durability, and streaming come pre-wired instead of hand-built.
Grokbot sits at the far 'super opinionated' end of the spectrum: fully hosted, zero setup, but you live inside its fixed bot-and-group-chat architecture.
Eve defines an agent entirely through files: agent.ts for config, instructions.md as the system prompt, and folders for channels, connections, and sub-agents.
A file-based agent architecture isn't just readable for humans, it's easier for an AI coding assistant to extend, which cut one rebuild down to about a week.
Eve isn't locked to one host: the same framework has been deployed fully on Convex, as the open-source 'Adam' project, and fully on Cloudflare Workers.
Giving an agent computer-use access to run a real purchase-research task is a way to stress-test whether it can self-heal after a failure, not just complete happy-path prompts.
The single biggest lever for agent power isn't the framework, it's the tool set: memory, email, payments, a sandbox, and connectors bundled behind one MCP server.
A tool layer built once (memory, email, computer, connectors) can be reused across every new agent framework, so switching frameworks stops requiring a full re-setup.
Chat may not be the right long-term interface for personal agents: voice and multi-threaded conversations already feel more natural for some tasks.
Designing agent UI without reference examples produces generically bad interfaces; feeding a design-pattern MCP concrete UI/UX examples fixed an onboarding flow an AI had originally designed badly.
Takeaway
Why the tools matter more than the framework
TOOL STACK LESSONS
Across a dozen personal-agent experiments, the deciding factor wasn't which framework or platform he used, it was whether the underlying tool set was reusable and reliable.
02OpenClaw and Hermes keep breaking
Self-hosted, open-source agent platforms that ship frequent updates carry real breakage risk: nearly every update introduced a new bug for this user.
A platform being open source and self-hosted means you also inherit its setup and maintenance burden, not just its flexibility.
Preloaded 'skills' or opinions baked into a platform can actively get in the way once your use case doesn't match what the platform assumed.
03The agent-building spectrum: AI SDK to Grokbot
Agent tooling sits on a spectrum from bare SDKs (full control, all plumbing on you) to fully hosted platforms (zero control, everything decided for you).
A framework built on top of a bare SDK, the way Next.js sits on React, pre-wires common needs like sub-agents, durability, and streaming.
Picking a framework over a raw SDK trades a small amount of control for a large amount of saved plumbing work.
05Placing OpenClaw, Hermes, and Grokbot on the spectrum
Fully hosted, opinionated platforms remove setup work entirely but lock you into their fixed interaction model, like bots-only group chats.
Open-source self-hosted platforms give you full customization rights, but every configuration change is your responsibility to build and maintain.
Coding-focused agent tools tend to cluster toward the opinionated end of the spectrum because their use case is narrower than a general personal agent.
06How Eve's file-based agent architecture works
Defining an agent purely through files, a config file, a markdown system prompt, and folders for channels and connections, makes the structure legible to both humans and AI coding tools.
A predictable file layout can meaningfully speed up how fast an AI coding assistant can extend or rebuild a piece of software.
Markdown instruction files work well as a system prompt format because they're easy for both a person and a model to edit and reason about.
07Pluto v2 demo: Percy handles a storm scare
Giving an agent real computer-use access and asking it to complete an actual research-and-purchase task is a stronger test than a scripted demo.
Testing whether an agent can self-heal after a tool failure reveals more about its reliability than testing only the happy path.
An agent with its own dedicated email and payment card can complete a task end-to-end without the owner acting as a manual relay.
08Running Eve on Convex and Cloudflare
A framework being opinionated doesn't have to mean vendor lock-in: the same agent framework ran unmodified on two different backend platforms.
Proving portability across hosts before committing removes a major objection to adopting a new framework.
An unoptimized first deployment being slow doesn't disqualify a framework, since basic performance tuning is usually a separate, later step.
09Bezalel: the tool layer that makes any agent powerful
The tool set behind an agent, memory, email, payments, a sandbox, connectors, matters more to its usefulness than which framework built it.
Building tools once behind a single reusable interface avoids re-configuring the same integrations every time you switch or test a new agent.
A tool layer that outlives any single framework protects your setup investment from becoming obsolete when the next agent platform launches.
10The best personal agent hasn't shipped yet
Chat may not be the ideal long-term interface for a personal agent; voice and multi-threaded conversations already feel more natural for some tasks.
Letting an agent hold multiple separate conversation threads with the same underlying model can matter more to usability than the model's raw capability.
The interaction model for personal agents is still unsettled, so betting entirely on one interface pattern carries real risk.
Glossary
Terms worth knowing.
AI SDK
A low-level software library most AI agent products use to build agent loops and features from scratch, giving full control but requiring you to build everything yourself.
Eve
An open-source agent framework built on top of AI SDK that pre-wires sub-agents, durability, and streaming so a developer only has to configure how the agent should behave.
MCP (Model Context Protocol)
A standard connector protocol that lets an AI agent call an external tool or data source, such as a design-reference library or a memory store.
OpenClaw
A self-hosted, open-source personal AI agent platform with built-in memory, integrations, and self-learning skills, criticized in this video for breaking frequently on updates.
Hermes
A self-hosted personal AI agent platform similar to OpenClaw, described as more stable but still carrying unnecessary preloaded features.
Grokbot
A fully hosted, opinionated personal-agent platform where you interact with bots and bot group chats through a fixed architecture, requiring no self-hosting.
Convex
A backend-as-a-service platform (database, queues, functions) that one experiment used to run an entire Eve-based agent without deploying to Vercel.
Composio
A connector service that lets an agent search for and call third-party tool integrations, such as checking email, through one unified interface.
Bezalel
A single MCP server built by the creator that bundles memory, email, payments, a computer, sandboxes, and Composio connectors so any agent framework can use the same tool set.
“OpenClaw, although what it did for the ecosystem was open our minds and show us what's possible... it's just too clunky, it doesn't work properly, it's not stable.”
blunt, quotable verdict on a popular but flaky tool→ TikTok hook↗ Tweet quote
03:29
“AISDK is the React, and the framework is Next.js. Eve is Next.js.”
instantly-graspable analogy for any web developer→ newsletter pull-quote↗ Tweet quote
16:09
“The one thing that I've come to realize is what makes your agent powerful is the set of tools you gave it.”
the thesis statement of the whole second half of the video→ IG reel cold open↗ Tweet quote
19:54
“I still think we're discovering the best ways to work and communicate with these agents. I'm not sure if chat is the best medium.”
open question that reframes the entire personal-agent debate→ newsletter pull-quote↗ Tweet quote
The Script
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.
17px
metaphorstory
I stopped using OpenClan Hermes as my personal agent. Instead, I built my own using a framework. And surprisingly, when I shared this with you all on Twitter, a lot of you asked questions.
Why? What? How?
And in today's video, we're going to talk about why I stopped using OpenClan Hermes, why I rebuilt my personal agent using Eve. the framework of choice, my Eve experimentation and results, meaning the different tests and experiments I did and the reason why I came to building my personal agent with Eve. Number four, we're going to talk about the problems with Grokbar.
It's beautiful, but I kind of have issues personally. I'm going to then share with you the tools I give my personal agents to be powerful. And finally, I'm going to share with you my personal opinion, why I think the best personal agent has not dropped yet.
Sit back, relax, get your snuggie and popcorn. Let's get into it. So first things first.
Why I stopped using OpenClaw enemies. I'm going to start with OpenClaw first. This is one of their recent tweets, September 3rd, 2026.
Let's read a couple of the responses. It was working fine until I ran this update. Yesterday's update cost me two hours.
No thanks for now. OpenClaw doesn't seem to work as per usual. OMG, every damn day with the new update.
Cannot just figure out how to build several things into a version before releasing? Done updating my bot last two days now. lol i think you get the gist open claw tends to break every update every single update i've made to open claw has always led to something breaking and not only that sometimes it straight up just doesn't even work now i want to take it a step further alex finn who's a good friend of mine a lot of people know him as the guy who just loves every single ai product and he chills every single area product but if you really met him the guy's just super excited about the time that we're in and there's a lot of great tools there and he loves them but even he was roasting open claw 2 .0 because of how much it was breaking and how some of the features were just nonsense right so open claw although what it did for the ecosystem was open our minds and show us what's possible even at a time where the models weren't that good it's just too clunky it doesn't work properly it's not stable and when it comes to Hermes Hermes is better
But everything I said about OpenClaw happens on Hermes just to a much lesser degree. Very much lesser degree, but it's still clunky, right? And some of the opinions and the features and the things that Hermes has, to me, is just useless.
Some of the skills that it comes preloaded with, unnecessary, right? So I just stopped using OpenClaw and Hermes, to be honest. There was a lot of back and forth on certain things I tried to set up on different machines on my network.
It just was not a good time. Therefore, I've made... decision to completely stop using these two now what you need to talk about is why i rebuilt my personal agent using eve but before i get into explaining what eve does i kind of wanted to explain to you a little bit of the landscape when it comes to building agents or agents in general personal agents to be frank you have aisdk which is like bare bones of bare bones, right?
This is the library that most agentic products use to build their agents with full control. It's very bare bones, but it gives you full control and full power. And then you have Eve.
which is built on top of the AISDK. EVE is a framework. For my web dev people, AISDK is the React, and the framework is Next .js.
EVE is Next .js. So with the framework, there's a lot more plumbing, right? There's a lot more things built out for me.
For example, with AISDK, if I wanted to add sub -agents to my agent, I would have to build that out. With EVE, it's built out for me. I just configure how those things work, right?
That is the benefit of a framework. When you think of web dev land, you had React, but you couldn't build... web app off react right because you need routing you need a state machine you need all these things that make a website a web app function the library doesn't give that to you the framework that's built on top of the library gives that to you so that's the relationship between ai sdk and eve and there's a lot of bare bone ones like cloud sdk open ai's sdk i'm sure cloudflare i think cloudflare does have an sdk as well that's where this lives eve is a framework built on top of the library, right?
And it does a lot of the nitty gritty heavy lifting for you where you only focus on how you want your agent to work and function, right? It handles durability for you, streaming and all that stuff that creates a good experience for your users. You just have to sort of architect how you want the agent to be.
You have an ugly baby. And I'm not talking about your child. I'm talking about your web or mobile apps, UI and UX.
Let's be honest here. Models are not really good with this. and UX without proper guidance, examples, and inspiration.
Now, if only there was a way to give agents guidance, examples, and inspiration. ding ding ding it's mobbin and their mcp now for those of you who don't know what mobbin is mobbin is the greatest collection of web and mobile apps with the best ui and ux to exist not only do they categorize it by different niches but they also categorize it by section let's say i want a fire hero section i have all these examples i have even this example right here of a site called family i love the playy cartoony feel i can give this to my agent via the mc as a reference.
They literally have every beautiful site. indexed on their site like this page right here is art now here's the thing it's not only just about ui in fact for me the big win with the mob and mcp is ux i'm building this app called bezel and this was the initial onboarding flow pretty ugly designed by ai i gave my agent access to mob and mcp and look at the difference how i add agents to bezel i click add i give it a name i create the agent it walks me through this flow i can connect my agent simply like Like all these models, the layouts, the structure, all of it was thanks to Mobin's MCP working with my agent.
If you're building serious apps, you owe it to yourself and your customers to use the Mobin MCP to restructure your layouts, your flows, your onboarding and your setup. And using it is so simple. All you have to do is click the link in the description down below.
Give your agent access to Mobin MCP. And I can't wait for you to cook apps that people will actually enjoy using with that. being said i'm glad you'll have no more ugly babies let's get back to the video but then we could take it a step further and you have something like open clone hermes that you know take care of the integrations that take care of the memory and the self -learning and the skills and all these type of things it's very opinionated but it's flexible a because it's open source you can literally download the code and make changes and not only that you have to deploy yourself now i think hermes has like a deployed version where you can you know use the dashboard i'm not completely sure i've always just ran these on my machine or you can run it on a vps or whatever right but you have to set it up yourself and there's too many opinions yet i have to set it up myself right so making changes or configuring certain things or removing certain things is a lot more difficult or open clone armies and then finally you have something like grokbot which is super opinionated you know it's hosted for you right they have a specific architecture on how they want you to use think of agency you have bots you communicate with bots bots have group chats and this is great this is great i will take grokbot over open clone armies any single day
the week now codex and cod co -work or whatever it's called i think would be around here maybe in between but more so on the super opinionated side but because those are mostly used for coding at least for me i just want to put grofbot here that being said this is sort of the spectrum from bare bones to super opinionated.
The reason why Eve is the perfect thing for me to build my personal agents is because it gives me full control over how I want my personal agent to work. And it's not super bare bones where I have to worry about how Texas streamed, right? So I have the best of both worlds where I'm fully flexible.
There's no opinions made. My opinions matter, but I don't have to worry about the low level plumbing. All of that is taking care for me.
So all that being said, I hope you understand. understand why i chose eve to build my personal agent and not only have i used to build my own personal agent the pluto agent is built on eve i stripped off open clock completely i rebuilt everything using eve and it's so much more performant that being said i think it's best to talk about how eve works the site is eve .dev and this Right here is the reason why I became bullish on Eve and start to build my agent.
The one thing I loved about Next .js is it used a file based routing. So in order to define the route of a page, you have to create a folder and under the folder, you have to create a file. With Eve, it's pretty much the same thing.
I have a template open here, one of their templates open here. It'll be easy to describe how the code looks. Now, I know some of you are going to be like, is it 2021?
Are we looking at code? Lock in. No, like seriously, lock.
in. Let's continue. So this is how an Eve agent looks.
This is what it looks like. This is how it works. You have an agent folder in your code base.
Under the agent folder, I have a couple files, agent .ts. In the agent .ts file, what's being defined here is an agent. There's a function called define agent.
I specify the model. Great. And then I have instructions .md.
These are the instructions fed into the agent. Think of this as the system prompt. Here I have the identity defined, how you write, how you work, et cetera, et cetera.
A markdown file. And then let's say I want to communicate with this agent in Slack. I create a folder called channels.
And under channels, I have Slack .ts. A TypeScript file that sets up Slack for me. And then let's say I want to connect an external service like Notion.
I just create a connections folder. And under connections, I have Notion .ts. Notion connection setup.
And let's say I want to create specific sub -agents. I just create a sub -agents folder. Define the sub -agent.
And then explain the sub -agent in code. But here's the thing. all this makes the code readable the files understandable and guess who this is great for not just me but my ai so when i was rebuilding pluto it was so easy to tell fable how i want the agent architecture to look i gave it eve and all i was reading was files and it was fantastic like when i tell you i rebuilt pluto in literally less than a week or two because the way eve was made is just fantastic.
The fact that they use the files to define the API just makes it so easy for not only a developer, but for the agent itself to understand how to build your personal agent. Now, that being said, because I'm so bullish on Eve, let me show you how Pluto V2 looks. First and foremost, like Grokbot, you can directly chat with the main agent or you can set up.
multiple different agents or you can set up a group chat i set up a bot named percy and percy is like my personal shopper now a couple days ago in toronto there was a crazy storm and like when i say crazy i'm just gonna show y 'all it was insane like this is what it looked like people's furniture flying off balconies hail like it literally looked like the second coming this is why we need to trust in jesus that being said this was a little bit terrifying and one of the reasons why i was terrified was because the power went out and when the power went out and i just have a newborn baby you know father instincts kicked in i'm like okay this cannot slide now the power came back instantly but nevertheless i panicked a little bit so i told percy hey percy there's a storm in toronto and two times the power sort of died and came back and that kind of scared me can you use your computer and find solutions for this on amazon .ca i live in a condo so i don't know if a big generator makes sense and then you use the computer right it used the computer oh the computer is watching a video of mine i promise this is not how i get my views i actually just explained myself i was just testing out some computer use stuff like i wanted to test out what happens if it failed like can it like
you know self -heal and do all that and it did and i asked it to open a youtube video and it did i should probably stop this so you guys don't think i do this to get views let's pause this right here that's possible actually just close the tab to make all accusations false that being said use the computer and it did a bunch of research and then it got me a couple products portable power station ups battery backup emergency lighting and surge protector And it gave me Amazon .ca links.
Now, here's the cool part. I could have taken this a step further. Every agent, Pluto agent has its own email.
Shout out to agent email. And then every agent has a card. I won't click on it because my credentials are there.
Shout out to agent card. So with its own email. and its own computer and its own card.
And I'll take it a step further. I can use iMessage, phone number coming soon, right? Its own memory.
This agent can really do stuff for me, right? So all these tools in combination with the specific agent architecture that I like. And the one thing, again, my issue with Grokbot is I like my threads.
I want to have different threads. I don't want to talk with the same agent only in the same chat. I want to talk to the same agent in different chats.
So all this. is my perfect personal agent this makes sense to me now i'm turning to this little product because i think there's going to be some people that agree with me and there's some people that don't maybe there's a specific way you want your agent to work and this is why eve just makes sense for it to be the framework you use to build your agents now i did a lot of experiments before i came to this conclusion and some of the experiments were as follows a lot of people were saying oh you're just going to be locked to the first cell ecosystem well on their page they say there's an example for you to self -host you can use different providers but i wanted to take a step further i wanted to see if i could host eve on convex and eve on cloudflare and both of them worked i literally told fable i want to run eve on convex and i have it it's open source it's called adam you can even go to hey adam dot xyz and literally the entire agent runs fully on convex so right here there's a builder and this
basically will build you an Eve agent and deploy it on your Convix account. I already have two deployed. I'm just going to open one right now.
And I'm just going to open a new chat and say, what can you do? Now, I want you to understand this. This is a personal agent built completely using the E framework, fully running on Convix.
Now, you might say, oh, but this was a little just slow. It took a second. Mind you, I did no optimizations.
This was a couple prompts. It fully is running. on configs.
I can even say, can you check my latest email? And because I have ComposeU set up as well, it can do tool calling and it can check my emails. Notice it did a connection search.
Just checked if I have any Composio tools. Now it's using Composio search tool. So I can build an agent that's fully deployed on Convict.
So I'm not locked up in Vercel land. I have two emails connected. It's asking me which one.
So you could see this clearly works. Again, you can fork this, do whatever you want with this. Another version that I have, this one, unfortunately, is not open source because there's still some tweaks I'm making.
And I call it Ezekiel. And this is basically Eve running. fully on Cloudflare.
Now I know Vercel and Cloudflare got beef, but the fact that I can run EVE fully on Cloudflare using Cloudflare's printers like workers, their sandbox, and even their gateway is incredible. It just shows you how powerful of a framework EVE is. And for those of you who want to have a personal agent on EVE, but you don't want to build it from scratch, you want a place to start, I have a repo.
I'll have all these linked in the description down below, eve -agents. This is basically personal agent build. it's not fully fledged like pluto but this will give you like a rough chat with some tool calling all that stuff and this fully lives in your own personal for sale account you can fork this mess with it do whatever you want but it's there for you to play with now although we have all these different personal agents and you know you have hermes open claw grok bod you have pluto coming soon all these different agents the one thing that i've come to realize is what makes your agent powerful is the set of tools you gave it.
And this is why I built Bezalel. Now, I shared with you guys in the last video what this was, but you can think of this as a single MCP server that gives you access, that gives your agent access to memory, email, the ability to spend money, text via iMessage, a computer, sandboxes, and connectors using Composi. Now, here's the thing.
There's a part of me that thinks that people are going to get tired switching personal agents, right? Somebody's going to win. But what's tiring right now with trying all the different agents, is I have to set it up again and again and again.
What if I had something like Bezalel where I set it up once, where I set up the connectors that I want, I set up my agent's email, right? I set up, oh, right here. This is free, by the way, and I have to up my agent mail account because more than 50 emails have already been created.
Imagine you give every agent its own computer, right? Its own sandbox, its own memory, its own, you know. cards coming soon you can have different channels you can communicate with your agent with right you have iMessage set up and there have been a lot of agents using this 440 users over 462 agents almost 4 ,000 operations and this is what I'm starting to understand you know the clods and the codexes are sort of entering the personal agent space and so is grokbot and all these different things and so is pluto right even i'm entering the arena and there's a part of me that's like maybe it's just a set of tools and people will have their own specific personal agents or maybe they have many one thing i came to realize with all these new tools because i'm not faithful and i keep switching and i want to test the new tools having to set up got tiring so i gave my agent bezlo and bezlo already has all the tools that I like.
I say that to say, if you are going to build your own custom personal agent or you're using open clone armies, I would really take into consideration researching tools that exist that if you gave your agent that tool, it would make your life a lot easier. I think Bezalel is a good answer. It's free right now.
Check it out. Now, the last point, the best personal agent has not dropped yet. And the reason being is I think we're still discovering the best ways to communicate with an agent.
I'm even going as far as to say chat might not be the best way. Let me give you an example on Pluto. We have this nice little cute guy.
Well, it's a girl because my agent's name is Ruth. And I'm just going to allow this time. Hey, Ruth.
Hey, Michael. How's it going? So I'm recording a YouTube video right now.
And you are going to be in the YouTube video. Can you introduce yourself and tell people what you do for me? Sure thing.
Hi, everyone. I'm Ruth, Michael's personal AI assistant. I help him with everything from managing his schedule to handling business tasks, researching opportunities, and keeping everything running smoothly.
Think of me as his right -hand helper for all things AI and productivity. Yeah, can you tell me some of the agents that we have, like the bots, the agents that we've created? Absolutely.
Michael and I have developed several AI agents for his consultancy. We've got custom bots for automating workflows, AI assistance for client support. and even agents that help with government contract analysis and bidding.
Each one's tailored to make his business more efficient and productive. And can you tell, like, give people a reason why they should use you? Because obviously you're going to be a product soon.
Like, what are some reasons why they should use you? Of course. One big reason to use me is I handle the busy work.
Automating tasks, digging up info fast, and keeping everything coordinated. Another reason... I'm customized to your business needs, so I help you move faster and make smarter decisions.
Plus, I'm on 24 -7, ready whenever you need me. I appreciate you, Ruth. You have a good one.
Thanks. You too, Michael. Talk soon.
And so after I finish communicating via voice, it's a thread. I like threads. These really help me a lot.
So the reason why I say the best personal agent has not dropped yet is because I still think we're discovering the best ways to work and communicate with these agents. I'm not sure if chat is the best medium. Multiple threads.
works but like having a singular chat like i know a lot of people love the i message i have i message configuration on pluto as well i just don't think it's the best way to get work done and therefore that's why i think the best personal agent hasn't dropped but yeah that's pretty much it for this video let me know your thoughts down below the next video which will be a build -up of this is we're going to talk about self -healing agents so if you haven't subscribed make sure to subscribe like comment all that stuff i appreciate you for watching to the end thank you so much i've been ross you've been awesome Thank you so much for watching this video.
I'll see you in the next one. Peace!
The Hook
The bait, then the rug-pull.
The creator ditched two popular self-hosted personal-agent platforms after they broke on nearly every update, and rebuilt his agent from scratch on a framework called Eve that is purely file-based and portable across hosts. What he landed on by the end wasn't a framework recommendation so much as a rule: the tool set behind an agent matters more than what built it.
Frameworks
Named ideas worth stealing.
02:45model
The Agent-Building Spectrum
AI SDK — bare bones
Eve — framework
OpenClaw / Hermes — opinionated but flexible
Grokbot — super opinionated
A four-tier way to place any agent tool: a raw SDK gives full control but no plumbing, a framework pre-wires common needs while leaving decisions to you, an opinionated self-hosted platform bundles integrations and memory at the cost of setup work, and a fully hosted platform removes setup entirely but locks you into its fixed model.
Steal forevaluating any build-vs-buy tooling decision, not just AI agents
CTA Breakdown
How they asked for the click.
VERBAL ASK
20:08subscribe
“if you haven't subscribed make sure to subscribe like comment all that stuff”
Delivered as a quick, low-pressure sign-off after the substantive content ended, not before it, so the ask doesn't interrupt the demo-heavy back half of the video.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A creator who pays for every major AI coding subscription ranks Codex, Claude Code, Cursor, and Devin on models, subsidies, and harness quality — then reveals how to get your employer to cover the bill.
How one developer chained a computer-use verification loop and an automated code-review loop to ship 75+ pull requests with an AI cloud agent, without reading most of the code.
A builder walks through every infrastructure decision behind his own AI-agent product, arguing that system design - not AI - is what separates a prototype from an app that survives real users.
00:19
00:37
00:49
01:09
01:21
01:56
02:11
02:27
02:44
03:00
03:16
03:31
03:44
04:15
04:30
04:57
05:11
05:25
05:49
05:59
06:20
06:29
06:51
07:22
07:37
07:55
08:14
08:24
08:55
09:03
09:26
09:41
09:57
10:07
10:24
11:03
11:17
11:30
11:45
11:54
12:16
12:27
12:47
13:10
13:14
13:34
13:43
13:57
14:14
14:36
14:53
15:19
15:32
15:57
16:11
16:31
16:41
17:03
17:04
17:29
17:52
18:08
18:44
19:00
19:15
19:31
19:49
20:01
20:17
20:33
21:24
Video of the Day17:29
19:22
Video of the Day20:40
17:06
30:18