Transcription
Here it comes again. Slack ping, Discord message, WhatsApp buzz, notification, email, calendar, Teams, another Slack ping. Every app wants your attention. Every platform has its own inbox. And if you want to use AI to help, you're copy-pasting between ChatGPT and everything else like it's 2019.
Yet, all of a sudden, out of nowhere, someone says, "Hey, have you heard of Moltbot?" and sends you a link. Allegedly, it's one assistant everywhere you already are. Message it on WhatsApp, get a response on Telegram, ask it to check your calendar from Slack, and it just works across everything. One assistant, one place can simply access all of your things, including the files on your computer. And it honestly has over 100,000 stars on GitHub already. 500 pull requests, 200 contributors, and a lot of people are calling it the future of personal AI. Some people are saying that it's some sort of AGI, and the hype is real.
The enthusiasm might be real, but here's what most people are not understanding me. If you open the codebase, all 50,000 lines of it, read every file, check every folder, you won't find a model. You're not going to find weights. You're not going to find training data. There is no AI inside of it. None.
So, what is everyone all excited about if it's this crazy AI agent that's handling everything? To understand what Moltbot actually is, we need to talk about a word that gets thrown around a lot in AI circles, but rarely gets explained well. Orchestration.
Think about an actual orchestra. Dozens of musicians, brass, strings, woodwinds, percussion. Each one a specialist, each one capable of making music on their own. But here's the thing. The conductor does not play an instrument. The conductor doesn't make any sound at all. The conductor's job is to coordinate who plays when, how loud, in what order, when to speed up, when to slow down, when to pause. That is the important part. The musicians are the talent. The conductor is the organizer. Now replace musicians with AI models, tools, messaging platforms. That's the orchestration.
Let me make this a little bit more concrete, and I kind of want to take you into here a second because there's a good lesson to learn. I'm going to use some simple shapes. A circle is your message. "Hey, what's on my calendar today?" The pentagon is an AI: Claude, GPT, Llama, whatever model you're using. A square is a tool: your calendar, your browser, your file system. And a rectangle is a channel: WhatsApp, Slack, Discord.
Without an orchestration layer, here's what happens. Your message has to find the AI. The AI has to figure out what it needs, a tool, when it should have a tool, what the tool is, and it has to be called. The response has to get back to the AI. And then the whole thing has to find its way back to whatever app you sent the message from. It's a mess. Every connection is manual. Every integration is custom. Things break constantly.
Now, if you add an orchestration layer in the middle, a central hub that speaks every language, your message goes in, the orchestrator routes it to the right AI. The AI says, "I need calendar data." The orchestrator calls the calendar tool, gets the response, sends it back to the AI. AI formulates an answer. Orchestrator sends it back to WhatsApp or Slack or whatever you started. Same task, clean architecture. Everything flows through one intelligent router.
But there's something important I want to hit on here. We should get our terms straight because people use these words interchangeably and they really shouldn't. An AI model is the actual intelligence. GPT, Claude, Llama, so on and so forth. The things that think, and I use that with a grain of salt. Moltbot is not this. A wrapper is a thin layer around an API. Minimal added value. And there's a lot of tools out there like that, but I don't think Moltbot is this either. An agent is an AI that can take actions autonomously. Moltbot enables this, but needs an external AI to do it. But an orchestration layer coordinates AI plus tools plus channels plus context. This is what Moltbot actually is, in my opinion. And the simplest way to think about it is Moltbot is the highway system. You still need to bring your own car to drive on it, to have it go places, make deliveries, all of that good jazz.
But not arguing over what it is or how it is, I think we should actually look at Moltbot's architecture, not just because of Moltbot, because it's a really great way to build all of your tools and apps in the future or in the AI world. Now, what I like about Moltbot is they really designate areas into layers, right? They have the input layer, which is kind of where your interaction space is. They have this gateway layer, where it's kind of this area where Moltbot is handling on sending those things, routing them, authenticating stuff. And then there's this, uh, kind of major output layer, where, you know, they're taking AI models and allowing you to connect them, looking at the storage. This is where it really wants all of that stuff to kind of be connected to and flowing really well.
But let's try to break them apart into their individual sections. So at the top, you have this channel layer. This is where Moltbot connects to all of those messaging platforms, right? WhatsApp, and they're using a library called Bailey's, Telegram through Grammy, Discord through Discord.js, Signal through Signal CLI, and so on. These are each of these really solid, maintained, open-source libraries that really know how to speak the platform's language.
In the middle, you have the gateway. This is the brain of the operation. And I don't mean an AI brain, although you could argue that. I mean a routing brain. It's a WebSocket server running locally on your machine. Every message from every channel throws through here. The gateway decides where it goes. It manages sessions so that AI remembers your conversation context and it handles authentication so random people can't just message your AI. Hopefully, there's some arguments that there's some big opening securities in it, but that's besides the point.
At the bottom, you have two things: tools and AI models. The tools are included: browser automation through Chrome's debugging tools, bash commands, file operations, cron jobs for scheduling, webhooks, all of them built-in. And the AI models, they're just hooks. They're empty. You have to actually bring your own, right? You have to bring in your own API keys. You plug in your Anthropic API key or your OpenAI API key, or you run a local model from Llama and you point that model inside of it. Moltbot will handle all of the API calls, but the intelligence, the actual thinking, is technically coming from somewhere else, somewhere you might be paying for already.
So what does this actual codebase look like if it's not a giant AI model? Well, about 35% of this is actually channel integrations. That's the code that speaks with WhatsApp, Telegram, all that stuff. About 25% is the actual gateway, right? Routing, logic, session management, security. 20% is tools: browser control, bash execution, file handling. And only 10% is really CLI configuration and AI, right? It's just basically a bunch of AI calls that you can configure into your endpoint and send messages.
This maps almost perfectly to a framework that I talk about in all of my workshops and lectures and even within my own company, which is this idea that 60-30-10. You should be building AI tools with about 60% traditional code, you know, platform integrations, networking, file handling, stuff that has nothing to do with AI. 30% needs to deal with rule-based logic, routing decisions, management, security rules, smart code, but again, still not these AI models. And realistically, basically only 10% actually needs to be AI calls. And in the case of Moltbot, that 10% is even pushed externally. It's API calls to someone else's AI.
But this isn't a criticism. This is actually a recommendation on architecture. Most production AI systems look like this. The AI is a component, not the whole thing. The value is often in everything around the AI, not the AI itself. Moltbot just makes this explicit, maybe more explicit than people realize when they first hear about it. It's why I'm hesitant to call it a wrapper and actually just call it what it is: good architecture. And of course, it's going to have security flaws, things like that. There's always stuff to be working on it. But honestly, because this is open-source, because you can see their GitHub, I think this is a great opportunity for people to learn how to create these orchestration systems that are actually going to last the decade. This is where we should be spending our time as developers. This is what makes sense.
However, just because this is the right way to code doesn't necessarily mean that Moltbot or something like it is going to stick around. And that's where we come into this question. I want to get real with you because there's a question hanging over this whole project that I don't think anyone seems to be asking. What happens when the big players decide to just build this themselves? If Moltbot is still requiring the other AIs, what happens when the other big guys play this themselves?
Imagine this for a second. Next year, Anthropic announces Claude Everywhere. Native integration with WhatsApp, built into Slack, Discord bot that works right out of the box. No configuration, no API keys, no self-hosting. You just have Claude everywhere. Why would anyone need Moltbot? Why use a middleman when the AI company itself is offering direct access? This, I often think we should call is getting sherlocked. When a platform adds a feature that makes your whole product unnecessary. And this doesn't just happen all the time in AI. This happens all the time in tech in general.
But there's also another way this plays out. Moltbot is in fact model-agnostic. You can swap Claude or GPT or Llama, or even run your own local model, and you don't have vendor lock-in. Plus, Moltbot is self-hosted. Your conversation can stay on your hardware if you keep a local model. Your data, your rules. Plus, Moltbot is open-source, 200 contributors, community-driven development, features that matter to users, not necessarily shareholders. So, Moltbot lets you not have lock-in, but Claude is straight to the source. So, the question is, in orchestration and frontier models, where is the value?
Moltbot has no AI inside. It's an orchestration layer, a router, a traffic controller, a universal translator between messaging platforms and AI surface. It follows the 60-30-10 architecture roughly. Most of the code is plumbing, not intelligence. Its value proposition probably is found in flexibility, model-agnostic, self-hosted, open-source, and customizable. And its future, I think, is genuinely uncertain. It could be replaced by native integrations from big AI companies, or it could thrive as the Switzerland of AI systems: neutral, interoperable.
At the end of the day, here's what I think. The engineering, it's impressive. Stitching together 10 different messaging protocols into a coherent system isn't simple, even with vibe coding. The WebSocket gateway architecture is clean. The tool execution sandbox is thoughtful. Is it revolutionary? I don't think so. But it is good infrastructure for the most part. I think there's going to be a lot of leaks of data and while you're using it, and also on top of that, I don't think it's necessarily defensible either. But that really depends on whether users value flexibility over convenience. And I think most people historically, or at least a good portion of the market, will always choose convenience. But developers and privacy-conscious users, I think they're going to choose flexibility.
My honest take is Moltbot is useful and today has a very specific audience. Power users who want control, developers who want customization, privacy advocates who want self-hosting. But as a mainstream product, it's one Anthropic announcement away from becoming a niche tool. And you know what? That might be fine. Not everything needs to be mainstream to be valuable.
At the end of the day, what do you all think? Is orchestration the real product? Is Moltbot solving a problem that will exist for years or filling a gap that's about to close? I'm generally curious what everyone else thinks. Drop your take in the comments. And if you want to see more breakdowns like this where we actually look at what's inside the hype, subscribe. I can do this for all AI tools or frameworks or whatever else is getting attention that might not deserve it. But anyway, other than that, I'll see you in the next one and have a wonderful day, everyone.