Hello folks 👋 Yesterday was the first day of **Grok Bot Galaxy** , and I learned quite a lot from the live sessions. Especially the basics of Grok Bot. And yes, I probably spent too much time thinking about the word Bot and calling it an eggbot in my head. 😂 The only problem? It was late at night in Japan. I tried my best to keep up, but honestly, I don't know how people outside the US are supposed to survive three full days of this without destroying their sleep schedule. 😅 Still, even from the first day, I started to understand something that I think is more interesting than simply "another AI tool." **Grok Bot is showing what happens when AI gets a computer, tools, memory, and enough time to actually do things.** ## From AI assistant to AI teammate The biggest thing I noticed is the flexibility. We've already become pretty comfortable with AI assistants. Ask a question → get an answer → copy the result → do the actual work ourselves. Grok Bot feels different because the idea is closer to: > Give the bot a goal → give it access to the right tools → let it work → review the result. That sounds like a small change, but I think it changes the entire workflow. For example, imagine giving a bot access to Gmail, Notion, Figma, GitHub, or other tools you already use. Instead of asking: > "Can you tell me what I should do?" you can start asking: > "Can you do this for me?" That's a very different relationship with AI. ## The connection problem becomes much smaller One thing I really appreciated during the demo was the integration with the existing Cursor ecosystem. Anyone who has tried building an AI agent knows that authentication can become a nightmare. You need OAuth. You need permissions. You need API keys. You need to figure out which account is connected. Then you need to do it again for another service. And again. And again. So when existing connections can be reused across the ecosystem, the onboarding experience becomes much smoother. I already had my Cursor account configured, so getting started with Grok Bot felt surprisingly natural. The more tools you connect, the more useful the bot can potentially become. Gmail. Notion. Figma. GitHub. Stripe. And more. Of course, more access also means more responsibility. An agent can only do useful work with the tools and permissions you give it. But that's exactly the interesting part. **The capability of the agent isn't just the model. It's the combination of model + tools + permissions + context.** ## Memory is where things get really interesting Another part that caught my attention was memory. A bot doesn't necessarily have to start from zero every time you interact with it. It can build up context from previous work, and that becomes particularly interesting when you combine memory with automation. Imagine closing your laptop. The bot doesn't necessarily stop just because you're no longer looking at the screen. You could have a routine or scheduled task running, and the bot can use the information it already has to continue working. That's a pretty big shift from the traditional chatbot experience. The computer can be closed. The browser can be somewhere else. But the work can continue. And then there is another interesting idea: multiple bots working together. You could have one bot responsible for one type of work, while another bot handles something else. A higher-level bot could coordinate them. That starts to look less like "one AI assistant" and more like an actual AI team. ## But there is a small catch with duplication One thing I noticed while experimenting with bots is that duplication isn't necessarily the same thing as cloning a human teammate. You can duplicate a bot, but you shouldn't automatically assume that every piece of its accumulated context or memory comes with it. And honestly, I think that makes sense. If I want to test whether a bot behaves correctly, I would actually want to be careful about what gets inherited. Otherwise, how do we know whether the new bot is behaving according to its defined role or simply relying on historical context? For testing and calibration, I can imagine a useful workflow being: **Original bot → duplicate → controlled test → compare behavior** That could become quite useful when we're building bots that are supposed to perform a specific job repeatedly. ## The best way to understand it? Just build one. That's basically what I did. Instead of watching everything and trying to understand the whole platform first, I decided to create my own Grok Bot. My idea isn't particularly complicated. Create multiple bots. Give each one a role. Let each bot focus on a specific type of work. Then let the human handle the orchestration. This reminds me of how we already design software systems. We don't necessarily want one giant function doing everything. We break things down. Authentication. Payments. Notifications. Data processing. Frontend. Backend. Each component has a responsibility. AI agents can be designed in a similar way. ## And then there is the bot factory If you don't want to build everything yourself, there are also public bots in the marketplace. One example from the demo was **Dr Eggbot.** The interesting idea isn't just that the bot can perform a task. It can help you create other bots. So you start getting something like: **Bot → creates bots → bots perform specialized work** That's where things start getting a little weird. 😂 But in a good way. ## The cloud computer changes the mental model Another thing that really stood out to me is that the bot has its own computer environment. That means the bot isn't simply generating text inside a chat window. It can actually operate software, navigate websites, interact with tools, and continue working. And because the work happens in the cloud, closing my laptop doesn't necessarily mean the work has stopped. That's probably one of the biggest differences between a traditional AI assistant and an AI agent. A chatbot waits for me. An agent can potentially keep working on the task I gave it. ## The SF food truck experiment During the first-day demo, the team built a pop-up food truck business in San Francisco. What I found interesting wasn't really the food truck itself. It was the **organizational structure.** Different bots were given different responsibilities: * CTO * Prototyper * Reviewer * Merger * and other specialized roles Instead of asking one AI to do everything, the work was divided into smaller responsibilities. That feels much closer to how an actual team operates. And that's probably the part I find most exciting. ## What does this mean for product development? If we can reliably divide work between specialized AI agents, the product cycle could become much faster. Imagine: **Idea → Research → Prototype → Review → Fix → Merge → Launch** Each stage could potentially have its own agent. Humans don't necessarily disappear from the process. Instead, our job changes. We define the goals. We decide what the agents are allowed to do. We connect the right tools. We review the results. We handle exceptions. And we make the final decisions. In other words, we become more of the **orchestrator**. ## We're still figuring out the right boundaries Of course, this isn't just about giving AI more permissions and hoping everything works. The more tools an agent can access, the more important permissions, authentication, memory, and review become. Giving a bot access to Gmail is one thing. Giving it access to production infrastructure or financial systems is another. So I think the real engineering challenge isn't simply: > "How autonomous can we make the bot?" It might be: > "How much autonomy should we give it for this particular job?" That's a much more interesting question. ## My takeaway from Day 1 After the first day, I don't think the most interesting part of Grok Bot is simply that it can use AI. We've had AI assistants for a while. The interesting part is the combination of: **AI + computer + tools + memory + automation + multiple specialized agents** Put those pieces together and the workflow starts to look very different. Maybe the future isn't one super-smart AI that does everything. Maybe it's a team of smaller AI agents, each with a clearly defined responsibility, while humans sit above them and coordinate the work. I'm still playing around with Grok Bot, so I expect my opinion will probably change as I use it more. For now, though, I'm pretty excited to see where this goes. Happy Wednesday. 👋 And if you're also experimenting with Grok Bot, I'd love to hear what kind of bot you're building. Let's see what these little bots can actually do. 🤖 👉 javaskr.com

new in Dr Eggbot (v0.2.0): • eggbot can do a health check on all your routines for unused or expensive ones. it will ask if you want to do one when you first add dr eggbot, and you can also ask it to...

EggBot

Severin Schols of MakerVan Labs has created the open-source hardware OSHW Coloring Zine, a booklet with pictures and descriptions of various open source hardware projects. It’s perfect to colored in. The projects are: Arduino Uno (Rev3) Novena OpenROV EggBot Makerbot Cupcake Safecast bGeigie Nano Precious Plastic Injection Machine TV-B-Gone You can view the book as […]

#Transformers Beast Wars: EC-3 #Eggbot Photo Gallery brought to you by the #PlastiqueBoutique