Skip to content
Bot jobsJob breakdowns

im a vibecoder and grokbot completely changed what i can do.

I want to start this out with some caveats. First, I'm not a developer, and I'm not a software engineer. I'm a vibe coder, and I'm proud of it. I'm a guy who's always had lots of ideas but never had

taylorImported from X7 min read
taylorhbuildsx article
See this runHouse 528 · 00727

Article

Job breakdowns

I want to start this out with some caveats. First, I'm not a developer, and I'm not a software engineer. I'm a vibe coder, and I'm proud of it. I'm a guy who's always had lots of ideas but never had the talent or ability to build the scripts and tools I wanted to help me in my daily life or at my job. Codex and Claude and AI have given me that opportunity.

I've had a Codex subscription for a year, and about three or four weeks ago I switched to Claude because of their 5.5 models: Opus, Sonnet and Haiku. It unlocked a whole other level of productivity for me.

I was fighting with Astra and Sol to not overengineer things, to not drift, and to not plan for use cases that were never, ever going to be an issue. I'm not building a SaaS. I'm not building something I'm trying to sell or monetize. I'm building something for me. I have a very unique and specific use case, and I just want the tools to build what I need. But Astra and Sol seemed more concentrated on building the infrastructure and the system around the tool than on building the tool.

When I switched over to Claude, Opus and Sonnet unlocked a wild level of productivity. Opus is the closest thing I've seen to actual intelligence, and I don't say that lightly. I'm not trying to be cliché.

Before I even try to correct it, it goes, oh wait, I need to change this. I've literally found myself describing what I want to Opus, I hit enter, and it just starts building. It doesn't ask me questions. It doesn't want to do research or investigation. I literally have to stop it and slow it down and say, wait a second, I wasn't done explaining what I want. Here's what I need. And its response comes back along the lines of, yeah, well, that's already in my build plan.

this is a real screenshot from opus "thinking" while i'm typing out how frustrated i am lol
this is a real screenshot from opus "thinking" while i'm typing out how frustrated i am lol

Just in the last week, X announced their unified subscription, where I could get Premium+ benefits plus Grok Bot and Grok in Cursor. It finally gave me the opportunity to play around with Grok Bot.

At first I was a little hesitant. Since OpenAI's Dev Day, I'd been trying to use my Dot, and I couldn't get it to work the way I wanted. It constantly stopped. I constantly had to tell it to keep going. It's incredibly slow. But when I started using Grok Bot, I found that it was just easy. It worked quickly. I asked it to do something and it just did it, and it was beautiful.

Then I went down the route of, well, I'm only on a $30 X subscription right now, so I don't have unlimited Grok Bot usage. So I said, wait a second, I have my Codex subscription. I have my Claude subscription. I need to figure out a way to get Grok Bot to drive Claude and Codex, and basically be the orchestrator of my orchestrators.

To give more context: when I switched to primarily using Claude because of the 5.5 models, I also found T3 Code, and it's everything I want in a harness. So in the course of about two weeks, I went to better, faster, more efficient models in Opus and Sonnet, and then I found the right harness for my workflow in T3. That increased my productivity even more.

I installed Grok Bot on my Windows PC, because it's always on and on my local network. From that Windows PC, Grok Bot can access the Linux box where all my Claude and Codex agents and T3 Code run (via tailscale). I don't really run Claude or Codex agents on my MacBook or my Windows machine. I run them all on a headless Linux box with a six-year-old processor and 48 gigs of RAM, so they're always on and always working, I never have to stop them, and my laptop never turns into a stove on my lap.

I asked it to figure out how I could access my T3 threads, my Codex agents and my Claude agents, and it just figured it out. It said, yeah, I can detect the T3 server. I can see the different agents you're running. I can read them, and I can even send them messages.

Then the ideas started to flow. Instead of managing anywhere from 2 to 4 to 10 to 15 threads in T3 Code myself, I could just go to my Grok Bot, which I affectionately named Dave.

Dave is incredible. Dave is the senior engineer on my team, and like I said, I'm not a developer, so having a senior engineer on my team is amazing.

It was actually incredibly easy to give Dave the tools he needed. All I did was go to @poteto, install her pstack plugin, and say, Dave, use this.

Here's her actual grokbot plugin as well:

So now Dave, my senior engineer, is using the tools, skills and workflow of someone who's an incredibly accomplished software engineer at some really big companies and way smarter than me. I didn't even have to teach Dave what I wanted him to do. I gave Dave someone else's experience and workflow, and then I just started talking to him.

I told him to be conversational, to speak in simple terms, and to not be too technical, so I could understand him. What has happened since then has been absolutely ridiculous. Now I tell Dave I have an idea, and he goes, OK, I'll build it using Claude or Codex with the pstack plugin. Here's what I need from you. It doesn't bother me. It just does it.

It orchestrates the agents, and it orchestrates which models my Claude orchestrator uses to actually build. Using T3 Code, my Claude Opus orchestrator can call a Sonnet agent, a Sol 6.1 agent, an Astra agent, a Haiku agent, a Luna agent, or even Antigravity, which is plugged in as a provider in T3 Code. Dave tells it all what to do. It's literally orchestrating my orchestrators, and my productivity is going through the roof.

I let it run for an hour just the other day. When I came back, I looked at my phone, not even my computer, and the message was basically: there are 19 PRs for you to review. Should we merge them?

I said, Dave, you used pstack on this, right? Did they test? Did they audit and verify? Are we good to merge them? Dave said, let me check on that. Dave checked the audits, the verification and the tests, and gave me the file paths where the tests, audits and verifications were. And me not being a developer, I said, OK Dave, I trust you. Merge.

Then I went and used the tools and the skills. I'm building things I use every day in my daily life, and I actually use them. I went and used them, and the fixes worked. They ran better. They did what I wanted them to do. It just worked.

Managing three, four or five different projects at one time, committing, and opening 19 different pull requests in an hour? I literally don't have the mental capacity to do that. I'm not smart enough. But Dave is. The difference comes from now I don't even have to manage my 3-4 orchestrators - Dave is. It's completely and utterly changed the way I work.

Here's the most exciting part - Dave helped me set up a webhook/notifier for my Claude/Codex agents. The part that has been annoying is that i've had to go back to Dave and say hey, 11 of the codex threads are done, check on them. Now with the webhook/notifier in place, when a Codex or Claude agent finishes their turn/what they're doing, it pings Dave and he can go thru and check it. If it's done, he can archive the thread. If its not finished or gets stalled, he continues it. If something needs to be merged, he brings it to me with proof that it's good or not good to merge.

I don't even have to be the middle man. There's no copying, no pasting, no asking him for updates. I just get a notification on my phone when he needs me.

Now I'm sitting here going, I have to upgrade my X subscription to get more Grok Bot usage. Now that I've seen the productivity gains, I don't know if I can go back to orchestrating my Claude and Codex agents myself. Dave breaks it down and makes it so simple. He has conversations with me about what I want to build and what I'm trying to do, and then he just makes it happen, using the subscriptions I'm already paying for. It's absolutely incredible. My productivity has dramatically increased.

Something I'm honestly really excited about is that I'm building skills, tools and things for me to use, and then I'm open sourcing them. If other people ever want to use them, feel free. If not, well, my GitHub will be full of public, open source repos. I'm not trying to build a SaaS. I'm not trying to monetize. I'm trying to build in public: give the scripts, skills and tools I'm building to the public and see where it goes.

I'm super pumped about this. I'm super pumped about how much my productivity has increased. And man, Grok Bot is on a different level.

Published on grokbot.sh. Cite the public log, not a prompt pack.

Command Menu