Stop copy-pasting between your AI agents

Here we go. And oh, oh, oh, oh, oh, I'm wounded. Hey, got him, yes. Welcome to this week's edition of Field Notes. Field Notes is the show where we go through some hot tips and tricks, things we've learned about working with AI, things we wanna share that, uh, to motivate and inspire you to innovate and be creative. And yes, you caught me playing a video game on this one, and that's because my AI agents helped me build this video game. In fact, they've helped me build an entire game studio, Tinmut Games Been a big week. What's going on? Uh, and you know, it uses a lot of tokens, and my Grok bot ran out of tokens midweek while we were still working on things. And so that's actually the point of what I wanna cover today. It's not even about the game. It's not that you can develop games. Like, not everyone wants to make games. Who cares? Why is every AI demo about games? It's because all the people making AI are a bunch of nerds that like video games. I'm pretty sure that's what it comes down to. Anyway, if you have this thing where you are, uh, deep in with, like, a Grok bots or any AI platform, one of the issues you always have is the fact that you've got this sort of, sort of lock-in. You're, you're stuck in that thing, and when you run out of tokens or run out of credits, what do you do then? Now, if you have multiple subscriptions for various reasons, that's great. But again, something like a Grok bot, you can't plug your ChatGPT in to power Grok bot. That's very Grok bot-ish. At the moment, OpenAI, if you've got access to their dots, dots are somewhat unlimited. So I've been taking advantage of that with Hamilton and getting Hamilton to do most of that game dev work. And I'm just using my Grok bots for managing the publishing, for doing feedback and reviews and testing, uh, and then publishing it up onto the internet, as well as building the websites and things. So I've been dividing the labor. But then how do I do that most effectively? If you have multiple bot agent infrastructure systems across different providers, how do you allow them to talk to one another? Are you just copy and pasting? Have you become the meat proxy, just copying and pasting between the two of them? That is not something I wanted to do, and so I use a project management tool. Now, there's plenty of them out there, whether you use Linear or Trello or Monday or Asana or whatever. I use one called Plain that is open source and I can run on my own hosting. Uh, and here are all the work items across my games. And so here is Hedgehop, that one you just saw me playing. And if we open that one up, let's close that away. You can see here's the initial notes we put up there. And this was put up th- by Grok bot, and the Grok bot's got questions for review for Dave and Hamilton. "What are your opinions on the fun?" And so I got my Grok bot to spec up some ideas, and it put them in here, and then I chatted with Grok bot back and forth. But you can see here as we go down through the activity, squire-valid-agenda is the login that my Grok bot uses, and david\+. is the login that Hamilton uses. And I am just me, I'm Dave, but I don't comment in here. This is where the two AIs share notes back and forth. So Hamilton issued a prototype, status and feedback, put it up on GitHub, gave some more feedback. Then, uh, Grok bot They had some art direction notes for, uh, for Hamilton and so on and so forth, back and forth. Even though they're on completely different ecosystems, I was able to get them to share notes and cross-collaborate. Now, I've done this for some time across all my different agent systems. I do this with my Hermes agents. My Hermes agents have access to this plain environment as well, and they can log tasks up there for my other SaaS chart reporter, and then I go to a different computer, which is running, uh, OpenAI Codex, and it accesses the, the, the reports there. I've gi- actually setting up a GroqBot to help manage my chart reporter instance as well, and it can also give tasks through to Codex via this sort of plain project management tool. It's amazing. You need some kind of shared spine. In this era where people are running across multiple different computers or AI backend model systems and harnesses and so on, how are you managing that cross-collaboration? If you were just copying and pasting things, if you are a meat proxy, there's a problem there. Now, you might be able to have some... Maybe it's just a Google Drive. All these things have connectors now. So if it's a Google Drive and a Google Sheet and something really simple like that, where you've got your markdown files just sitting in a shared repository that they can all access. Using tools like GitHub is really handy 'cause you get that versioning as they commit and push, you get all that versioning and difference, and so that way if they start editing the same files, you get that collisions and you get that version history built into that platform. Uh, I like this where I'm using a tool where I've given them separate user logins, and they can comment back and forth to each other, and that way I get the full audit trail of the chat between agents. And it's great. GroqBot, GroqBot, you have multiple bots, and they all talk to each other, and that's awesome. But this is GroqBot speaking to my OpenAI dot, back and forth, back and forth. Um, I'm still me proxying a little bit, I'll admit, where I go back and say, "Hey, there's feedback from Hamilton. Hey, Hamilton, GroqBot's got some feedback for you," and prompting them. But if I wanted to, I could tell them to set up schedules to check plane on a regular instance. I'm not, because I'm being hands-on with the development of these things. It's important, especially when it comes to things like video games, I wanna be the human in the loop. I don't wanna be one-shotting these things off and just going, "Yeah, build a game, that'll be fine. Out you go." Admittedly, I would say some of these Timmy games, they still need a bit of play testing and a bit, a bit of balancing. I don't think they're all completely perfect and exactly like the best games ever yet, but they're pretty fun already, and I wanted to get some real user player feedback on these three prototypes. Just different games, different genres, but amazing that Hamilton was building these every night. I'd give it a bunch of feedback and say, "I'm off to bed now. I'll see you in the morning with the updates." And I wake up in the morning to three new builds, all the, all the updates there, and then I could then play test, give feedback. We'd iterate, we'd iterate. I'd get GroqBot to, to do a play test, and it would iterate and give its notes, and then we'd send those back through to Hamilton. It was a collaborative process with me and my bots But ultimately, it was made possible, and we talk about managing context, because we had the, the codes being shared and then backed up through GitHub, and so they both had access to the same code because Hamilton would push the games into GitHub, and then Grokbot could look at the GitHub. And also from a actual what was taught, what, like, the, the scope and the analysis and the feedback was all happening via the tasks in Plain. So yes, I built some games. Good for me. tinmygames.com exists now. The games are free to play up on itch.io. Get out there, have a play, give me some feedback. I'd love to hear from you on what you think of the work, my collaboration with my agent team on getting these games out. I've got some bigger, meatier game projects I'm working on, but I just wanted to get some lightweight, quick, simple games out there to begin with. A bit of fun, just to test the waters and see what game development is like with these agent friends. So, uh, that's the field note. What do you think? Are you using some kind of shared knowledge spine between your different agent platforms? Are you going, "Who has multiple agent platforms, Dave? I just have Claude Code." Great. Great. That's awesome. Um, but again, if you run out of Claude Code tokens, where are you going next? And if you're just running out of f- folders and repositories on your hard drive, of course, you just open up the other harness in that, that thing. But this is bigger at scale things. Working with hosted agents, you need different tools, and that's what we're here to share. So anyway, uh, if you like this different kind of tool, um, get into it. Uh, drop some comments. Make sure you like, subscribe. Tell us what are you building with bots? I mean, this is a common trend that we're seeing in the field notes at the moment. I'm doing more talking about bots and agents and what I'm building with them, whether that was doing full video editing of me out with the kangaroos or now building games, uh, and now this cross-collaboration between different bots to build bigger applications and to have different work streams where I can leverage quotas and skills and specialties. That's, that's sort of the next iteration that I'm going through. So I'd love to hear your stories. Let me know, and we'll see you next time.