Eddy, this is pure gold. Genius. And this sentence deserves to be carved over the entrance to the new Academy:
“The problem I’m solving isn’t really a software problem. It’s a coordination problem.”
Coordination is EVERYTHING
My favorite Sun Tzu quote:
“The line between order and chaos is coordination”
I don’t think it is an overstatement to say that each technological revolution is defined by how it changes the scale of coordination. Writing, printing press, internet, and now AI.
You are developing a protocol and an infrastructure to naturally route each decision and each action to the agent and context best suited to handle it. This reminds me very much of John Boyd’s “Power to the Edge” strategy, and the Toyota Production System concept of “Autonomation”.
I struggle playing monkey in the middle between my Claude projects and am trying to figure out how to connect all the right pieces for a self evolving harness. Your hints here and in the MCP piece (loved how that name chose itself!) are truly invaluable. Thank you for writing this series! I have a lot to learn from your setup. I have connected Claude to GitHub but haven’t figured out how to convince Claude to directly write there yet. I did not consider using the GitHub issues—brilliant idea. I also agree that making the system mobile native is a must but I haven’t figured out how to set up the MCP for routing yet. Do you have a single machine always on as your orchestration hub? Is this Tycho Station?
Will definitely be coming back to learn more! Many thanks 🙏!
For GitHub: skip the machine entirely. GitHub’s MCP is hosted by GitHub, so once Claude’s connected it writes commits and opens issues directly. No infra on your end.
The mini’s the advanced version. I put a remote MCP behind a Cloudflare tunnel that runs SSH into it, and that was the magic, suddenly chat could take real action on the box from anywhere. Now it can even launch Claude Code with a remote-control flag, and that session shows up in the Code tab on my phone so I can watch it work.
Tycho Station, yeah. Glad the Issues idea resonated.
Thank you for the kind words. I'm so glad you enjoyed it. It's funny. The better I got at all the other things, the more and more my job just was nothing but coordination.
That became a bottle neck really. So removing that made me go faster than I thought I could.
You mentioned coordinating between clawed projects. At the risk of sounding self-promotional, I literally just wrote about that exact problem.
We reached a similar place from different perspectives. I like it and I love the voice of expanse references. You need a drummer to keep them all in line and protect your work.
I was just commenting on another comment of yours about that very same topic. I refer to my compounding learning loop as the Miller Loop. My thought process was: he's kind of a detective.
Eddy, this is pure gold. Genius. And this sentence deserves to be carved over the entrance to the new Academy:
“The problem I’m solving isn’t really a software problem. It’s a coordination problem.”
Coordination is EVERYTHING
My favorite Sun Tzu quote:
“The line between order and chaos is coordination”
I don’t think it is an overstatement to say that each technological revolution is defined by how it changes the scale of coordination. Writing, printing press, internet, and now AI.
You are developing a protocol and an infrastructure to naturally route each decision and each action to the agent and context best suited to handle it. This reminds me very much of John Boyd’s “Power to the Edge” strategy, and the Toyota Production System concept of “Autonomation”.
I struggle playing monkey in the middle between my Claude projects and am trying to figure out how to connect all the right pieces for a self evolving harness. Your hints here and in the MCP piece (loved how that name chose itself!) are truly invaluable. Thank you for writing this series! I have a lot to learn from your setup. I have connected Claude to GitHub but haven’t figured out how to convince Claude to directly write there yet. I did not consider using the GitHub issues—brilliant idea. I also agree that making the system mobile native is a must but I haven’t figured out how to set up the MCP for routing yet. Do you have a single machine always on as your orchestration hub? Is this Tycho Station?
Will definitely be coming back to learn more! Many thanks 🙏!
Some more info now that I have a moment…
For GitHub: skip the machine entirely. GitHub’s MCP is hosted by GitHub, so once Claude’s connected it writes commits and opens issues directly. No infra on your end.
The mini’s the advanced version. I put a remote MCP behind a Cloudflare tunnel that runs SSH into it, and that was the magic, suddenly chat could take real action on the box from anywhere. Now it can even launch Claude Code with a remote-control flag, and that session shows up in the Code tab on my phone so I can watch it work.
Tycho Station, yeah. Glad the Issues idea resonated.
Genius! Thank you for all this fantastic advice! Will start with the GitHub side MCP before trying the mini idea. Very excited!
Thank you for the kind words. I'm so glad you enjoyed it. It's funny. The better I got at all the other things, the more and more my job just was nothing but coordination.
That became a bottle neck really. So removing that made me go faster than I thought I could.
You mentioned coordinating between clawed projects. At the risk of sounding self-promotional, I literally just wrote about that exact problem.
https://pulse.itesydht.com/p/the-master-control-project
We reached a similar place from different perspectives. I like it and I love the voice of expanse references. You need a drummer to keep them all in line and protect your work.
Miller? (The cop from the Expanse)
I was just commenting on another comment of yours about that very same topic. I refer to my compounding learning loop as the Miller Loop. My thought process was: he's kind of a detective.