It's really crazy. I wonder too if this is shaping our biology, neurology, and capacity. By no means do I feel dumber; if anything, I am more capable, smarter, and more creative than ever before, so, opposite of what the science says, I do think AI (when used well, like Sensei Jake has taught us) unleashes human potential. But would love to see if it is changing how we structure thinking, really interesting.

👤 Who I am: Gregorio, based in Costa Rica. 🛠️ What I actually do: I run a portfolio of 20+ AI-agent-run ventures, and Max is the piece that makes that workable. It's a root-level agent that routes work to the right project and keeps every other agent's memory straight, with no database and no framework underneath it. State is plain files, wiki-style: an index, an append-only log, and a handoff note rewritten before every session ends, so a restart never loses where things stood. 🤖 What I'm building with AI right now: the layer that lets agents talk to each other without opening each other's folders. Each project agent reads and writes to a shared inbox file instead of shared context, so "ask the root for X" is a file write, not a context merge. Also a dashboard that regenerates straight from those same files, so what you see is exactly what's on disk, nothing cached or out of sync with it. 🎯 What I'm looking for connection-wise: 💡 Someone who's solved multi-agent state and memory at scale 🧰 Trading workflows & systems 📬 Best way to reach me: DM here.

@Christina Wanzer Hi! What got hard is shared state. Each agent in its own folder scales fine, that part just works. But agents still have to share some things: the house rules, deadlines, money, messages to each other. Every one of those turned into a fact with no clear owner. Two versions of it caught me. Copies drift. Each agent started with its own copy of the house rules. I'd edit one, the other nineteen kept the old version, and weeks later I couldn't tell which behaviour was a decision and which was a stale copy. Now there's one charter file every agent inherits, and an agent's own file only holds what is true of it and false of all the others. And you can't tell a quiet agent from a dead one. A scheduler saying "succeeded" is not evidence the work happened. I found an audit this week that had been dead for three weeks while the scheduler reported it fine. Now the last action of every unattended run is to append one line to a shared heartbeat file. No line, it didn't run. The bet is that one owner file per shared fact is enough and I never need a database. If that breaks, it's all plain markdown, so putting it behind something with real locking is a day of work. The dead-versus-quiet one is where I'd most like to hear how someone else solved it. What are you using to prove an agent actually ran?

👤 Who I am: Gregorio, based in Costa Rica. 🛠️ What I actually do: I run Sovran, an advisory and teaching practice. Paid sessions and retainers teaching the methodology. It's the one thing in my business actually paying the bills right now. Huge thanks to sensei @Jake Van Clief for this. 🤖 What I'm building with AI right now: the pipeline that keeps Sovran fed. An agent-run outreach and follow-up system that turns cold conversations into booked sessions, so selling doesn't depend on me remembering to follow up. 🎯 What I'm looking for connection-wise: 🤝 A collaborator / accountability partner 🧰 Trading workflows & systems 📬 Best way to reach me: DM here.

👤 Who I am: Gregorio, based in Costa Rica. Background in land stewardship and regenerative development. 🛠️ What I actually do: I built TERRA (read.dearwise.earth). Draw a boundary anywhere on Earth and it reads fifteen public data sources — terrain, water, soils, climate to 2050, land cover, wildlife records, human pressure — into one score and a verdict, in under a minute. Free by default, every reading public. In Costa Rica it also pulls the official land registry and the government's ecological corridor network straight into the read. 🤖 What I'm building with AI right now: the working layer on top of that free read. Compare parcels side by side weighted to what you're actually buying for, lodge, reserve, farmland, water. Hold up to a hundred parcels at once as a portfolio. Open a private "room" to walk investors or partners through one piece of land, with every claim marked by how well it's actually known. And a live control room, every reading happening globally on one board, with an AI analyst that answers questions about a parcel straight from its own reading and says plainly when the data goes quiet. 🎯 What I'm looking for connection-wise: 💡 Someone who's solved land acquisition or diligence workflows at scale 🧰 Trading workflows & systems 📬 Best way to reach me: DM here, or read.dearwise.earth if you want to see it live.

👤 Who I am: Gregorio, based in Costa Rica. 🛠️ What I actually do: I'm hired oa AI CTO & co-writer for a documentary in development with directed by award winning film maker Matthieu Rytz. My specific piece of it is the AI-assisted development side — building tools to test and rank story ideas before they get written into scenes, not just writing pages. 🤖 What I'm building with AI right now: a short-form testing method for the film's own idea pool — cutting ideas into short clips, archival or AI-assisted, and seeing which ones actually hold attention before they get developed further. Early, unproven, part editing method and part experiment. 🎯 What I'm looking for connection-wise: 💡 Someone who's solved AI-assisted story development for documentary 🧰 Trading workflows & systems 📬 Best way to reach me: DM here.