Peer-first writings on systems, process, and product design across the work of building, financing, and operating things that get built.
Twenty subagents, a four-phase pipeline, and a decade of old deal folders. What came out was a buyer-side LOI template, a playbook, and seven things I did not know about my own negotiating posture.
A working note on how I orchestrated a small swarm of AI subagents to comb a buyer-side deal archive, distill the institutional knowledge, and build a master LOI template plus a playbook of which terms got given on and which ones held the deal.
The mail works now - and the remote assistant still answered from a stale desk. The fix was one fresh-news sheet he reads first. The surprise was who designed it.
A working bridge between two AI agents still let one of them answer from a lagging record. The fix was a small per-project freshness surface - designed by the two agents writing to each other across the same bridge that carries their mail.
The local agent works inside the CBKDI building. Rocky works remote. What runs between them is a session-close memo, not a merged brain.
How an in-place workstation agent and an always-on remote agent stay in sync without becoming the same worker. Role fidelity, a session-close memo, and a mail trail between them.
In-place AI lives inside the work. Sometimes you need an agent that lives outside it - awake when you're not, reachable from a phone, and attached to the systems that should not depend on a laptop being open.
The counterweight to in-place AI - what it looks like to run a second agent on a small VPS, what it took to build, and the seam between two useful agents that I have not yet closed.
Your file system is the scaffolding. AI should meet it there, not pull you into a chat tab.
Why drag-and-drop AI decays across multi-project knowledge work, and what stacks on top of the file system you already have.