The figure below shows a work item's path through the fleet's judgment: every route it could take, the context that changes the odds, and the pick it logged. Each pick goes into the log under the panel, so the next run of the same work starts from what was already decided. A person keeps the judgment.
sampled · logged. the next instance of this work picks up from the log.
update drafted · numbers first, risks on page one · held for your edit
[auto] candor rule → 1 framing locked · sampled · logged
- ph. open · claims [01] [02] · the unit, the standing fleet
Work comes in with every route it could take. The unit is a standing workspace; the fleet is many of them, one per operation. Nothing is decided yet.
- ph. context · claim [03] · the overlap is the work
The workspace already knows this work. The context chips appear one by one, and the probability bars move toward the call the fleet will make.
- ph. done · claim [04] · a person decides
The field narrows to one logged judgment, sampled and logged. You take the pick; the next run starts from it.
- 01[auto]the monthly board update · lead with cash position · logged
How we think
The work gets rebuilt so the agents do it from the start.
None of this started as a framework. It came from twenty years running companies in and around pharma, devices, and software, and then from rebuilding the work inside those companies so the agents do it from the start: the reports, the monitoring, the follow-ups. Every company that has run this way so far is one Braydon owns or co-owns. The four claims below are what survived. They are the mental model for how AI helps.
[01]
the unit
The unit is a standing workspace with memory, a repository, a task list, and the systems it is connected to. When work comes in, the fleet that takes it already knows the work.1
[02]
the standing fleet
A fleet is many of those units standing at once, kept separate by company and purpose. When work comes in, it goes to a workspace that already knows the operation. Agents get built and retired as the work changes; what persists is the memory.2
[03]
the overlap is the work
The work is in the overlap: one agent is rarely the whole job. A question goes through several worldviews and gets challenged. A build moves through research, planning, dependencies, and review. The shape is the same: context, task, execution, then a handoff a person can inspect. Work that runs through this shape is on the examples.3
[04]
a person decides
A person designs the contexts, assigns the work, redirects it, and decides what becomes part of the next state. The agents keep the work moving.4
Notes
- [1] The fleet figure above shows one work item collapsing: the fields behind it are standing agents, durable context, and running services.
- [2] The count (10+ fleets, 2–50 agents each) and the daily texture of the fleet live on How we run. Agents are built and retired as the work changes.
- [3] The cases behind those links are drawn from the work records of the companies Braydon is inside, under codenames.
- [4] Operating sources: STRATEGY.md, OPERATIONS.md, notes/braydons-view-of-ai.md (internal).
The same pattern shows up in different kinds of work.
Agent Designer
A multi-role harness for designing agents, panels, prompts, and assignments.
Leaves: Design and reviewShell Scenario Panel
Worldviews, scenarios, impact analysis, specialist seats, and a quality pass.
Leaves: Scenario briefFDA and patient work
Research, rendering, planning, dependencies, and review moving through a graph.
Leaves: Reviewed work product