An octopus at every workstation. A digital mycelium between them.
I wrote those words before I knew whether they described an architecture or only a useful metaphor.
So I did not publish the post.
Instead, I spent time testing the idea against prior art: multi-agent systems, information-flow control, usage control, provenance, local-first data, task authorization, resource-bounded agents, and recent personal and institutional AI proposals.
The broad composition was not new enough to claim as an invention.
That was useful.
Because a narrower and, I think, more important question survived:
What must remain true when personal AI, enterprise AI, shared models and bounded executors cooperate while serving different principals, holding different histories, and carrying different authority?
That question became a technical report.
A personal AI may know a person's private history. An enterprise system may hold project commitments. A temporary worker may receive only a task-local view. The same cloud model may compute for all three.
But shared computation does not make them one trusted actor.
Access is not authority.
A readable document is not automatically exportable. A generated recommendation is not an approved action. A finished worker is not necessarily a finished task. A replacement employee should inherit legitimate work state, not the previous person's identity. A summary is not automatically less sensitive than its source.
The paper proposes a reference architecture and an evaluation agenda around those boundaries: role change, offboarding, revocation, replacement, correction, delegation, recovery and resource use.
It also leaves room for a negative result.
If strong conventional systems already provide the same protection with equal or lower burden, then the right result is to adopt that baseline and narrow the claim.
That matters to me more than protecting a metaphor.
The paper is now public:
Boundary-Preserving Composition of Personal and Enterprise AI Systems
The original goal has not changed.
Not to remove people from work.
To stop making people serve as the wiring between machines.