Computational Boundary

In plain language: What your org can See.

Definition

An Agent’s Computational Boundary is the collection of all pathways through which information reaches the Agent from outside itself. It is the input side of the Agent — the perimeter through which signals from the surroundings enter and become available for the Agent to model. The Boundary is defined by the Agent’s channels: a person’s senses (eyes, ears, touch, and so on); a single biological cell’s surface receptors and bioelectric channels; an LLM’s input messages from its environment.

What the Computational Boundary is not:

• Not what the Agent understands about what it receives — that is the Cognitive Boundary.
• Not what the Agent does in the world with what it receives — that is the Causal Boundary.
• Not internal artifacts that come from inside the Agent. An Agent’s own internal records, dashboards, and reports are not on its Computational Boundary, because the Boundary is the perimeter with the outside of the Agent. Internal artifacts may feed the Cognitive Boundary as Tools, but they do not enter through the Computational Boundary.
• Not tied to a moment or context. The Computational Boundary is treated as time-and-context-free — an analytical convention that makes the Boundary separable from what it is doing right now. The questions of what could those channels carry at maximum and what are they carrying right now are perspectives on the same Boundary — they are Capacity and Realized. The specific signal flowing through the Boundary at a particular position and time is Actual Input.

Relations

One of the three Boundaries every Agent has, jointly definitional with the Cognitive and Causal Boundaries. Tools come in three directions: Computational-side Tools (telescope, microphone, scientific instrument) extend the Computational Boundary by adding channels or amplifying signals at existing channels; Cognitive-side Tools (a map, a written report, a CRM, a dashboard) operate on already-received information and live inside the Cognitive Boundary, not the Computational; Causal-side Tools (gun, lever, broadcast antenna) extend the Causal Boundary instead. Capacity and Realized are the two perspectives on the same Boundary: Capacity is the theoretical maximum signal the Agent’s channels could perceive; Realized is what they perceive in a given context. Actual Input is the instantaneous specific signal crossing the Computational Boundary at a particular position and time — the Computational Boundary is the channel surface, Actual Input is what is flowing through it right now.

Example — CEO

A regional sales leader’s Computational Boundary is the collection of channels through which information from outside him reaches him as a person. Conversations he has with customers and reps. What he sees and hears in meetings. Emails that arrive in his inbox. The window he looks out. What is not on his Computational Boundary, even though information exists out there: a hallway conversation he never walks past; a competitor’s internal strategy that has never been reported externally. The website-form queue that exists but he does not read is more subtle — at the Capacity layer the channel is on his Boundary because he could open the queue, and at the Realized layer nothing flows because in practice he does not. The channel exists; he chooses not to use it.

A note on Tools, since this is where the framework’s three-direction Tool category first matters: a CRM, a sales dashboard, and an analyst’s weekly report are not on the leader’s Computational Boundary directly. They are Cognitive Tools — software, documents, and frameworks that operate on already-received information to extend his modeling capacity. The data inside the CRM originally came from outside the company, but it reached the company through the Computational Boundaries of individual sub-Agents (employees, salespeople), was processed and recorded by them as Causal output, and is now an internal artifact the leader’s Cognitive Boundary uses through the Tool. Computational input arrives at the perimeter with the outside; once inside the Agent, it is no longer Computational input.

Example — Research

A biological cell’s Computational Boundary is the collection of pathways through which it detects signals in its environment. Surface receptors detect chemical signals — pathogen molecular signatures, cytokines, surface markers on neighboring cells. Bioelectric channels (per Michael Levin’s work on bioelectricity as a developmental and cognitive substrate) detect voltage gradients and ionic patterns across the cell membrane. A signal that lands on a different cell type’s receptors is not on this cell’s Computational Boundary, even though it is in the same human body and the same instant — there is no channel into this cell for that signal. The Boundary is defined by the cell’s actual channels, not by what is “happening nearby.” When the cell up-regulates a receptor in response to inflammation, it is expanding its Computational Boundary at Capacity — adding channel capacity. When the cell internalizes a receptor (pulls it inward, off the surface), it is contracting the Boundary at Capacity — removing channel capacity. Tricks the environment plays on the cell — a pathogen disguising itself to slip past a known receptor pathway, for example — do not change the cell’s Boundary. The pathogen is fooling the cell’s input function, not changing what the cell can perceive in principle. The bouncer is being deceived, not removed.