Target Surface

In plain language: What you’re aiming at.

Definition

A Target Surface is the set of outcomes an Agent wants to see in the world. It names what the Agent is trying to accomplish — the results it is aiming at — independent of whether the Agent can actually produce them.

Any Agent can hold a Target Surface, and the same Agent can hold more than one. A CEO has a Target Surface for the company’s next quarter (the revenue, retention, and market-position outcomes he wants). The same CEO has a Target Surface for the VP of Sales role he is creating (the outcomes he wants that position to produce). A VP has a Target Surface for his own work. A deployment engineer has a Target Surface for the AI system she is building. In each case, the Target Surface is the same kind of object: desired outcomes, stated from the perspective of whoever holds the objective.

Target Surface is independent of the Agent’s Boundaries. The CEO can want 30% revenue growth whether or not the company’s Causal Boundary can produce it. A deployment engineer can want accurate, fast ticket resolution whether or not the AI system’s Cognitive Boundary extends to the domain. The gap between what the Agent wants and what the Agent (or the position, or the system) can actually deliver is where the framework’s diagnostic value lives — but the diagnostics are downstream. The Target Surface itself is just the objective, cleanly stated.

What Target Surface is not:

• Not a Boundary. Target Surface describes desired outcomes in the world. Boundaries describe what the Agent can sense, model, and affect. The relationship between Target Surface and Boundaries is diagnostic — how much of what I want can I actually reach? — but they are different categories.
• Not fixed. Target Surface changes when the Agent’s objectives change. A company that pivots from growth to profitability has changed its Target Surface without any change to its Boundaries. The world looks the same; what the company is aiming at is different.
• Not limited to what the Agent articulates. An Agent’s stated goals and its actual Target Surface may diverge. The CEO who says “we’re focused on customer satisfaction” while every decision optimizes for short-term revenue has a stated Target Surface and an actual Target Surface, and they are not the same. The framework treats the actual objectives as the Target Surface — the outcomes the Agent is functionally aiming at, not the outcomes the Agent claims to be aiming at. (This parallels the revealed-versus-stated-preferences distinction in economics — a connection worth exploring downstream, but not developed here.)

Relations

Target Surface gets its diagnostic value by being compared against other framework objects. The primary comparisons:

• Alignment — the degree to which one Agent’s Causal output approximates another Agent’s Target Surface.
• Success Overlap — the intersection of an Agent’s Positional Causal Boundary with a Target Surface. What the Agent can actually reach, of what someone wants reached.
• Three-Boundary Failure Decomposition — when an Agent misses a Target Surface, the miss is diagnosable by which Boundary failed.

The gap between Target Surface and Positional Causal Boundary at Capacity is one lens on organizational design risk — the position enables effects the designer did not intend, or does not enable effects the designer needs. The gap between Target Surface and Actual Output is a lens on whether the outcomes are being produced in practice.

Frame intersects with Target Surface at the self-model layer. An Agent’s Frame includes a model of its own Target Surface — what it thinks it is trying to accomplish. That self-model can be wrong: the Agent may believe it is pursuing one objective while its decisions functionally pursue another. This is a Frame-level mismatch, not a Target Surface mismatch — the actual Target Surface is whatever the Agent is functionally aiming at.

When two Agents interact, there are two distinct questions that use Target Surface differently. First: does the Agent’s Causal output approximate the other Agent’s Target Surface? (This is alignment of outcome — the Agent is producing what the other wants.) Second: do the two Agents’ Target Surfaces approximate each other? (This is alignment of intent — the two Agents want the same things.) The two can diverge: an Agent can want the same outcomes but fail to produce them (aligned intent, misaligned outcome), or an Agent can produce the right outcomes while pursuing entirely different goals (aligned outcome, misaligned intent).

Example — CEO

A CEO creates a VP of Sales role. Before he hires anyone, he defines what the position should produce: grow revenue 30% in the next year, expand into two new verticals, and retain 90% of existing accounts. These outcomes — the revenue growth, the vertical expansion, the retention rate — are the CEO’s Target Surface for the role.

The Target Surface exists before anyone sits in the chair. It is what the CEO wants to see in the world, not a description of the person who will pursue it. Two different candidates could occupy the role; the Target Surface stays the same. Whether either candidate can actually hit the Target Surface depends on the candidate’s own Boundaries (Cognitive, Causal, Frame) and on the Positional Causal Boundary the role confers — but those are separate questions from what the CEO is aiming at.

The VP who is hired may have his own Target Surface for the role — he may want to build a high-performing team, develop his own leadership skills, and exceed quota significantly to earn a substantial bonus. His Target Surface and the CEO’s Target Surface for the same position may overlap substantially, partially, or barely at all. The degree of overlap is not a moral question — it is a structural diagnostic. Where the two Target Surfaces diverge, the CEO and the VP are aiming at different outcomes, and the interventions are different depending on which Boundary is producing the divergence.

Example — Research

An engineer designs a customer support AI system with two agents: a triage agent that receives incoming tickets and routes each one to the right handler, and a resolution agent that takes the routed ticket and produces a response. The engineer’s Target Surface for the system is: tickets resolved accurately and quickly — the right answer, delivered fast.

The Target Surface exists as a specification before either agent is deployed. It describes the outcomes the engineer wants to see in the world — satisfied customers with resolved issues — not the capabilities of the agents that will pursue those outcomes.

When the system runs, a customer submits a ticket about a billing error caused by a recent product change. The triage agent consults the routing table that maps ticket categories to resolution agents. The routing table is outdated — the recent product change created a new category of billing-related technical issues that the table does not account for. The triage agent classifies the ticket correctly according to the table it has, but the table sends the ticket to the technical support resolution agent instead of the billing resolution agent. The resolution agent does what it can — it produces a technical answer that does not address the billing issue. The customer gets the wrong kind of solution (less accurate). A human support manager notices the error and intervenes to redirect the ticket (less fast). Both dimensions of the Target Surface — accuracy and speed — are missed.

The miss does not live in either agent’s individual judgment. The triage agent classified correctly given what it had. The resolution agent answered the question it was given. The miss lives in the routing table — the artifact that connects the two agents — which did not account for the new category. This is a coupling problem: the interface between the agents carried the ticket to the wrong destination, and the downstream consequences cascade from there.

The engineer’s Target Surface has not changed. What changed is the Actual Output of the system — the outcomes it produced in practice. The gap between Target Surface and Actual Output is the starting point for diagnosis. Where the gap comes from — in this case, an outdated artifact in the coupling between agents — is a downstream question the framework’s diagnostic tools answer. The Target Surface itself just names what was aimed at.