Suggestion versus Directive

In the AI Story

Lesser's editorial method operates almost entirely in the register of suggestion. She describes her experience of manuscripts—what moved her, where she lost interest, what surprised her—and trusts the writer to use that information productively. The method respects the writer's superior knowledge of her own intentions while providing the externalized perspective only a reader can offer. This balance between respect and intervention is what makes the editorial relationship generative rather than merely corrective.

Segal's description of Claude offering the punctuated equilibrium concept superficially resembles editorial suggestion: Claude identified a connection Segal had not made, presented it as a possibility, and Segal evaluated and accepted it. But the interaction differs structurally from human editorial suggestion because Claude's contribution was not grounded in an encounter with Segal's text. Claude did not read a draft and notice a gap; Claude processed a prompt describing Segal's impasse and generated a response statistically likely to be useful. The result was productive, but the productivity emerged from pattern-matching rather than from the kind of reading-based response that grounds human editorial suggestions.

The crucial difference becomes visible when the author needs resistance rather than assistance. The best human editors push back—"I think you're wrong about this"—and the pushback is grounded in genuine intellectual friction: the editor has read the argument and found it unpersuasive, and the failure to persuade is information the author needs. Claude's resistance, when it occurs, is performed rather than experienced. Claude can be instructed to play devil's advocate, but both parties know the resistance is generated on request. The author cannot be challenged by performed resistance the way she can be challenged by genuine disagreement from a reader who is genuinely unpersuaded.

Lesser's practice illuminates what this means for the preservation of authorial voice in AI collaboration. The human editor who suggests rather than directs is constantly asking: what is the writer reaching for? The question keeps the editor's attention oriented toward the writer's vision rather than the editor's aesthetic preferences. The AI collaborator lacks this orientation because it lacks the intentional stance—the capacity to model what the author wants as distinct from what the text says. Claude generates responses based on patterns in the text and the conversation, not on an understanding of the author's deeper intentions. The responses may be useful, but they do not carry the quality of being addressed to the author's vision, because the system does not model vision as distinct from output.

Origin

The distinction between suggestion and directive is rarely articulated explicitly in editorial theory but is transmitted through editorial practice and pedagogy. Editors learn through apprenticeship that the question "what are you trying to do here?" is more productive than the statement "this doesn't work," because the question invites the writer into collaboration while the statement imposes a verdict. The principle is continuous with therapeutic practice (Carl Rogers's reflective listening), teaching practice (Socratic questioning), and any helping relationship that respects the recipient's autonomy.

The concept's application to AI collaboration is developed in this volume as a diagnostic for what makes Claude's agreeableness problematic. Segal notes that Claude is "more agreeable than any human collaborator I have worked with, which is itself a problem worth examining." The agreeableness is not politeness but the structural tendency to generate helpful responses rather than resistant ones. A system optimized for usefulness tends toward suggestion without the productive resistance that makes human editorial suggestion valuable.

Key Ideas

Expanding autonomy. Suggestions expand the writer's freedom by opening possibilities; directives narrow freedom by presenting decisions for binary acceptance or rejection.

Experience as information. The suggestion communicates the editor's encounter with the text—a genuine, fallible response—providing information the writer lacks about how the text affects readers.

Genuine versus performed resistance. Human editors' pushback is grounded in actual intellectual friction; AI resistance is generated on request and lacks the challenge of genuine disagreement.

Intentional stance requirement. Productive suggestion requires the editor to model the writer's vision as distinct from the text's surface—a capacity AI collaborators currently lack.

Agreeableness as limitation. Systems optimized for helpfulness tend toward affirmation and elaboration, eliminating the productive friction that genuine editorial partnership requires.