The Specific Event

A tech CEO recently circulated an internal memo asking employees to submit what he called the "D+ version" of their ideas before AI tools polish them into something presentable. The policy is straightforward in intent: show your thinking first, then let AI refine it. The reasoning behind the memo is that AI-generated output obscures the cognitive work that supposedly justifies employing humans in the first place. This is a specific, operational intervention, not a vague cultural statement about AI. It is worth taking seriously as an organizational design choice, and then asking whether it solves the problem it claims to address.

What the Policy Actually Does

The CEO's instinct is defensible. If employees route all written communication through AI before anyone sees it, managers lose observability over reasoning quality. The "D+ version" requirement attempts to restore that observability by inserting a checkpoint before the polishing stage. In organizational theory terms, this is a monitoring mechanism designed to reduce information asymmetry between workers and supervisors about cognitive output. The problem is that it conflates two distinct things: the quality of a person's thinking and the quality of their unassisted writing. These are not the same variable, and treating them as equivalent produces a policy with real measurement validity problems.

The Awareness-Capability Confusion, Organizational Edition

My research on Algorithmic Literacy Coordination draws a firm distinction between algorithmic awareness and algorithmic capability. Knowing that a system exists does not tell you how to respond to it effectively (Kellogg, Valentine, and Christin, 2020). The "D+ version" policy commits a structurally similar error at the organizational level. It assumes that observing unpolished output reveals something stable and diagnostic about a worker's underlying reasoning. But what it actually reveals is that worker's current proficiency at translating thought into text without AI assistance, which is a skill that has been depreciating for knowledge workers for at least three years. The CEO is measuring a capability that his own industry helped erode, and then using that measurement as a proxy for cognitive quality.

Routine Versus Adaptive Expertise in Knowledge Work

Hatano and Inagaki (1986) distinguish between routine expertise, which is the reliable reproduction of practiced procedures, and adaptive expertise, which is the ability to restructure knowledge when context changes. A policy that demands unassisted first drafts selects for a specific kind of routine expertise: the ability to produce legible written output under conditions that increasingly resemble a deliberate handicap. Workers who have genuinely developed strong schema-level thinking may produce terrible unassisted prose while still reasoning well. Workers who have always relied on clear writing as a substitute for clear thinking may pass the D+ test without difficulty. The policy inverts the signal it is trying to detect.

What a Structurally Sound Version of This Policy Would Look Like

If the goal is to maintain observability over reasoning quality in an AI-mediated communication environment, the intervention point should be at the level of argument structure, not prose quality. Hancock, Naaman, and Levy (2020) frame AI-mediated communication as a category where the key theoretical question is what the AI is doing to the signal, not just to the surface. A structurally sound policy would ask workers to submit a logic map, a claim-evidence outline, or a decision tree before any drafting begins. That would capture reasoning prior to both AI polishing and prose execution. The D+ memo as described does not do this. It captures unpolished prose and calls it thinking.

The Organizational Theory Problem Underneath This

There is a broader issue here that the memo symptomizes without diagnosing. Organizations are attempting to build governance structures for AI-mediated work without adequate schema for what AI is actually changing about knowledge work. The result is policies that are reactive and topographic, meaning they map surface-level behaviors, rather than topological, meaning they address the structural relationships between inputs, processes, and outputs. Rahman (2021) describes how invisible algorithmic constraints reshape worker behavior in ways that management cannot easily observe. The D+ memo is, in a sense, an attempt to make that invisible constraint visible. The problem is that the CEO has identified the right concern and implemented the wrong diagnostic. Capturing the unpolished draft does not reveal the reasoning; it reveals the writing, and in 2025 those are not the same thing.

References

Hancock, J. T., Naaman, M., and Levy, K. (2020). AI-mediated communication: Definition, research agenda, and ethical considerations. Journal of Computer-Mediated Communication, 25(1), 89-100.

Hatano, G., and Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, and K. Hakuta (Eds.), Child development and education in Japan (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., and Christin, A. (2020). Algorithms at work: The new contested terrain of control. Academy of Management Annals, 14(1), 366-410.

Rahman, H. A. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. Administrative Science Quarterly, 66(4), 945-988.