AI-Generated Docs: The Emperor's New Clothes in Big Tech Meetings

Big tech employees report that meetings have devolved into exchanges of AI-generated documents. The author describes a typical workflow: an employee starts with a vague idea and feeds project requirements (often also AI-generated) into an AI with a prompt like "Analyze the difficulty of this project and how it should be broken down." The AI produces a planning document. The employee then refines it with further prompts: "Plan according to a 5-step roadmap; the first step must have a PoC (Proof-of-Concept) deliverable; detail only goals achievable within six months, outline the rest." The AI revises. Another prompt: "Add diagrams: Chapter 1 architecture, Chapter 3 deployment..." The AI adds illustrations. This iterative prompting continues until the employee is satisfied.

The process saves significant time compared to hand-crafting documents over days, and the output has no typos and flows well. However, the author questions whether this truly improves efficiency. For responsible workers, it is a powerful efficiency tool; for the team overall, it lowers the barrier to "slacking off" — producing plausible-looking but empty content becomes trivially easy.

The core value of any document is its core viewpoint . AI can only simulate a viewpoint; trained to provide averages, it yields middle-of-the-road perspectives unless explicitly fed human insights. The author argues that AI expands five sentences of genuine insight into five pages of fluff, diluting signal with noise. Now everyone uses AI to package their views (if they have any), further diluting the original thinking.

This is especially evident in AI-generated slide decks. The author includes screenshots of dense, decorative PPTs that obscure the message. AI lacks genuine viewpoints, so it compensates with flashy graphics and verbose text, making it impossible to convey the point in one or two sentences. The result is slides that require the audience to extract the core idea from a heap of decoration.

The author contrasts this with Steve Jobs' approach: one sentence and one image per slide, conveying the core viewpoint clearly. All the peripheral decoration wastes time and tokens. The author laments that Jobs did not live to see the AI boom and speculate how he would react to the rampant misuse of AI for document generation.

As AI-generated documents flood the workplace, the ratio of core insights to decorative fluff shrinks. When a final document is 1% substance and 99% AI-generated ornamentation, extracting information consumes excessive tokens. The author resigns to the status quo: since everyone accepts the "emperor's new clothes," the system will continue until it breaks down.