Opus Is a God-Tier Coding Partner. But Its Landing Pages Need an Intervention.

Anthropic's Claude Opus, while excelling in coding and reasoning, produces landing pages that are factually accurate but lack emotional resonance, according to a marketing experiment by an unnamed practitioner. The author argues that as AI models optimize for logic, their ability to craft persuasive, empathetic marketing copy may decline, raising questions about the limits of generalist AI agents.

I’ve been running intensive marketing experiments with Opus lately—landing pages, ad copy, product descriptions, you name it. The verdict? It’s blazing fast, factually precise, and boasts an incredibly high information density. But when you actually sit down to read the output, something feels entirely off. The Generalist Era Used to Work My typical approach to AI agents is blunt: no role separation. The same agent writes the code, makes product decisions, crafts the marketing copy, and handles customer support. It’s not born from some grand architectural philosophy; it’s a pure convenience play to cut down on agent-to-agent communication overhead. The older versions of Opus handled all of these hats beautifully. It was a true generalist—never perfect, but always competent. You never felt it was particularly “out of character” in any one role. Enter the Sheldon Problem Recently, however, that delicate balance has shattered. Opus’s coding ability has leveled up astronomically. Hand it a complex requirement, and it produces in an afternoon what used to take a week. The logic is airtight, the structure is impeccably clean, and there’s almost nothing to nitpick. But ask that same genius model to write a landing page, and the vibe shifts entirely. Think of Sheldon Cooper from The Big Bang Theory. Everything he says is factually flawless, yet it completely fails to land with the human beings in the room. Opus’s landing pages suffer from the exact same affliction: they are accurate, logically sound, and structurally polished. But a landing page shouldn’t read like a peer-reviewed academic report. It needs to revolve around a single, compelling idea and offer a relatable solution. Opus just transmits data. It doesn’t build a connection. A great landing page makes the reader think, “This is exactly about me.” Opus’s landing pages make the reader think, “This is about a standardized user persona experiencing a generic scenario.” Evolution’s Unintended Side Effect To be clear, this isn’t a case of Opus getting worse. Quite the opposite. Its leaps in coding and reasoning are undeniably brilliant. The problem is that marketing is a domain that fundamentally rejects pure logic. The older, slightly “weaker” Opus actually wrote marketing copy with a much more human touch. Because it didn’t possess hyper-advanced reasoning, it didn’t try to turn every single selling point into a rigorous, air-tight argument chain. It didn’t format every emotional benefit as a technical spec sheet. Its structural imperfections ironically made it sound more natural in marketing contexts. Now? The smarter it gets, the more it reaches for logical frameworks to solve every problem. Ask for a landing page, and you receive a well-argued, logically cohesive thesis. But people don’t pull out their credit cards because they’ve been logically cornered into a purchase. They buy because of emotional resonance. Evolution in one dimension has created a regression in another. The more a model optimizes its reasoning capabilities, the more it treats every prompt as a logic puzzle to be solved. But marketing isn’t logical persuasion. It’s empathy. What This Means for the Future This raises a fascinating question: How far can the “one-agent-for-everything” model really go? If models keep pushing the absolute frontier of logic and coding, will their performance in “non-logical” domains like marketing, storytelling, and persuasion continue to deteriorate? Maybe the model didn’t actually lose its touch. Maybe its “logic persona” has just grown so dominant that it completely overpowers its “empathy persona.” Sheldon can learn quantum physics in an afternoon. But he never quite figures out how to talk to people. We might be watching our AI models do the exact same thing.