AI can now produce polished charts, dashboards, and infographics in minutes. But as generating visuals becomes easier, a more important question emerges: what part of data storytelling still requires a human?

This session presents the results of a personal experiment built around that question. Fifty infographics and data visualizations, created over the past decade for real clients across journalism, NGOs, corporate reports, and public communication, were rebuilt using Claude under the original project conditions: the same brief, the same dataset, the same audience, and the same design constraints.

Each AI rebuild underwent a two-stage evaluation. First, it was assessed using a predefined framework measuring five dimensions: time to insight, factual accuracy, visual hierarchy, audience fit, and introduced errors. Second, the human and AI versions were compared in blind evaluations, where independent participants judged the visualizations without knowing which version had been created by whom.

What this personal experiment revealed:
• The commodity layer: the parts of the design process AI already performs as well as, or better than, a human.
• A taxonomy of AI failures: the recurring mistakes that appeared across dozens of AI-generated visualizations, from context blindness and generic aesthetics to misplaced emphasis and fabricated precision.
• The new human advantage: the skills that consistently resisted automation
• The blind-test verdict: what people actually trusted, understood, and remembered when they didn't know
whether a visualization was created by a person or by AI, including the uncomfortable cases where AI was preferred.

This is not a debate about whether AI is good or bad. It's an experiment that looks at how data storytelling is changing, built from almost ten years of real client work reconstructed under the same conditions. The result is a practical map: what to automate, what to supervise, and which human skills have become more valuable than ever.