AI UXFailures

Open any AI product launched in the last two years. You'll find the same interface: a text input at the bottom, messages scrolling up, maybe a sidebar for conversation history. They all look the same because they all copied the same template.

The Template Trap

When ChatGPT launched, it set the template. Every competitor copied it. Not because it was the best possible interface, but because it was familiar. Familiar is safe. Safe doesn't require design thinking. And so an entire industry converged on mediocrity.

The chat interface works for general Q&A. But AI does more than answer questions. It analyzes, compares, plans, creates, and decides. One interface can't serve all of these modes. Yet that's exactly what every product ships.

If your AI interface looks like every other AI interface, you didn't design it. You copied it.

What They Get Wrong

• Walls of text with no visual structure
• No way to compare options side by side
• Conversations that lose context as they scroll
• Markdown rendering as a substitute for real design
• 'Great question!' and other condescending filler

These aren't minor complaints. They're fundamental failures to match the interface to the task. A decision needs a tree, not a paragraph. A comparison needs columns, not bullet points buried in prose.

The Way Forward

The fix isn't to polish the chat interface. It's to abandon it for tasks it was never designed for. Different questions deserve different visual responses. Persephonie starts with decisions because they're the most obvious mismatch between chat UI and user need.