One influencer campaign CSV, two AI-built client dashboards

One CSV from a real influencer seeding campaign: 21 creators across Singapore and Malaysia, each sent product and promised a Story or a Post. Creator names and handles are replaced with ids; the figures are real. I gave two AI agents one brief: turn this into a dashboard I could send the client, in one HTML file, sized for a 1080 x 780 window, that comes alive on its own.

Claude Fable 5.1 ran in Claude Code, GPT-6 Astra in Codex CLI. One attempt each, no retries, no coaching, a fresh empty folder per run with the same CSV in it.

Both produced a working dashboard, and they look nothing alike. Fable built a dark operations board with a flow of promised against delivered, a follower drift chart and a spotlight that cycles through all 21 creators. Astra built a light, branded report with the full creator table, and it refused to call follower counts 'reach', saying so in a footnote. On this data that was the better call.

Fable took 35 turns and 12 minutes and billed $5.90. The reel shows $4.91, from my older script that under-counted Claude runs. Astra answered in one pass in 3m22 for $0.73. They ran in different tools, so part of that gap is the tool and the number of turns, not the model.