If a vision agent costs 45x more to serve than an API agent for the same task, what fraction of next year's marketing budget should you redirect from SEO to AEO? That's a real question now. The cost asymmetry is structural. The traffic mix is shifting. And no one inside the marketing team has a defensible number to bring to the CFO. This post tries to give one. Not a spreadsheet. A framework, with a starting band that you can argue with. The cost data, briefly The previous post in this sequence ( The 45x Argument ) walked through the Reflex benchmark in detail. Two agents, one task. Vision agent: 550,976 tokens, 53 steps, ~1,000 seconds. API agent: 12,151 tokens, 8 calls, 19.7 seconds. Cost ratio: 45x. ( Reflex, April 2026 ) That number isn't going down. Better models reduce error rates but they don't reduce step counts. A vision agent must render every intermediate screen to interpret the next move. The cost is paid by whoever runs the agent, which means: the buyer. Now the budget question.…