Ask HN: Request limits vs. token limits for AI-powered apps?

Currently, I’m working on a web app for managing documents, databases, and whiteboards—the typical app that aims to be like Notion.

However, right now I’m facing the dilemma of creating a plan with AI usage limits, since the idea is to make it more agentic: able to edit and query context across an entire workspace and move it into a document, for example, maybe draw something on a whiteboard, etc. Still, I have the feeling that consumption could easily get out of hand. I plan to use DeepSeek for AI chat, but use Gemini 3 Flash for agentic usage and editing because it’s more intelligent. Lately, I’ve seen how many core AI apps have shifted their pricing models from per-request pricing to a fixed usage limit, but I’m unsure whether that’s frowned upon, creates a less pleasant user experience, or maybe even gives the feeling that you’re not getting what you paid for. So I’d like to hear some opinions on what decision I should make.