tbh i was super worried about the exact same thing when I started my FastAPI setup last month. I've been really happy with how it's turned out tho... I ended up sticking with the IDE approach just because the CLI felt like a token vacuum if I wasn't careful. If you're trying to stay on budget before your launch:

• stick to Cursor AI Code Editor Pro Subscription for the day to day stuff since the monthly fee is predictable
• use the Anthropic Claude 3.5 Sonnet API only for the really tricky async refactors
• keep a close eye on your usage dashboard so there's no surprises on your bill It works well for me and I haven't had any complaints about how it handles dependency injection. Honestly, keeping it simple is probably better for your stress levels right now lol. I'm pretty satisfied just staying in the IDE where I can see what's happening and not blowing the budget...

Building on the earlier suggestion, I've found that Claude Code CLI definitely has a leg up on complex FastAPI migrations because it sees the whole dependency graph. Cursor is great for writing logic, but when you're changing a Pydantic model that affects ten different endpoints, the CLI handles that ripple effect way more reliably. Ngl though, the token burn is a legit concern when you're indexing everything right before a launch. If you want a budget-friendly middle ground, Aider AI Coding Assistant is a decent option. It gives you CLI power but lets you manually pick files for context so you aren't dumping the whole repo into every prompt. It handles async refactors well and understands those backend patterns just as good as the official tool. Basically comes down to whether you want full automation or more manual control over the spend.

This is exactly what I needed to hear. Youre a lifesaver honestly.