What surprised me isn’t that Claude is handing out API credits. It’s that the author is already talking like a budget allocator: one model for writing, another for review, another for orchestration, and a willingness to move workloads around based on who is cheaper or better at a specific task. That feels more like how teams will actually use these products than the marketing versions we usually get.
The Claude credit angle is the part I’d be most cautious about. “$50 to $200” sounds generous, and if it really recurs every month, that changes the economics of small automation quite a bit. But the write-up is one person’s experience, and the numbers are workload-dependent in a way that makes them hard to generalize. Burning about $40 a day on information distillation and data extraction is believable, but it also tells me this kind of setup can get expensive fast once you stop doing toy demos. Free credits don’t remove that reality; they just postpone it.
What I found more interesting is the Antigravity bit, because it reads like a quiet admission that Google’s AI stack is still uneven. The web-side experience apparently hasn’t improved much, and the real-time search remains weak enough that the author prefers incognito in some cases. That’s not a flattering comparison. If that’s accurate, Google still has a product coherence problem: the underlying models may be competitive, but the user-facing experience doesn’t feel trustworthy or consistently better.
The VS Code-based editor side sounds more promising, though I’m not sure I’d share the author’s confidence that it “can replace Sonnet 5.5 in certain scenarios.” Maybe it can, for the kinds of everyday feature work and migrations they describe. But code review is where these systems tend to reveal their limits, and the fact that one review consumed half of the weekly Claude quota is exactly the sort of detail that makes me think the ecosystem is still being shaped by quota management as much as capability.
The most convincing part, honestly, is the cross-review idea. Write with Claude, review with Gemini, and let the models catch each other’s blind spots. That sounds less like hype and more like an actual workflow insight. I think that’s where this whole market is headed: not “which model wins,” but “which combination is least annoying, least expensive, and least wrong for this exact job.”
Reference: $50–$200 in Claude API credits, and Google's Antigravity
