Every single data fetching library seems to make the exact same mistake:
- Realize that nested data retrieval can be reliably decomposed into multiple retrieval calls
- Settle on this approach in the library design and build around all the tools around it
- "Oh right, we need support for mutations too"
- Try to extend the existing fetching model with a 'mutation mode'
- Fail to realize that nested mutations *do not* decompose into multiple mutation calls... and end up with a completely bolted-on mutation API

Is this like, the data fetching API design equivalent of "forgetting to put on the heatshrink before soldering the connector every time", or something?