Related to this, for our use case, setting thinking to high instead of low made tasks complete faster and cheaper (Gemini 3.0 flash).
Other aspects are caching, often at 0.1X cost, where providers really differ in how efficient they are (Anthropic really good, Google not so much) and how chatty a model is (costing output tokens).
I really like this but feels like it does not really work on a Swedish keyboard layout (macOS), cannot input '$'. Tried both Alt+4 (how you actually input '$') which works in vim and Shift+4 (US layout).
While the accessibility tree is great in many aspects it has its own limitations for example when it comes to stacked views or lazy loading outside the viewport.
That is why I like the idea of having as much of the documentation as possible in code. Tests that describe how things are supposed to work, infrastructure as code the describes the parts of the system and so on. Then you are forced to keep them up to date.
Could you maybe make use of Simon Willsons [LLM lib/app](https://github.com/simonw/llm)? It has great LLM support (just pass in the model to use) and records everything by default.