@lgrammel@mattpocockuk Will assume that better input leads to better output. Personally, I noticed that it's better than writing the rules in the agents' MD files.
When you encounter something new, and it does not 'click', in most cases, it happens because you likely do not have a mental model to process it.
Some people learn fast because they have built mental models, knowingly or unknowingly. For example, a new distributed systems paper lands, and they immediately map it to the trade-offs and patterns they have seen before.
Here is how to build the right mental models:
1. Start with first principles, of course :-)
When you learn something, ask not just "how does it work" but "why does it work this way." Understand the problem it is solving, the constraints, and the trade-offs that were made. For example, Redis chose push-based replication because of its speed-first philosophy. That is not a fact to memorize, but a model to carry forward.
2. Teach it.
Writing or explaining something forces you to find the gaps. If you cannot explain it clearly, you do not have a model and understanding yet; you have a collection of loosely related facts.
3. Build small prototypes.
Abstract understanding solidifies into a model only when you make it concrete. One prompt will help you build a quick prototype and demonstrate the concept. This will help you move from "I read about this" to "I know how this behaves."
4. Connect it to what you already know.
Every new concept you encounter, ask where it fits. Is it a variation of a pattern you have seen? Does it contradict something you believed? Try to place it in relation to your existing knowledge - and this is how the model forms.
Over time, this will compound. Each new mental model helps you build the next one. More importantly, you stop re-learning the same class of problem and start recognizing patterns and structure faster.
To be honest, the goal is not to know more things. It is to carry fewer, stronger frameworks that explain more things.
You will always start confused, but mental models will not let you stay in that state forever.
Hope this helps.
We've moved our entire code review lifecycle onto Automations.
Cursor will approve and unblock changes that are low risk and automatically assign reviewers based on commit history for changes that are medium/high risk.
There is so much power in transforming historically static processes (*all* changes must be reviewed by X people) to dynamic ones (# of reviews scales with risk of the change).