Don’t overload AI with instructions/ context
you are still the code owner, you must understand why AI made the changes it did otherwise you cannot spot errors/ over engineering
Strive for consistent and sensible patterns that AI can find / reuse
use unambiguous language when speaking with the agent and in terms of context do not be vague about your goals, consider planning mode for larger tasks
agree on requirements and keep a running list, at the end of the task you should reverify before signing off
avoid asking AI things like “do you think this could work” or “do you think this is a good idea” – you may bring out AI’s tendency to want to please you, instead ask “do you think this is logical” or ask it to break the possible solution down into pros and cons
After a solution is implemented ask the agent to perform a retrospective looking for possible pitfalls