wiz
@wiz
- wake up, turn on agent mode, auto approve all requests, go make cup of covfefe PROMPT: "we need to [DO SOMETHING]. i added this block of code that outlines the functionality to implement. this requires interacting with multiple parts of the application stack so please review how it is implemented for [THIS OTHER FEATURE] before starting. you will need to design a [GIVE HINT TOWARDS DIRECTION] to [TASK DESCRIPTION]. follow existing conventions" - 5 min later new feature is done. start code review. it works without any changes on my part. no details missed, creates db migrations too, adds metrics, and does all the tedious stuff that i didn't mention in my prompt so i am very happy. - go on a walk because i feel like i accomplished a lot today
1 reply
0 recast
4 reactions
seneca
@seneca
what’s your workflow like? im only now getting into this stuff
1 reply
0 recast
1 reaction
wiz
@wiz
literally the above but to get there requires a little bit of setup. i recommend cursor and reading all the docs on context (especially rules) https://docs.cursor.com/context/rules code is much higher quality when it has existing implementations to reference and has as much information in context as possible. i don't let it design architecture because it is garbage at that so that is mostly what i spend time thinking about. o3 generates the highest quality output rn but sometimes gemini-2.5 comes in handy if o3 gets stuck (which is rare). you still need to review code because sometimes it goes down the wrong path. add any little quirks you notice the model doing that you don't like to cursor rules. sometimes its worth writing a short spec for the model if it's a more complicated task, or telling it to think in a certain direction take all these on the surface tho, you'll find your own flow the more you work with it
0 reply
0 recast
0 reaction