@simonlast Also keen to hear how you deal with/verify visual UI changes? Do you integrate with figma and do manual testing or something completely different?
@mattpocockuk Correcting the agent when it says "the simplest fix is..." or "the quickest solution is..." to ask for the best technical solution instead.
The "simplest" is almost never the right solution and usually leads to you and the agent getting stuck.
@mattpocockuk Once I get to a decision/solution (I don't think in terms of "plan" anymore) I persist the document and it forms a reusable context for other chats. That's usually the end of that conversation.
I'm always just building context for the next step and there may be many steps.
@mattpocockuk I'm giving agents less and less context up front. I don't even tell it what I'm about to work on.
I start with "fact finding" - I ask indirect questions which builds up it's context about an area of the codebase.
Then I begin the discussion about what change I want to make.
@mattpocockuk Do you have a rough idea of how many tokens you're using per loop? I've been getting this all working with cursors cli agent using opus and I've seen it run up 2M tokens in a single loop. I'm not sure if I'm doing something wrong 🤷♂️
@hardy27@mickynufc93 He's got the quality but he's looking for an easy route every time. Needs to stay on his feet, put the effort in and work hard. His head isn't there.