👁👁👁
8K posts

👁👁👁
@JonHearty
▶️ exploring ai, building golfrise, investing in scifi (microfactory, cavalla, digital metal)👽🦾 ⏪ odf1 & early mercury, origin, datanyze, redbeacon; ucsb🚫🎓
california 参加日 Mart 2010
3K フォロー中3.1K フォロワー
固定されたツイート
👁👁👁 がリツイート

@LeahLibresco Something way cooler than a minivan is coming
English

👁👁👁 がリツイート


I do believe having it change driver profiles (where it drives faster/slower, more aggressively, etc) via those modes is definitely telling it how to drive, but I get your point.
I assume you mean "speed up", or "get over one lane", etc maybe? How would you tell it how to drive if you could?
English

@JonHearty You can literally tell your FSD where to drive you by talking to it (among other options)

English

Two things I learned from @steipete on the @lexfridman podcast that immediately helped me with my v̶i̶b̶e̶ ̶c̶o̶d̶i̶n̶g̶ agentic engineering:
1. Asking your agent "Now that you're done building, what would you do differently?"
It's wild how it will immediately find edge cases, bugs, etc. on a second pass after it's done building.
Fun fact: It works well in Plan mode also (in @replit at least). "Now that you made your plan, how would you rethink it?"
2. Remember that your agent is always starting from scratch.
You need to have some empathy for your agent and where they're coming from.
Every time you start talking to your agent, remember that they don't know your codebase or what your app does without doing some initial research.
A small tactical example: You can save you and your agent a lot of time by providing exact file names instead of vaguely referencing screens/pages (i.e. "the profile page").
English



