@sebdedeyne@AmpCode I decide per task. If I'm having a conversation in a thread trying to come up with a plan I have that thread spawn new threads to do the actual work. If I'm just knocking out something quick I just use a single thread.
Codex cloud agents feel half-baked. No easy way to enable signed commits, no global MCP servers or skills. Being able to close your laptop is only a small part of what makes cloud agents so powerful. @OpenAIDevs
@sqs@mweinbach@AmpCode I’m pretty sure they still have a lot to learn from Amp. The only compelling feature is the “everything in one app” appeal, but orbs are still too useful to switch to Codex Cloud Agents.
@QuestTheLine When BW was introduced to hunter in vanilla it was one of my favorite classes. Probably going hunter then. I’ll probably spend between now and release changing my mind on the race though
@ryancarson@DevinAI@linear@HelloUntangle I want to try Devin so bad to see if it’s usage limits work for me but I’m not willing to give up my $200 ChatGPT account without knowing 🙃
@Chris_Wozniczek i agree with this in concept, but devin's $200 plan cant match chatgpt or claude token for token. saying just use SWE-2 because it's free (for now) doesn't solve the issue with frontier model usage limits.
@morgantepell@dabit3 I feel this same way using amp orbs. I recently tried cursor and it definitely feels behind. I still want to try Devin at some point