@mntruell@mardehaym That feels not fair. If I will launch some opus for hours and then make a post that it didn't work - nobody will return me money, and difference will be just because it's some enterprise users with more public impact ๐
@jancurn@apify Yeah, I understand, but practically only reliable way to do that properly is to put agent inside actor, and it makes it expensive and unpredictable. Or I can hardcode calling one particular mCP by tool name, but then I need to babysit that integration for3ver
@jediahkatz Problem is that in reality it watches terminals pretty bad and needs constant nudging. Discovered during last few weeks that codex can just launch app and never forgets about it and just keep working on issue. Composer in Cursor just forgets about running app in few turns.
I really like how ActorHunt improves how I build Apify actors. My agents can use it to validate ideas by keyword data, compare competitors, and get better Store context before building. If you are also automating Apify actor building, it is definitely worth trying.
@fredrikalindh Yes, already found. But it doesn't want to use it without nudging, (maybe system prompt diff per model) and for me stuck on trying to focus input on login screen, while gpt5 doing this without problems for last months
@leerob Indeed, if directly asked - it says it has tools, but it doesnt use them without being told to.. Like if i spawn gpt and say 'add button' - it will provide PR and video of button etc. Composer just adds code changes. Is this diff in system promt per model?