Super Agent Not Ready for Primetime
building now
Chris Keseling
I've been watching a ton of videos and community events to try and get a sense of actual use cases. In the end, most ended up being marketing hype more than anything OR "we use agents to do x, y, z", but never showing them in action.
I finally decided to give one a shot last week. I started with a template (personal assistant I think). After some iteration on the setup, I thought I was set. Each morning it would send me my focus items for the day and a list of my meetings.
Well it changed the format from the first day to the second and then again on the third (with no prompting). I decided I liked the second day the best and asked it to take what it gives in the chat and create a parent task with sub-tasks each day so I could actually check something off (the checklist in chat is not interactive).
EVERY day since, it messes up something. The tasks gets created and it's in my Overdue section. Bot says "You're right...let me fix that" and says it'll be correct moving forward. Sometimes it's missing the subtasks, other times its missing tags, other times the meetings. When I call it out, it tells me I'm right and that it'll fix it and update it's preferences, but it continues to not be consistent and therefore not reliable.
I've burned through 4K credit this week because the agent can't lock in and stay consistent with it's instructions.
As the company's ClickUp "champion", I can't recommend this to anyone else on the team at this time. It's all great in theory but it's wasting more time and effort in fighting with the agent to get things right than is worth for the output it provides.
Intrigued with the concept (I use AI for a lot of things), but disappointment with the experience so far.
Lastly, I found it interesting when I had a call with some of the product team this week when I asked if/when do the credits renew, neither had an answer. I still don't know if they renew monthly or we have what we have and when we run out we have to purchase another batch of credits.
Log In
Michael Van Doorn
Hey Chris Keseling
Thanks for the feedback
The team is actively working on agent reliability, instruction following, and speed right now (In person with ~20 folks as I type this). We'd love to take a deeper look into your agents to identify where we can improve.
Sending you an email now!
Michael Van Doorn
Merged in a post:
Super Agents not so super
Tevia
I found using it incredibly complicated. I watched youtube videos and spent over 30 minutes communicating with it to try to get it to do something. One question always led to another question with no actions. Finally I thought I got it to do something but in the end it did not do what I asked and actually bothers me more than helps me as I get notifications daily that I do not want. You need a coach just to try to figure out how to use it. I gave up on it and would rather have it deleted from my workspace. I am not a new Clickup user as I have been using it for over 5 years.
Michael Van Doorn
marked this post as
building now
Michael Van Doorn
Hey Chris Keseling
Thanks for the feedback
The team is actively working on agent reliability, instruction following, and speed right now (In person with ~20 folks as I type this). We'd love to take a deeper look into your agents to identify where we can improve.
Sending you an email now!
V
Vincent D'Amico
Michael Van Doorn is this related to some of the shortcomings of Open AI falling short with their model.
Obviously since your so integrated with Open AI it makes it frustrating since Anthropic's Claude would be a better fit.
Chris Keseling
I'm done with "super" agents for now. Every single day it messes up, we go back and forth (it gives me the "You're right I messed up and to be frustrated" line), fixes whatever the issue of the day is and says it has updated it's preferences. Then the next day its either the same issue, an old issue that was previously fixed or something new. It it was a real coworker they'd be on a PIP and likely not part of the company much longer.