do I need to loop agent tasks so that the exit condition runs through ALL meeting attendees before sending the slack message? even though I set the exit condition for the agent to exit once ALL attendees were researched, it still is slacking me after the first one, then running through the rest, which means I got three slack messages summarizing:
The first attendee
The second attendee
All the attendees
yeah the LLM might make mistakes here → more robust to use a quick loop
all good… I had wondered if that month be necessary… what’s the best way to loop an agent then? got any tips?
and where does the “swarm” come into play for these kinds of workflows?
also, how would one add an existing agent into a loop?
you can’t copy and paste it into the loop at the moment
not sure I understand the question :S are we talking about the AI Agent step? Or Looping?
…both… in this Lindy, what’s the best way to loop an agent to run research on all meeting attendees to then get that full list of researched attendees via slack
i think you need to have your agent inside the loop then the slack message prompt be output from the loop no? That's what you had in your presentation wasn't it?
with a platform like Zapier or Make, there’s no concept of looping an agent and keeping all its output in the context window for a future single output… you’d have to instead load into Google Sheets or something like that
in make, you'd call a scenario then use an iterator/aggregator and return all that data via webhook response (or scenario output, I forget what they call it)
similar concept here, the enter loop is your iterator, exit loop is your aggregator
and the legendary Michael S. enters the chat! good call… I think that’s what I’m wondering on the aggregator
does the exit loop work that way outta the box?
in my preso, I hid the fact that, as built, I’m getting three, separate Slack messages
I've got 5 min, quick zoom?
i assume it works that way out of the box if you put in the format of the output as something like you had in your prensetation, but I'll test it tonight
but I think I see what you’re saying… exit loop acts as the aggregator… which I’ve used in other Lindies, just wasn’t sure if the Agent throws that for a “loop” ( 😅 ) and acts differently
it shouldn't, I think your agent's exit conditions will be something like you've found been through every step to find the relevant information
thanks for the demo today Rex H.! seeing your post here - looks like the 1 message was coming from 1 shotting agent
calendar >> agent >> slack (3 different slack messages)
were you also getting 1 message when you did loop?
enter loop >> research, email, summary >> exit loop >> slack message
^ this one seems cleaner as you can set the instructions in the loop
re: exit conditions - wonder if keeping at auto complete or simplifying could help
i've used custom starting conditions for loop starting, but haven't touched exit parameters and it's been working as expected
thanks, Michael! here's the agent I've got looped:
this is more for Marvin A., but here's one of the tasks that got stuck: https://chat.lindy.ai/ultra-agentic-meeting-prep-6806aa1e406954b995899450/tasks?task=6808e9db24cd54257f5546c1&e=68083097df85b90a265ff56c
michael r. - this is the "non-agent" Lindy I was getting three, separate slacks from: https://chat.lindy.ai/home/?templateId=680949524472cc8e083f96c9
the steps seem all over the place and I don't know what "Email processing" means here
any thoughts from the crew here? hopefully we can all figure this out together!
Here's the prompt I started using: You are a meeting prep researcher. For one attendee's information, you will look them up on Perplexity, find their LinkedIn profile via Google, scrape their linkedin profile, then look up email history with this person. From all of that, you'll create an output summary in the following template for each person: - Name and title: the first/last name and their role title, if found. If not title, leave it blank - Company and company summary: if not found, leave it blank - Professional Bio: if not found leave it blank - LinkedIn profile URL: if not found leave it blank - Summary of the last 5 email conversations leading up to the meeting: if not found leave it blank If any of these are not found, simply say "not found" after the colon for each line item. Here are the variables you'll want to use for each task {attendee name} {attendee email} {attendee company taken from the domain part of their email address, ignore if its gmail or a similar person email domain} What I found is that if the exit conditions are not met and are continually not met, it appears to freeze here's my less strict exit conditions: you've attempted each tool for the user
ooooh… nice… have you had much luck with the newer exit conditions?
and you just did an agent with those skills, inside a loop or not?
🙏 🙏 🙏
alright, this new "exit condition" prompt worked like a charm to fix the agent getting hung up and stuck inside the loop... I did, however, notice some odd behavior
for instance, in the UI, even after I got the slack message, I'm still seeing a spinner on "Loop Entered" in the Task List
also, it only shows that it processed one email, then I get two outputs (Danny and Adam) in different formats
the different formats I could see just being an agent not adhering to the instructions, but is the missing loop in the UI just a bug?
page refresh fixed it 🦾