David (00:00)
How Lindy ruined your date night. Okay. I can't wait to hear this.

Ilan (00:03)
yeah, ⁓ yeah, I was in the doghouse last night

Hey everyone, welcome to Prompt and Circumstance I'm Ilan

David (00:12)
And I'm David.

Ilan (00:14)
And today we're talking about Lindy.ai.

All right, on today's episode, you're going to hear about how Lindy ruined my date night. You're going to hear what's wrong with the tool. I'll also tell you what's right with the tool. What does work well there. And I'll do a quick comparison with ChatGPT's agent mode.

David (00:46)
Hey, Ilan, all right, so you've been playing with Lindy for little while. I'm curious to hear what went well and what didn't and how it compares to other similar tools.

Ilan (00:55)
I'm very happy to go into that. Now, let me start by framing this with the season that we're on, right? We're still working on discovery of this product that we want to build. And what I was thinking was that Lindy would be potentially a really good tool for helping to find the right people to talk to. The first thing that

you really need to do as we know as product managers, talk to your users, understand where are they coming from, try to dig into their needs.

David (01:28)
Now, guess, I guess the thinking is, that, ⁓ so Lindy, ⁓ positions itself as the first AI employee, right? And so this would be, ⁓ implying, course, that they, they make agents, which is, I believe what they, what they do. So that kind of a task where, Hey, find me information about, about these people. That's totally something that you would give to some intern, ⁓ maybe an intern with a PhD, several PhDs, that, ⁓ that you would expect them to be able to accomplish.

Ilan (01:50)
Right.

Ilan (01:57)
Now is a great time to tell you about our sponsor for this week, Querio. Querio is a data analysis tool which sits on top of your data stack and allows you to ask natural language questions about your product data. It gives you beautiful visuals and allows to really pull out threads so that you can get to the bottom of all of your product questions.

You can try it too. Just go to querio.ai and let them know David and Ilan sent you for two months free of Explore. That's a thousand dollar value for our listeners.

Ilan (02:31)
as you said, Lindy should probably have access to multiple PhDs and what I need is some basic SDR work. I thought this should be pretty easy. They've been growing like crazy. They hit 10x growth this year. It's a company that's been around for a couple of years.

Last last numbers I saw were around five million in revenue at the end of last year So they're probably you know in the 50 million revenue as of a couple of months ago

David (02:58)
Hmm.

I'm thinking about how some companies are ⁓ restructuring. And I wonder how much Lindy has had a hand in some of those jobs getting impacted.

Ilan (03:08)
Mm-hmm.

Well, let me tell you that it will not be an employee at Prompt and Circumstance

David (03:23)
hahahaha ⁓

All right. And now you were telling me about how Lindy ruined your date night. Let's dive into that. What's, what was, what was that about? What happened?

Ilan (03:34)
I've been in the doghouse, David. ⁓ It's not been fun. So.

David (03:39)
The sofa isn't so comfortable. ⁓

Ilan (03:41)
⁓

So yeah, especially this one that's right behind me.

David (03:48)
and

Ilan (03:48)
The issue that I had here is that Lindy makes it sound so easy to build an agent. And I had a lot of confidence in their hype and the marketing material. I read a few guides, watched a couple of videos of people making their agents. And I was like, no problem. I can do this. So what I did first is I thought, OK, let me think about this.

discovery use case. ⁓ Let me have this pull some information about about target customers.

David (04:28)
Alright, so let's set the stage here. ⁓ So you had an ICP defined, ⁓ and you're hoping for Lindy to be able to give you a list of people to contact who fit that profile.

Ilan (04:42)
That's right. And I really wanted to use their agent builder to accomplish this. And after a couple of minutes, I realized probably not the right use case for an agent, right? This is more like a bulk action. This is more like an agentic tool rather than a single agent who is going to accomplish a task over and over again, given a goal. So I pivoted away.

I really wanted to show how Lindy could be helpful though. So I was like, Hey, let me automate something for the podcast. I'm going to make it so that after each episode posts, we use a tool called Transistor to host our podcast. ⁓ I'll have Lindy find the episode, find the transcript and then recommend a LinkedIn post based on our history.

of LinkedIn posts, So something in our voice, ⁓ similar to other things that we've done, and send me an email with that recommended post.

David (05:47)
Okay, so

it would draft any ⁓ LinkedIn post for you.

Ilan (05:51)
Exactly.

Seems pretty simple, right?

A trigger, an episode is created, an action, take the transcript, another action, look at our previous history of LinkedIn posts, and then another action, generate a new post that highlights the value of this episode in the same style as our previous posts. And then finally, a conclusion, send me an email with that first draft. All right.

David (06:22)
That's

something that a university student would be able to do. I'd hire a college student to do that maybe. ⁓

Ilan (06:26)
Yeah,

exactly. I think you'd be overpaying if you hired a PhD from that. And Lindy gave so much confidence that this was possible. I went into to ChatGPT deep research to get it to give me step-by-step instructions. I went into Lindy. I tried to use their agent builder And my time box was...

David (06:31)
Yep.

Ilan (06:52)
had to get this done before dinner because I had a date night planned with my wife. So I was going to finish the Lindy workflow, get it all ready for the podcast and then go have dinner, play a board game, watch a movie and go to bed. And I come down to dinner and I'm like, listen, I'm so close. I am so close. This may be 20 more minutes after dinner.

David (07:21)
Right.

Ilan (07:21)
and

I will have finished everything I need to finish.

And guess what? Did not take 20 minutes. Yeah. So then I spent two more hours trying to debug this Lindy workflow while...

David (07:26)
⁓

man. some vibe

debugging level of time consumption.

Ilan (07:38)
My friend, this was

an unpleasant experience, let me tell you.

So with that in mind, let me give my unbiased review of Lindy and let you know what really didn't work about the tool, why I struggled so much. And let me also tell you what did work and where I did find that it was helpful. And, ⁓ you know, we can let the audience decide whether they want to use this tool.

David (08:09)
All right, let's dive in.

Ilan (08:11)
All right. Let's start out with Lindy's pricing. They do have a free tier. It gives you 400 credits per month. And we've talked about in the past how we dislike credits as a form of monetization because it's quite ambiguous. And you'll see a little bit of that as we get into the tool.

David (08:24)
Yeah.

Ilan (08:27)
from experience, you use up those credits pretty quickly. in my case, I had to upgrade to the pro subscription so that I could finish off this task that I wanted.

David (08:37)
So not only did you spend way more time than needed out of your date night with your wife, you also burnt $50 US on it.

Ilan (08:44)
Mm-hmm.

That's

right. Now I spent the $50 and I had reached out a few days ago asking for a demo of Lindy to see if they could show me how to use this tool well. And as a paying user, as a $50 a month user, no answer. So...

David (09:03)
As a paying user, you reached out.

Yeah.

Maybe that's for the business. like,

hey man, you got to upgrade to the 200 per month tier. Only then do we bother with you, you small potatoes.

Ilan (09:19)
That's right.

That's right, David.

But anyway, I was on the pro $50 a month tier. And I gotta considering tools are pricing themselves these days, know, $20, $25 seems to kind of a standard.

tier that people are offering. Now, it's pretty common to hear these days that a lot of AI tools are not profitable because the underlying models that they are using cost much more than these, ⁓ cost much more than they're making from their paid tiers.

So maybe Lindy is accounting for that, right? They wanna be a sustainable business. They wanna make sure that their agents don't just stop working one day, because the company runs out of cash.

David (10:08)
Now, do they explain what credits map to? Is it roughly tokens or is it something else?

Ilan (10:17)
They don't explain what credits map do. It is completely arbitrary.

David (10:20)
It's completely arbitrary.

All right. That's, amazing. Just a very opaque pricing system.

Ilan (10:28)
So going into the tool, what I was thinking was, OK, Lindy has a bunch of employees, right? That's the whole point of this tool is that it's like your agentic employee. So probably their agent builder will be all the help I need to be able to build the agent that I want to build.

I mentioned before that I had this workflow that I wanted to build, right? An episode gets created and then it pulls the transcript and then it ⁓ checks against our previous LinkedIn posts and then it uses our latest transcript to suggest or draft a new LinkedIn post. And finally that it would email me

that draft post. So the first thing here is a trigger, right? The trigger is from Transistor from this platform we use to host our podcast. When a new episode gets created, trigger the workflow. So I start out and hey, look, Lindy has a Transistor trigger built in. Very cool. I come into here and when you're setting this up,

David (11:14)
Mm-hmm.

Lovely.

Ilan (11:41)
You add an account. It gives you a little window. It tells you use pipe dream. It gives you a window to put in your API key. And then, ⁓ on the right hand side of the page here, it will give you your, Or it'll give you your account. And the first thing you need to do is choose a show within your account. Well, after three attempts and two different API keys,

David (12:02)
Hmm.

Ilan (12:06)
turns out you can't get the show. There's some unknown server error when you get the show.

David (12:11)
So you're stonewalled before you even start.

Ilan (12:15)
⁓ Now, start to talk with the agent, it gives me a bunch of steps to try out. And ⁓ today it's given me a ⁓ different suggestion, which was first to check the Lindy community, which had nothing, then contact Lindy support who doesn't answer, and then finally verify with Transistor that the API

David (12:24)
Mm-hmm.

Ilan (12:41)
key is working by testing it. And I did that. I did make sure that the API keys are testing directly with Transistor.

David (12:47)
⁓ okay.

Ilan (12:48)
Yesterday, it told me, you know what? You can just copy paste your show ID and put it into agent step and it'll work just fine. I tried that. It does not work. You cannot copy paste a show ID into here and have it just work. So herein lies the first two problems that I found with Lindy. One is they have all these triggers.

David (12:56)
Hehehe.

Ilan (13:17)
They advertise that they are connected with 5,000 different tools. But I don't think they've really tested those integrations very well.

David (13:21)
Mm-hmm.

just like a fly by integration. It's like, hey, what's the URL? What are the params? And go, we're done. Next. Yeah.

Ilan (13:29)
That's.

That's right. Exactly.

the tools don't work. And the second one is that the agent hallucinates. And this was really the crux of the problems that I had. Because I really tried to use this agent to help me along and get me there. Assuming that it was trained on their documentation.

So it would really be able to understand, right? What are the ins and outs of how to use Lindy?

David (13:59)
Mm-hmm.

That seems like an obvious thing to do. And also the, ⁓ when the agent builder came back and said, content lending support,

I would have expected, okay, where's the CTA to that? Right. Okay. Can I do it right now? Like, am I not right? And you would think also that maybe their support would be using AI agents, which would mean that they'd be very responsive.

Ilan (14:18)
Ha ha ha ha!

David (14:30)
I know, I'm kind of squinting my eyes thinking, ⁓ hang on. Does that mean that Lindy's not using their own tech?

Ilan (14:36)
I know, it does make me wonder. It does make me wonder. ⁓ Now, meanwhile, throughout this whole process, I was burning credits. So each request to the agent builder burns some credits. ⁓ There is no, even when you're chatting with it, yeah.

David (14:52)
even when you're chatting with it. Yeah.

See, which is, which is unlike, the vibe coding tools like Replet and lovable and bolt, right? Because when you put it in, put those into chat mode, you're not spending any of your actual credits. You're just talking about your code base, right? It's only when you build, at least last I checked, it's only when you build and write code, do you spend those credits. So this is, this is kind of uncool.

Ilan (15:17)
right, so then I put my tech hat on and I was like, how else might I solve for this? And actually this is where ChatGPT's deep research, which I used to give me the steps that I should take, offered another suggestion. And one of the things that it said was that it was not clear if Lindy's Transistor

integration existed or actually worked. So what you should do is use a webhook trigger. So I went into the documentation, I figured out how to make a webhook trigger work. I'm not going to get into that. You know, it kind of gets into some pretty nitty gritty technical details and technical setup

David (15:51)
Hmm.

Ilan (16:05)
But already this is a warning sign for somebody who may not be so technical and wants to build an agent to accomplish tasks within Lindy

David (16:15)
Yeah, exactly. mean, I'm thinking the same thing, right? Like go to a random marketing manager who wants to do something on the marketing side, which by the way, like it's marketing and sales that's seeing ⁓ significant impact from this AI revolution. ⁓ And yeah, so just go to a marketing manager and tell them to do this and ⁓ good luck. Good luck with them being able to get, make any kind of progress on this.

Ilan (16:22)
Mm-hmm.

Mm-hmm.

David (16:45)
So yeah, this is very far from the easy to use, no code ⁓ claims that some people have for Lindy

Ilan (16:52)
Mm-hmm.

So the next thing that happens is I need to go get the transcript. And again, here, because the Transistor integration is not working, I got to do that on my own. So back into the Transistor documentation, find out, okay, here's the API call that you need to make. It's actually pretty simple. I need to grab an ID from the webhook response and put that into a URL. And that will give me the episode information.

So I go ahead and do that.

And then I test and it doesn't work. And it fails over and over and over again. And

David (17:29)
Hmm.

And

it can't fix itself, right? It's not like, you know, some vibe coding tools, they'll say, I noticed that there's an error. Let me go try to fix it. Right. Did Lindy do that?

Ilan (17:36)
A Kif.

⁓ well, funnily enough, I told the agent that I had this problem. ⁓ in fact, you can see here where I gave it the information about what I was looking to do. ⁓ this whole JSON that I copy pasted and told that I need to retrieve this field and pass it somewhere else. And it tells me perfect. I've added an additional API call to your workflow.

Here's what I added and then it tells you all of the steps that it's followed. Except when I go into the workflow, actually nothing has changed. It has just hallucinated. It has spent at this point 60 credits and hallucinated its entire action. ⁓

David (18:29)
you

love your response there. The all caps response

of how the F do I make sure? The HTT request is URL included, encoded, and not JSON or XML. That's awesome. I can tell you were just really excited. That's what.

Ilan (18:41)
Mm-hmm.

That's right.

I was just really excited.

This was T plus, you know, one hour after dinner This should have been my ⁓ hint that that was probably the time to shut down the computer and move on.

David (19:09)
It's one of those things where it's the sunk cost fallacy, right? It's like, look, I've already burned an hour of my date time with my wife. This better be worth it. I'm going to make it worth it by investing another hour.

Ilan (19:14)
Mm-hmm.

Right.

Exactly.

And then this points to the last problem with Lindy, which is that

they over-rely on AI in my opinion for the various functionality.

David (19:35)
Mmm.

Ilan (19:38)
So when you go into this HTTP request, which again, I don't think I ever should have had to use. I never should have had to figure out, but I did.

I go in and there are only specific fields that I can manually set. And for a lot of them, it wants you to prompt the AI

tell it what to do from your previous step with the information from your previous step. And it's like, don't worry, the AI will figure it out. But this is not probabilistic, right? It's not like, ⁓ if this magical thing happens in the previous step, then I want you to do something else. This is deterministic. Like, Hey, I know that I'm, it's always going to be like this.

David (20:17)
It's always going to be like this.

Ilan (20:20)
So I've been using no code tools and workflow builders for years now, well before the AI revolution. And making an API request is while technical, a pretty standard piece of functionality that these tools have. And there are different ways that you make an API request.

There are different fields that you might have to have filled out. This is settled ⁓ functionality, but Lindy doesn't give you access to all of the fields or all of the ability to set these parameters deterministically. They want AI to figure it out for you. You know, on an old episode with Rami from Querio, he mentioned this about their tool.

It's also an AI agent and they had to UN-AI the product.

Rami (21:16)
I think it's really easy to be a bit lazy and use a bazooka to destroy a paper and use an LLM for flows that just don't require

David (21:24)
Yeah, that makes a lot of sense. know, the, this craze to add AI to everything. it, ⁓ it really has made people lose maybe some touch with, the, like the original sort of product management sense of things, which is, ⁓ what's the best way to solve this problem for the business. Right. I mean, sure. You could, you could solve it.

Ilan (21:49)
Mm-hmm.

David (21:52)
in a way that loses money for the business, but that's then you're not doing your job. So ⁓ yeah, like the fact that they really over indexed on AI and that, you know, like we're seeing that perhaps elsewhere is, I think something that our audience could take away.

Ilan (22:11)
Now, one thing I will say here is that maybe this was a business decision from Lindy because every time that they are calling a, an LLM on your behalf, they're spending your credits. So, ⁓ this is maybe the opposite problem where Lindy is thinking about the business viability over the customer value. and you know, putting

putting ⁓ spending your tokens or spending your credits over, making it easy for the user.

David (22:46)
Yeah, that may very well be. Yeah, personally, I'd be worried about retention if they do that. So we'll see where they end up a year from now.

Ilan (23:00)
With that, David, I will have to say wasn't all bad with Lindy and it actually did accomplish a couple of tasks that I needed really well. And so let me talk about those.

When you first come into Lindy, it's going to give you this window where it says build an agent or perform a task. And so we've gone through the build an agent side and my thoughts on that. But let me tell you the perform a task is really good in Lindy. It's comparable to ChatGPT's agent mode, but I would say that

David (23:35)
Hmm.

Ilan (23:40)
Lindy is really well connected to the types of tools that it needs to accomplish these tasks and accomplish them pretty quickly.

David (23:51)
What would be an

example ⁓ of a task as well?

Ilan (23:54)
So concurrently with the building of the agent workflow, I asked it to extract the posts from our feed page on LinkedIn. And it did it really well. I'm going to scroll a little bit here and show you.

David (24:11)
Hmm.

So for those listening, there's a conversation back and forth and it looks like Lindy provided a markdown document with all of the posts.

Ilan (24:25)
Exactly.

So at the end of it, ⁓ it got 13 of our posts, which was pretty good. It was good enough for me. And exactly as you said, it provided it in markdown format. So I was able to copy paste this into Notion and I had my ⁓ first version of the document that I needed. And actually it was ⁓ a great win for me otherwise, because I think that we should be maintaining

this history in a way that's easy for us to access via LLMs in the future.

Let me talk about another example where this agentic tool that they have did really well was I wanted to find people with specific titles or types of titles from series B startups because we

have a hypothesis that we want to test. And these are the kinds of people who we want to talk to. And we don't know a ton of them just out of our network. So I figured, hey, let's go out and figure out who's out there.

David (25:21)
Mm-hmm.

Ilan (25:28)
The agent tool did an amazing job at giving me ⁓ just a list of folks who I might want to talk to and.

David (25:39)
And

I like that it actually asked you some clarifying questions first, even though you did not prompt it to do so.

Right. So there's questions such as how many people are you looking to find in total? And are you targeting any specific geographic regions or industries?

Do you have a preference for company size that is employee count range? So, I mean, those are all great questions to ask before going on this kind of a search for the people who you're looking for. I think that's quite good.

Ilan (26:12)
It was, and it did a great job.

And so here's a snippet of what it did. It gave me more than this, but it found people at Series B startups. It told me what's their job title, what company are they at, the company size, the location. And it gave me their LinkedIn URL and even tried to figure out their work email and personal email, which it's a little scary.

David (26:37)
Hmm.

Ilan (26:42)
information is available out there from lists that Lindy has access to. But this really gave me the idea that what Lindy might be best for is sort of sales outreach marketing material. And it did a great job in another case testing out an idea we had for a tool for teachers.

It gave me a list of a hundred people who we could potentially reach out to as well as creating the outreach plan So this is where I think that the tool really shines is in its ability to serve.

those kinds of more sales-focused, marketing-focused users. And I did compare this with the agent mode in ChatGPT The Lindy response took maybe a couple of minutes at best. It has access to a bunch of tools, has access to a bunch of information already. These are some of the pre-built connectors are to sort of lead scraping sites.

Uh, on the other hand, ChatGPT took well over an hour to return me a list of 10 people. And the list of 10 came from three companies. So, um, you know, Lindy was a little bit more understanding that, Hey, I'm looking for a broad range of folks to speak to. Whereas, uh, ChatGPT's agent mode focus very deeply on, uh, specific companies that it was able to find.

David (28:01)
Hmm.

Ilan (28:18)
information on and target all of the engineering people in the org at that company.

David (28:24)
interesting. Okay, so I was wondering what GPT spent the hour actually doing. I mean,

Ilan (28:31)
It's

spent a lot of time doing research online about Series B companies.

David (28:36)
to find which is the right Series B company. Even though, I I assume you gave it the same prompt, which is basically I need to find people with the following titles in Series B startups who have received funding in last 90 days. So, you know, there isn't a whole lot to really think about, I'd imagine. ⁓ So it's interesting that it's spent an hour on it. I mean, that's an hour of it.

Ilan (28:39)
That's right.

Mm-hmm.

David (29:05)
⁓ generating text that imagine somewhere in the back, right? Through tool use and so forth.

Ilan (29:11)
Yeah, going into ChatGPT, I gave it the exact same prompt. And you can see where it was thinking that it was reading through web pages of different companies and reading through news articles, trying to understand ⁓ what companies have raised rounds in the last 90 days. So.

David (29:16)
Mm-hmm.

I see. Interesting. Yeah. I mean, maybe Lindy has access to some special tools too that GPT doesn't, right? Like maybe, ⁓ you know, some kind of ⁓ tool for, what is it? Crunch base? Yeah.

Ilan (29:47)
That's right.

So with that, that's the good, the bad, and the ugly of Lindy. To summarize, I don't think their Workful Builder is that great. It will not be an employee at ⁓ P and C Podcast and...

we probably will not be renewing our subscription to them. We will be a churned customer.

David (30:08)
Yeah. I wonder whether, whether an agent would reach out to you to try to save your turn.

Ilan (30:13)
That's right.

On the other hand, their task mode is pretty good, right? It does burn through your credits, but if you're looking for customers in your ICP that you could chat with, that you could reach out to, Lindy will do a really good job at that. And I'm sure that there are more sales focused automations.

that their agent builder could even do well at.

David (30:44)
Yeah, yeah. Well, thanks for walking us through that, Ilan You know, I did a bit of ⁓ research on sort of the market response to Lindy in terms of people actually using it, people who are building AI agents and so forth. And the reviews are mixed. There are some people who say that, ⁓ you know, just try N8N or put something together in Zapier. Those seem to be the go-to.

for some people who are doing the kinds of things that you might be looking to do, which is to give them the trigger, do this, access this API. It sounds very much in Zapier's wheelhouse. ⁓

Ilan (31:31)
Absolutely, and Zapier has done a lot with their ⁓ agentic tools. So maybe one for another episode in the future.

David (31:39)
Yeah, I'm looking forward to it. Well, that's it for for today's episode. ⁓ Please remember to like and subscribe and give us a review. You can find us on the socials at @pandcpodcast

Ilan (31:54)
Let us know if there are other tools that you think could be useful in discovery that you'd like us to test out.

David (32:00)
Alright, we'll catch you at the next one.

Ilan (32:01)
See you next time.