EVERY SPOKEN WORD
80 min read · 16,316 words- 0:00 – 3:16
Why AI agents can be 100× cheaper and better than humans (and why adoption is still low)
- AGAakash Gupta
What if AI agents could replace members of your team? The hottest startup for building AI agents right now is Lindy, and we have on the CEO, Flo Crivello. And we are going to show you how you can make tons of money by building AI agents in twenty twenty-five.
- FCFlo Crivello
I think people should increasingly, with AI agents and with the rise of AI in general, they should increasingly think themselves as orchestrators and not as doers. I just think with AI agents, it's like everybody just got a promotion to manager.
- AGAakash Gupta
People might be asking, "Well, what types of integrations do I need to do? How do I integrate my CRM? What does that look like?"
- FCFlo Crivello
Like, we have six thousand integrations, so pretty much any tool that you want to use, you can use it in Lindy.
- AGAakash Gupta
What are the things you should be thinking about when you create an AI agent so that it isn't going out there and potentially ruining your brand name within your company?
- FCFlo Crivello
That's where human in is important. You can double it down, and you can be like, "Before you do this specific thing, you ask me for confirmation."
- AGAakash Gupta
LLMs, they tend to be unreliable. There tends to be some lossiness.
- FCFlo Crivello
I think LLMs have a worse reputation than deserved because they have improved so rapidly. They don't screw up much more often than a human would.
- AGAakash Gupta
You hired perhaps the most famous engineer right now-
- FCFlo Crivello
[laughs]
- AGAakash Gupta
... Soham Parekh. If you haven't heard of him for working, like, five jobs, tell us the story of hiring him.
- FCFlo Crivello
He was really good. Within a week of joining, he started making excuses. I woke up one morning, and I saw that conversation on Slack. That was like, "Wait a minute. Like, is, is this our Soham?"
- AGAakash Gupta
What it's like to use the agent builder or the flow editor.
- FCFlo Crivello
So you literally just click Start from scratch, and I'm just gonna... You just talk to it in English. And so I'm gonna say-
- AGAakash Gupta
Flo, welcome to the podcast.
- FCFlo Crivello
Yeah. Thanks for having me, Aakash.
- AGAakash Gupta
So you recently went viral on LinkedIn when you said that Lindy agents can be a hundred X more effective and cheaper than humans. Is that true?
- FCFlo Crivello
It's true for certain roles, for su- for sure. And I, I, I feel like it should hardly be controversial. Like, we accept that machines are, are, by and large, like, a thousand X better. Like, a calculator is just, like, infinitely better at calculating than a human, right? Uh, a computer is, is just millions of times faster at whatever it does than a human. Um, I, I think it, it's just a matter of this range of things at which computers are better than humans is just extending over time, and it's just... In a way, that's just a continuation of, of the march of technology over the last couple of decades. So absolutely. Like, the things you can ask AI agents to do today, and there's many of these things. There's, like, thousands of use cases, like customer support, like sales prospecting, recruiting and whatnot. Absolutely, AI agents are going to be way, way, way better than humans.
- AGAakash Gupta
I think that you and I understand that because we're in the day-to-day building and using AI agents. I polled my Slack community of working product managers and tech professionals. I asked them, "How many of you guys have built or used an AI agent?" Literally, eight out of one hundred had done it.
- FCFlo Crivello
Yeah.
- AGAakash Gupta
So-
- FCFlo Crivello
And that's great
- AGAakash Gupta
... the penetration is very high on Twitter. It's very high on YouTube. But in the-
- FCFlo Crivello
Yeah
- AGAakash Gupta
... real world of people working, the penetration is not there, and I think seeing is believing.
- FCFlo Crivello
Totally.
- AGAakash Gupta
So would you be up for kinda demoing these for us?
- FCFlo Crivello
Yeah, absolutely.
- AGAakash Gupta
And while you're doing that, a lot of people, they're not even sure what an agent should be used for. What are, in your mind, the top five AI agents that every entrepreneur should be using?
- 3:16 – 6:57
Agent #1: Meeting recording as a ‘second brain’ with automated follow-ups
- FCFlo Crivello
Yeah. I think the, the ones we see most commonly used, there's the ones that people use for their companies, and then there's the ones that people use for their lives. For your life, you should use an agent for meeting recording. It, it blows my mind that people... Not everyone has one yet. It's like you need to have an agent sitting with you in every meeting, and it's, it's incredible because it becomes like a second brain. Like, you can ask it questions about any past meeting. You can, like, have it, like, send action items. You can have it, like, you can configure it so that, you know, we, we, we have, like, Slack channels for every project in the company. And so if we have a meeting about a project, my agent, and I'll show you what it looks like. This is my meeting recording agent. This is actually a, a recording that we had for... It's a chat with Anthropic. And y- you can see right here, this is actually a question I asked this morning. So we, we are organizing a hackathon, uh, on, uh, August ninth. Uh, if you are in San Francisco and if you wanna join, please join. Um, and I just asked this morning, it's like, "Hey, like, how much did Anthropic say they would contribute to the hackathon again?" And here they were like, "Oh, you know, they're contributing, like, ten thousand dollars worth of credits." And that's just me, like, opening a new task, but I can just also scroll down and see all of my meeting history here. We're gonna have to, like, hide the list here. But I can see here this task, which shows, like, Lindy recording the meeting and then, like, summarizing the entire meeting here.
- AGAakash Gupta
And what's the differentiator people might be saying between a Lindy meeting AI agent and just using Granola or any of the other sort of famous note-taking apps out there?
- FCFlo Crivello
Yeah. It's a good question. First of all, I think all of those AI, uh, all, all of those, like, note-taking apps, they're all including AI in it. So fundamentally, I think, like, whether you know it or not, you are actually using an AI agent. It's, it's perhaps like a more minimal AI agent in the case of other apps, but it's still an AI agent. So the advantage of using a platform like Lindy, which by the way, like, there's a lot of cases where you should use other solutions. The advantage of using a solution like Lindy is that it's going to be way more powerful and way more customizable. So I'll, I'll show you, and I'm almost afraid to show it 'cause it's, it's going to seem intimidating. This is, this is my, uh, meeting recording agent. Like, I've, I've added to it over the years at this point, and so it's very rich. Like, there's a lot going on here. So I'll show you an example of a... I just mentioned we have project channels on Slack. And so the way it works is, like, I have this agent here. Like, I have this trigger that's like, when a calendar event begins, you record the meeting, you summarize the meeting, and then at the end of the meeting-I have this condition that's like, was this a project meeting? And if so, okay, well, then, like, look at my list of channels on Slack, and if you found a matching channel for this project meeting, then you send a message to that channel with the, the summary of the meeting. And so it's super magical. Like, we just talk about project X, and then automatically I hang up the call and automatically there's a summary of like, "Hey," like, "these three members of the team had a conversation about project X. This is what they discussed." This may seem intimidating, but, and this is part of what we just announced that we're so excited about, is the agent builder. So it used to be that in order to get this kind of behavior, you had to, like, use this, we call it the flow detail yourself. It, it's not too hard. Like, you can see it's no code. It's just, like, drag and dropping stuff. But still, it's, it's a little bit of work. And so what we are just announcing now is this agent builder. So now you can just click this button here and type in natural language. It's just a prompt. You say what you want your agent to do. So now I could just say in English like, "Hey, after all my meetings, I want you to check if it was a project meeting, and if it was, I want you to look at the list of channels on Slack and send the summary to the right channel on Slack."
- 6:57 – 8:54
Reliability & brand safety: hallucinations, human-in-the-loop approvals, and learning from edits
- AGAakash Gupta
And I think that the first hesitation that everybody has is LLMs, they tend to be unreliable. There tends to be some lossiness. How often is it that your meeting recorder is putting it in the wrong Slack channel, or how do you engineer this agent so that you can minimize that lossiness?
- FCFlo Crivello
That's a good question. First of all, I think LLMs have a worse reputation than deserved because they have improved so rapidly, and I think OpenAI got the crown today of number one in AI, in large part through their aggressivity in releasing things very, very, very early, which I, I admire that just for the record. Like, I think it's awesome. Like, the, it's an organization that's got, like, a tremendous bias for shipping, which I think is awesome. But as a result of that, look, they shipped ChatGPT when it was GPT 3.5, which was, like, an embarrassingly, I'm sorry, but, like, by today's standards, like, an insanely bad model, right? Like, the very term hallucination I feel would not really exist or, like, would not be as prominent in people's minds if they had waited just six months to release ChatGPT. So look, this is not to say that LLMs are perfect. They're obviously not perfect. They still hallucinate sometimes. They're, like, not deterministic very, very, very obviously. But, but they're quite good. Uh, I think they're, they're about as good as a human insofar as they are not deterministic for these ty- for these tasks. They don't screw up much more often than a human would.
- AGAakash Gupta
Right.
- FCFlo Crivello
And number two, like, w- we offer, like, a human-in-the-loop primitive, so that means that you can select any step that your agent performs, so, like, here in this example, sending the message on Slack. And so here you could be like, "You know, I'm not sure you're gonna pick the right channel. I'm not sure what you're gonna do." And in literally one click here, like, boom, you can toggle, ask for confirmation, and this means that before Lindy performs the action, she's gonna show you what she's about to do, and you're gonna, you're gonna get a chance to review it and to change it. And we're also working on the ability for Lindy to learn from the changes that you make, so, like, little by little, you need to make fewer and fewer changes.
- 8:54 – 10:32
Choosing the right model: speed vs intelligence vs cost (plus token ‘burn’ for high-stakes tasks)
- AGAakash Gupta
And I'm noticing here that you're on Claude 3.5 Sonnet. How do you make the decision of which models to use for different tasks?
- FCFlo Crivello
Yeah, it's kind of weird. I know I'm on, I don't know why I'm on Claude 3.5 Sonnet. I should be on Claude 4 Sonnet. Frankly, I, I think, like, 95% of the time you shouldn't have to worry about it. Like, just select whatever is the default model that we offer, which today is Claude 4 Sonnet. Sometimes if you need a lot of speed and, like, low latency, you should use the fastest model, which today is Gemini Flash 2.0 or 2.5. That's very useful for, like, phone calls. We do support phone agents. That's, like, one of our main use cases. Sometimes you need an agent that's particularly smart, in which case w- you should use either o3 or Gemini 4 Opus. We just released as part of this announcement, we have an, an agent that makes bets on Calci, like the, the betting marketplace. So every day it wakes up, it looks at the bets available on Calci, it decides which ones to bet on, and then for each bet it's about to place, it thinks very hard. It does a lot of research online, and it thinks very hard, and it burns, like, a couple dollars' worth of tokens to think about the bets that it's about to make. And for that, I, I just need, like, the absolute smartest model and, and for that we just use either o3 or Gemini, uh, Claude or Gemini.
- AGAakash Gupta
Okay. So model choice is really dependent, it seems, on how smart you need the model to be and how fast you need it to operate. If those are the two important dimensions, can we go ahead and see, like, what it's like perhaps maybe even to use the agent builder or the flow editor for our second AI agent from scratch, how it looks like to build this all together?
- 10:32 – 13:26
Agent builder walkthrough: creating an email triage agent in plain English (and ‘agents building agents’)
- FCFlo Crivello
Yeah, for sure. Let's, let's do it. I'm gonna create a new agent here. So you literally just click start from scratch, and you have the agent builder that appears on the right, on the left. And I'm just gonna y- just talk to it in English, and so I'm gonna say, we're gonna start to, like, a very simple use case that's like the email triage use case. That's one of our main use cases. So when you receive new emails, you want an agent to triage them for you. So when I receive new emails in Gmail, I want you to apply labels, labels either urgent, FYI, archive, or investor. I also want you to... Nah, I mean, here you can just, like, add more and more prompting. You can be like, "When do you want me to archive? When do you want me to urgent?" But, like, here, I'm just going to keep it simple with, like, this one simple prompt.
- AGAakash Gupta
Yep. That was some of the flow that we were seeing earlier, and I think this is really the power of the Lindy platform. Actually, why I'm using Lindy so much is when you edit the flow, right, you're really gonna be able to personalize it to how you want it to behave. We talk about context engineering so much. That kind of is context engineering in action.
- FCFlo Crivello
That's right. That's exactly right. And here again, the sky is the limit. You could be like, "Hey, by the way, I'm not gonna keep giving you instructions here in Lindy for-"How to triage my emails. What I'm gonna do is I'm gonna add the instructions to a Google Doc, and I want you to consult the Google Doc every time you receive an email to know how to triage emails. And so I'm just gonna go and, and keep editing the Google Doc.
- AGAakash Gupta
Oh, that's genius. I'm not doing that right now. So you can just have it reference one Google Doc, and you could even have maybe members of your team or whoever update that at any time.
- FCFlo Crivello
That's right. Boom, I've got my agent. And it's like, all right, when I receive an email, there's an email analyzer. It's really cool because you can see it's basically an agent prompting an agent. So people always ask me for, like, "How should I prompt my agent? What's the best?" And I'm like, "Don't even worry about it." Again, I think, like, prompt engineering is one of these terms that would not have existed if, if, if OpenAI had waited six more months, which I don't think they should have. And so here it's just like, don't worry about it. Just speak in natural language, and you're gonna have an agent building your agent. By the way, an interesting cool detail is that this agent builder here is a Lindy. It's literally just like we've built it into the platform, like we have... It's not even an admin tool. Like, we have, like, a team account, and we use the same interface as every other user uses, and that's the interface that we used to build the agent builder. So it's, it's just agents all the way down. So you can see here, like, the trigger is, like, when they're receiving an email in Gmail, you know, I go into this agent step and analyze the incoming emails and define how to label them. And here-
- AGAakash Gupta
Yeah
- FCFlo Crivello
... there's, like, a skill to label the email. That's it. Super easy.
- AGAakash Gupta
And so what would be like, if somebody wanted to add a more complex flow, what would be an example that they could add in here?
- 13:26 – 17:25
Agent #2–#3: CRM manager agent + 6,000 integrations (and using docs as living instructions)
- FCFlo Crivello
Yeah. Just yesterday. So one of my personal favorite agents that I have is my CRM manager agent, and I, I created it before we had the agent builder. But I often have friends who ask me, they're like, "Flo, I love this agent." Like, I show it all the time 'cause I love it so much. It basically manages my personal contacts. So every so often I ping it, I'm like, "Hey, I just met this guy. He's really cool. He's, like, a sales guy. You know, remind... Remember him as, like, a sales guy." And then every so often I go to, to, to the CRM agent and I'm like, "Hey, who are salespeople I know?" Like, we are hiring salespeople again. What, what, what are some good ones I've met, you know? So I- it's just become irreplaceable for me. And, like, when I'm about to fly somewhere, like if I book a flight to Paris or whatever, it automatically detects that I'm flying to Paris through my inbox, and it sends me an email proactively. It's like, "Flo, I see you can- you're going to Paris. Here are some people that you know in Paris that you should, you should hit up." So it's, it's such, such an awesome agent. And so I have friends all the time ask me, "Ah, like, can I, can you share this template with me?" And so just yesterday I wanted to see if the agent builder could create this agent basically from scratch. And so you can see the prompt here that I sent it. It's, it's... Again, it's not rocket science. Th- there is no prompt engineering here. It's literally just talking. And I'm like, "All right. I want you to create an agent to manage my CRM. Every so often I talk to it, and either I want to add a contact to my CRM or I want to ask for a list of contacts, like which investors do I know? You add contacts, you retrieve contacts. And by the way, also, once a week you wake up, you look at the people that I met on this week. So you look at my Google Calendar, you look at my CRM, and then you tell me who are some people that I have met this, this week that I don't have in my CRM, and you ask me which ones to add." And it just worked. Like, it's like, it, it asked a couple of clarifying questions like, "Hey, like, what do you use for your CRM? Is there, like, a Google Sheets? Like, what do you use for your calendar?" And I just answered the questions and it, it created the agent for me. Took, like, two minutes.
- AGAakash Gupta
Before we dive deeper, let's talk about something every PM faces: getting alignment on product decisions. You know that feeling when you're trying to explain a user flow to engineering or justify a design choice to leadership and you're just describing it with your hands? That's where Mobbin comes in. Mobbin is the world's largest library of real-world mobile and web app designs from industry-leading apps like Airbnb, Uber, and Pinterest. Instead of spending hours taking screenshots or hunting for inspiration, you can instantly find exactly how successful products handle onboarding, paywalls, checkout flows, whatever you're facing. Over 1.7 million product builders use Mobbin to benchmark against best-in-class products and show their teams proven solutions. Whether you need to convince stakeholders there's a better way to handle user activation or research how top apps approach feature discovery, Mobbin gives you the visual proof to back up your product decisions. Check out mobbin.com/aakash. That's M-O-B-B-I-N.com/A-A-K-A-S-H and get 20% off your first year. Today's episode is brought to you by Jira Product Discovery. If you're like most product managers, you're probably in Jira tracking tickets and managing the backlog. But what about everything that happens before delivery? Jira Product Discovery helps you move your discovery, prioritization, and even roadmapping work out of spreadsheets and into a purpose-built tool designed for product teams. Capture insights, prioritize what matters, and create roadmaps you can easily tailor for any audience. And because it's built to work with Jira, everything stays connected from idea to delivery. Used by product teams at Canva, Deliveroo, and even The Economist, check out why and try it for free today at atlassian.com/product-discovery. That's A-T-L-A-S-S-I-A-N.com/product-discovery. Jira Product Discovery: Build the right thing. And people might be asking, "Well, what types of integrations do I need to do? How do I integrate my CRM? What does that look like?"
- 17:25 – 20:42
Computer Use: bypassing missing APIs (Twitter spam blocking) and surprising cost dynamics
- FCFlo Crivello
Yeah. I mean, that's one of our strengths. Like, we have 6,000 integrations, so pretty much any tool that you want to use, you can, you can use it in Lindy. And with Agent Builder, you don't even have to worry about it. Like, Lindy's just going to pull up, pull the integrations into the agents, and then she's going to ask you. So like, "Okay, I'm gonna need you to authenticate to Salesforce or HubSpot or whatever." And then, and this is also, like, the very big announcement that we made this week, for the rare cases where we don't have an integration, we just announced computer use. So basically it's the ability for your Lindy agents toUse their own computers in the cloud. So anything that you can do on a computer, it can do as well. I'll show you this, this example. Twitter pretty famously doesn't really offer an API. Like, it always sucked and, and, and then they made it suck even more. Like when Elon took over, he basically, he basically completely gut the API. And I have this problem on Twitter, which is I feel like the, the, the spammers are back, man. It's crazy. Like, I'm getting so many spammers in my mentions, and it's always the same stuff. It's like, "Look, like this gentleman is awesome. His analysis is so good. Like, what this guy said makes sense. Fire emoji. Fire emoji." And so what I did is I created an agent using Agent Builder, and you can see here I have like a video of when I created it for the first time. The prompt is, is, is literally just, again, like not rocket science. I have a bunch of spammers in my mentions on Twitter. They say like, "This gentleman's analysis is great. What are you waiting for? Follow him." I want you to create an agent that wakes up every three hours, looks at, look at, looks at my mentions, and blocks the spammers, basically. And you can see here, um, this is a task where it does it, and I, I no longer have spammers. I, I simply don't see them anymore in my mentions. I can see here the, the history of the task, and like it just like keeps scrolling and finds spammers and like, boom. "I met a blogger who releases top analysis. Like, you should totally follow him." It's like, boom, block. Never wanna hear about you again. So there's, there's basically no real limit anymore. Like, since we have started beta testing computer use internally, like I have never been bottlenecked by the lack of an integration.
- AGAakash Gupta
Wow. So you actually don't even need an API with computer use. Does that add to cost or anything like that?
- FCFlo Crivello
So that's been the huge surprise for us and for me. Initially, we thought that computer use would be a lot more expensive than API integrations, and in fact, it's often cheaper. And the reason for that is because, um, it's, it's, it's a little bit of a, of a technical detail, but like we, we basically, with API integrations, when Lindy talks to integrations, we keep everything in the context window at all times. With computer use, we actually remove everything but the last three screenshots. So you can see this one task here cost me $0.30, which is, is like pretty low. Like, most other computer use agents are like a dollar fifty or two dollars per task.
- AGAakash Gupta
Yep. Very interesting. So we covered, I guess, in that personal use case bucket so far, right? We covered meeting tria- or email triage, meeting note-taker, blocking people [chuckles] who are spamming you on social media, and then in the business bucket, we covered your CRM manager a little bit, although that could also be a personal use case. What are the other really interesting business use cases entrepreneurs should know about?
- 20:42 – 24:46
Business agents that replace real headcount: support automation, elastic scaling, and recruiting at scale
- FCFlo Crivello
Yeah. Support is a really big one. I think like that's the most obvious one that people are, are automating 'cause it's, it's costly. It's a pain in the butt to scale. Like I think one huge advantage that people don't discuss often enough with agents is it's just like the cloud. It's like AWS for labor, right? It's like you never have to worry about scaling again. It's just, it just like scales elastically up and down with, with your, uh, level of demand. Like we, we have some customers who are like e-commerce shops, and Black Friday, for example, is a very big day for them. And they love Lindy because they're like, "Black Friday used to be such a headache. It was like such a mad scramble of like I needed to find like more support people and like the temp workers that just work for like a week, and I have to train them just for this week. It's just, it's just a nightmare." Whereas with Lindy, it's like, hey, 24/7, 365, I reply to support tickets in like literally 30 seconds. Like this is what I mean when I say it's like 100X better than the human. Like at best, your humans, like a good support team is going to reply to tickets in like an hour. Like this replies to every ticket still in 30 seconds, 24/7, 365. And if you get a lot of tickets, if all of a sudden you have a spike in traffic, and you get a million tickets, then it's just going to reply to the million tickets. And if you have zero tickets for like a week because your, I don't know, maybe your business is doing very poorly, maybe you're going through like a down season or whatever, you're gonna pay nothing, right? So support is, is a huge one. Recruiting, I, I love like my recruiting agent. I couldn't live without it. I can show you my, my bulk recruiter here I call it. You're gonna see basically I can literally just go to it, and I can ask it, "Hey, who else..." Um, it's like here. How can I help you? Who can I help you recruit today? And I'm like, "I need you to find 10 software engineers from Zapier." And it's like, "Okay, I'm gonna search for them," and so it looks for them. And then it's like, "Okay, I found these people." And it's good. Like I can, I can open these LinkedIns here. It's like boom, boom, boom. Like those are like actual software engineers at Zapier.
- AGAakash Gupta
Those are real links.
- FCFlo Crivello
That's right.
- AGAakash Gupta
Sometimes LLMs hallucinate, but I think that's again people kind of being six months in the past. We're getting real links.
- FCFlo Crivello
That's correct. You actually get like actual software engineers at Zapier, and then it's like which of those... Oh, I'll, I'll show you this other one. Um, it's like, okay, which of those should I actually reach out to? And I'm like, "Yeah, it's all good," like reach out to all of them. And then it kicks off this other thing that we reach- that we released about six months ago or so, like three or four months ago, which is Agent Swarm. 'Cause agents suck at performing actions in bulk over long time horizons. So a human, if I'm like reach out to these 20 or 40 or 50 people, just gonna sit down and do it and like just do it one by one, and it's gonna work. An agent after the like 10th or 20th or so, it's going to start screwing up. So with agents, and also it's gonna be very slow. With Agent Swarms, like it can perform a bulk of tasks all at once in parallel and doesn't screw up. And so here it's like, okay, I'm gonna enter this loop once for each of these people that I found. And here if I expand the task in the sidebar, I can see all of the subtasks for each of the agents in the swarm. And so basically, the agents in the swarm, what they do is first they search for emails, for previous emails that I have sent to this candidate before to make sure that they've not reached out. If I have not reached out, then they draft an email to the candidate.
- AGAakash Gupta
Smart, because the last thing you wanna do [chuckles] is just send them an email, and they say, "You messaged me three months ago."
- FCFlo Crivello
That's right. And then that's the beauty of Swarms is like even though each of these agents executes in parallel-They then all merge back into a single agent when they're done. And so you can see the single agent has this awareness of everything that the swarm did. And so it's like, "I did it." Right? "I completed a comprehensive outreach to 20 software engineers. Here's what I did." And it did know, it's like, "Hey, I skipped two whom we already reached out to." It's like, boom.
- AGAakash Gupta
There we go.
- FCFlo Crivello
These people you, you'd reached out. It's like it even knows these, like, ah, I reached out previously three times in May, right? So, yeah.
- AGAakash Gupta
So you're working heavily on recruiting right now. [laughs]
- FCFlo Crivello
That is... Yes, absolutely. We're growing quite fast right now. Hey, we are recruiting, by the way, if you... Every function, just hit me up at, uh, flo@lindy.ai.
- 24:46 – 26:54
Operational guardrails & evaluation: avoiding Gmail bans, activity logging, and weekly digests
- AGAakash Gupta
So I think people would want to know, like, what are the, like, areas where it can mess up that we need to, in the flow editor, kind of account for?
- FCFlo Crivello
Yeah. Well, I think that's really like the... Oh, no, there is another thing that I added here, which is I use Gmail to reach out to people. Gmail is not really supposed to be used as, used as like a mass outreach thing. And so what I did is, basically, you can get blocked on Gmail. If you start to email thousands of people, like Gmail is gonna block you, and you're gonna go in spam as well.
- AGAakash Gupta
Yeah.
- FCFlo Crivello
And literally, like your Gmail account can get deactivated, which is catastrophic. So what I did is I, I made it keep track of the campaigns that it has running in a spreadsheet. So basically, what it does, it's like, okay, I tell you what I'm looking for, you search what I'm looking for, and once I've told you explicitly who to reach out to, you first look at your running campaigns. And then you tell me if you have too many running campaigns at a time right now, you tell me like, "Hey, do you want me to do this, or do you want me to wait until the campaigns have run and basically queue this one up?" And then once you proceed, either I've told you to wait and the wait is over, or like I've told you to continue, then you, you do the whole outreach.
- AGAakash Gupta
Nice. Now, how do you measure the system and evaluate the system so that you can improve it over time?
- FCFlo Crivello
Yeah, it's a good question. Today, you sort of have to go to the tasks and, like, look at it. If you want, you can update your agent to send you updates, right? So you could have it log all of its activity to a spreadsheet, and then every week or whatever, it'll send you an update of like, "Hey, this is what I did this week." I actually did that with my meeting recorder. So just to go back to it really quickly, when it records meetings, it also logs the meeting in a spreadsheet. And then every week, you can see I have this trigger here that's like, okay, every Friday, you look at all the notes of the meetings for this week, and you send me an email with like... It's like a weekly digest of everything that happened this week for me. And I can't tell you, I look forward to this email. It's like at the end of my week on Friday at like 6:00 or 7:00, I receive this email that's like, "Flo, this is, this is the week. This is what happened. This is where you are in the company. This is the decisions you're making," and so forth. And it really helps me take a couple of steps back and think about what's going on.
- 26:54 – 31:06
More ‘replacement-grade’ agents: sales prospecting, LinkedIn outreach, and cross-post marketing automation
- AGAakash Gupta
Amazing. So that is five agents that basically any entrepreneur can benefit from. Are there any other agents people should be aware of that could replace members of their team and help them save money?
- FCFlo Crivello
Yeah. I think, like, sales is a really big use case. So sales prospecting, sales outreach, especially with computer use now, you can pretty much reach out through any mean. Like, it used to be like, oh, you can only send SMS or, like, email. Now, it's like, hey, even if the company is behind, like, a contact us form online, you can do it.
- AGAakash Gupta
Oh, wow.
- FCFlo Crivello
Or even if, if, if it's only like a LinkedIn, like we can send LinkedIn DMs now. Like LinkedIn, also famous for, like, not really offering an API. Well, you can just offer a LinkedIn DM now. If the person has a Calendly open on the internet, you can have the agent find the Calendly and book some time on the Calendly for you.
- AGAakash Gupta
Wow.
- FCFlo Crivello
So I think, like, sales is, is just a, a massive use case.
- AGAakash Gupta
Think people are gonna really want to see that LinkedIn one. Can you just walk us through that? I actually haven't used that one yet either.
- FCFlo Crivello
I don't have it handy here. It's on an- on another account. Yeah.
- AGAakash Gupta
Okay. No problem.
- FCFlo Crivello
Yeah. I think marketing and, like, content generation is a really big one. So I, I feel like it's a little bit of like my, my dirty secrets here. But I'm very active on, on, on X/Twitter. I'm not very active on LinkedIn, and I, I don't really want to be active on LinkedIn. And I, you know, part of it is, is really just like brain cycles. It's like I, I, I feel guilty enough being active on Twitter. Like, I spend like 20 minutes a day on it or something. And so I created this agent that wakes up every day at 9:00, uh, starts a computer, looks at my tweets that I posted over the last 24 hours, and then analyzes tweets to know which ones to post on, on LinkedIn. Because I also feel like LinkedIn is a different vibe from Twitter. Like, everything I tweet on, on Twitter, I can't post on LinkedIn. So like I, I shitpost, I post a bunch of like shitty jokes, like I, I post about politics and all of that stuff. Like, I don't really want that on my LinkedIn profile.
- AGAakash Gupta
Yeah.
- FCFlo Crivello
And so I'm like, "Okay, you analyze my Twitter, and you propose the tweets to cross-post to LinkedIn." And so once you have collected these tweets that you think should be cross-posted, you send me a message, you send me this list, and once I have told you which ones to post, then you go to LinkedIn, and you paste, you, you, you, you, you, you, you, you cross-post from Twitter to LinkedIn. So I have nothing to do now, just like post on Twitter, and automatically stuff finds its way to LinkedIn.
- AGAakash Gupta
Wow. That's a huge agent for co- anyone creating content. We talked about entrepreneurs. We talked about, a little bit about content. A lot of people watching or listening are product managers, and you were a product manager. If you were a product manager again, like, what would be the top agents you would build? Today's episode is brought to you by Maven. The problem with most courses online, like Udemy, is there's no live component, and the instructors aren't experts in their fields, they're professors. At Maven, you get direct live access to experts and operators from the world's best tech companies. You can't get that access anywhere else, in any university, and you usually can't find them on YouTube either. I've featured so many of Maven's experts in the newsletter and podcast for that reason. To help you out, I've put together a collection of courses I recommend at maven.com/x/aakash. This includes courses like AI Prototyping for PMs, Product Sense for PMs, and getting an AI PM Certification.Visit it now at maven.com/x/aakash. Today's episode is brought to you by the AI PM Certification on Maven. Run by Miqdad Jaffer, who is a product leader at OpenAI, this is not your typical course. It's eight weeks of live cohort-based learning with a leader at one of the top companies in tech. OpenAI just doesn't stop shipping, and this is your chance to learn how. Run along with product faculty and Mo Ali, the course has a 4.9 rating with 133 reviews. Former students come from companies like OpenAI, Shopify, Stripe, Google, and Meta. The best part? Your company can probably cover the cost. So if you want to get $500 off, use my code AAKASH25 and head to maven.com/product-faculty. That's M-A-V-E-N.com/P-R-O-D-U-C-T-F-A-C-U-L-T-Y.
- 31:06 – 35:52
PM-specific agent stack: voice-of-customer digests, virtual users, and a ‘mini-PM’ decision agent
- FCFlo Crivello
Yeah. I think that the meeting recording is a really big one because as a PM, your role is so cross-functional, it is so collaborative, and you basically spend your life in meetings. And I think a lot of your job comes down to making sure people are aligned, people are in the loops, like things don't fall through the cracks. And so this agent that I just showed that like disseminates meeting notes through the right Slack channels is-- would've been a godsend when I was a PM. So that's one. The other one is like the voice of the customer. So we have this agent that I, I love that like every evening it wakes up and it looks at all the support tickets that were received over the last 24 hours, and it sends a, a digest in like the general channel on Slack of like the main things people are saying about, about the product, the main issues they're encountering, the main things that are confusing them, and so forth. And so it's just, you know, very often in big companies in particular, product teams end up weirdly disconnected from the customer, which is like the only thing that matters. Like you should be in very, very close touch with your customer and your user. This just-- You have no excuse left. You just receive this like couple of bullet points every day. It's like, "Hey, this is what's going on with, with our users." Like, and then it, it's very obvious all of a sudden if there is a big divergence between what your team is working on and, and what our customers are complaining about. Another one that I really like is like a virtual user. So this one basically, especially if you have a, an agent to record your meetings, you can then every time you have a chat with a meeting, you can add it to a knowledge base, and then you can plug that knowledge base to another agent that becomes a sort of virtual user. So it's like your internal expert about customers and users in your company. And you can then share that agent with the rest of your team, and you can be like, "Hey, anytime you have a question about a user or like a feature or like how people use the feature or like what they like, what they don't like, how they think of competitors and whatnot, uh, you can just talk to this agent." And, and finally, one last one that I like, and this is part of what we just announced this week, is like this ability for people to create agents and share them with the rest of their teams. And so one agent we have internally is, is like our PM has created it. It's like, you know, engineers ping you all day, like asking you for like, "Hey, can I have help for like the copywriting? Like PMs are doing the copywriting right now at Lindy. Like can I, can I know what copy to put in this button?" Right? Or like, "I'm facing this like micro product decision that we didn't define in the designs. Like what should we do here?" And so this PM created an agent that's basically like a mini, a mini her. It's like, "Hey, this is how I make product decisions. This is the principles I use. This is a bunch of examples of, of product decisions I've made before." And so anyone can talk to this mini PM every day to ask it. And then if the mini PM is not sure, it's like, like, "I'm sort of on the fence on this one. Like my instructions don't really tell me what to do here," it actually loops the PM in. It's like, "Hey, Michelle..." Her name is Michelle. It's like, "Michelle, like, like someone's asking me this question. I'm not quite sure what to say. Can you tell me?" And then I'm gonna add that to my own knowledge base. And so like the, the agent continuously self-improves. So I would say, I would say those are like the big ones I would use as a PM.
- AGAakash Gupta
A couple other ones that, uh, come to my mind for a product manager, like potentially like a competitor updates agent, because there's way too many competitors to keep track of. So you could have an agent that is keeping track of these are the relevant competitors. Message me or Slack me when I need to know about those launches.
- FCFlo Crivello
Yeah.
- AGAakash Gupta
I think that another agent that I often recommend that PMs build is an agent that is taking their millions of bug requests, Slack requests, email requests for features-
- FCFlo Crivello
Right
- AGAakash Gupta
... and then putting those directly into a Jira backlog for them. [chuckles]
- FCFlo Crivello
Right. Yes.
- AGAakash Gupta
And even coming up with like the orig- the initial version of, "Hey, this is the user story. This is what I think the potential impact is. This is what I think the potential severity is." Because you can actually train that agent on prior examples, and it can just do the first version for you.
- FCFlo Crivello
100%. I'll show you, I'll show you another example of this is sort of like the adjacent to like the, the, the virtual user or like the voice of the customer one that I just mentioned. Um, we released-- This one was a while ago. Like we released a, a feature for Lindy to support email attachments. And so I was able to talk to that, like we called it like the Lindy brain, right? And so it was like, "Hey, we just released email attachments. What are some customers that have requested it? Give me their name, give me when we met them, and give me a couple of sentences on what they were building." And boom, here it sends me a list like, okay, these people, this is what they've requested email attachments for. And so we got like this entire list. We're gonna have to like hide like the, the names, but like-
- AGAakash Gupta
Okay
- FCFlo Crivello
... got this entire list of people and, and why they wanted email attachments.
- 35:52 – 40:14
Frameworks for when to build agents: repetitive work, dislike-to-do tasks, and safety via permissions
- AGAakash Gupta
So many use cases. And so if there was like a general framework that we had to put together right, I think that you wanna use an agent for anything that is like manual or repetitive work, anything that you can potentiallyHave instruct it and give it instructions. Anything that you could potentially train up like a VA or somebody on, you should think about, can I build an agent? Is there anything else people should think about, okay, this might be an agent use case?
- FCFlo Crivello
Yeah. I think, I think you, you nailed it, you nailed it on the head. Like, anything, uh, we always say, like, you never do the same thing twice. Like, if you, if you find yourself doing the same thing multiple times, like, you should just create an agent for it. Anything you don't enjoy doing, I think, like, that, that's part of why we're automating stuff. That's why we have machines, right? It's like, like, we-- I wanna be a human. I don't wanna be a machine. I, I don't want... I wanna have fun at work. So anything you don't enjoy doing, you should, you should build an agent for. Anything that's done at scale in your company, you should definitely build an agent for. Yeah.
- AGAakash Gupta
And we have to include the obligatory, how do you make sure that your agent is kind of safe in your work environment? What are the things you should be thinking about when you create an AI agent so that it isn't going out there and potentially ruining your brand name within your company?
- FCFlo Crivello
Right. Yeah. Well, that's where human in the loop is important. Again, one click, you can toggle it down and you can be like, "Before you do this specific thing, you ask me for confirmation." Agents can also only do the things you gave them access to do. So if it's integrations, like, u-unless you gave the agent the ability to issue a refund of, on Stripe, for example, it's not gonna be able to do that, right? But by and large, I think, and in particular with this, like, human in the loop primitive we have, I think people should increasingly with, with AI agents and with the rise of AI in general, they should increasingly think themselves as orchestrators and not as doers. Like, you should really basically more and more and more and more as, as time goes and as these AI systems develop, you should basically spend more and more of your time monitoring and orchestrating your basically empire of AI agents, uh, instead of doing the stuff yourself and, and clicking the buttons yourself. And, and that's just going to be, like, a much higher leverage activity for you.
- AGAakash Gupta
I think everybody's been trying to figure out, they're trying to been getting their hands on, like, how is AI gonna change the nature of work? I think in the middle of 2025, this is the most clear tactical example, right? You're switching from doer to orchestrator.
- FCFlo Crivello
100%. 100%. It's, it, it, you know, it's, it's the same thing that happens when people get promoted to managers. Like, the more, the more high up the chain, like, the less you do, the more you orchestrate. I just think with AI agents, it's like everybody just got a promotion to manager.
- AGAakash Gupta
Everybody just became a manager, which is a totally new skill set. And so you kind of need to think about developing these skills first by doing yourself, but then learning from others, asking for feedback from others, getting inspired by the flows that you just saw from Flow, asking other people. Like, one of my favorite things to do these days, I'm in a couple content groups, is, like, share the screenshot of my Lindy with them. So, like, you had your X to L- LinkedIn cross-post Lindy, and it's like, "Oh, look at this cool thing I did where, like, I created a condition, if I'm talking about politics, do not post on LinkedIn." And that's how you really start to improve this skill set.
- FCFlo Crivello
Yeah. The interesting thing though is, like, when you become a manager, I think there's, like, two big categories of skills you must acquire. And in general, when you read management books, you sort of see these, like, two schools of thought. There's like, I call them, like, the hard school and the soft school, right? Like, the hard school is like, okay, this is how you design your, your, your, your, your company as a system, as a machine, right? Like Ray Dalio, like, principles, like, talks a lot about this. Like, pay attention to your system, pay attention to your machine, right? Like, how do you make it repeatable and so forth? Like, how do you set up the incentives and the reporting and all of that stuff, right? So, like, that's the hard school. And then there's the soft school, which I actually think, in the case of humans, is more important, which is, okay, how do you inspire people? How do you create the culture that you wanna see in your company? How do you communicate with them? How do you handle conflict, right? And I, I think with AI agents, obviously, you, you basically remove the entire soft component of it, and it's just hard stuff, right? So I actually think that the transition is going to be a little bit easier for ICs to become these, like, agent managers than it would be for them to become human managers. Because, like, as an IC, you are used to the hard stuff, especially if you're, like, technical. You know how to think about systems and how to design these things. I think it's the soft thing that very often people struggle with when they make the leap to management.
- 40:14 – 44:01
Agent platforms vs chatbots: orchestration, permissions, performance, and cost management
- AGAakash Gupta
So the agent mode, going back to that Slack thread, when I followed up with people, I said, "Okay, what agents have you been exposed to?" Almost all of them said ChatGPT and Claude agents. So of those a- even fewer are actually using these true AI automation platforms. How should they think about what they're seeing in ChatGPT and Claude versus an AI automation in Lindy?
- FCFlo Crivello
Yeah. I almost think of it as the difference between Slack and iMessage, right? Like, one is really a, a tool that you sort of use personally and, like, you know, and the other one's like a work tool. And, and it's a lot more complex as a result of being a work tool. But basically, the fact that it was designed in this work context has a million downstream ramifications. It's just, it's just a much more complex application. It will give you a lot more control over exactly what your agents see, what they do, when they ask for permission. Um, and, and again, I think this, this world orchestrating is really important. You can basically think of, like, Claude and ChatGPT as, like, single agents. Like, what we let you do is we let you create and orchestrate a whole lot of agents, and these agents can also, by the way, work together. You can set them up so that they collaborate together. And so, and, and orchestrating basically means managing three things. It means, and ag- again, it's, it's sort of similar to being a human manager. You manage their performance. So you, you set up your agents, and then you, you check in on them. You're like, "How are you doing? Are you screwing up? I'm gonna, I'm gonna give you more instructions. I'm gonna tweak your logic and so forth." So there's the performance management. There's the permission management. So it's like, you just, you just talked about this. Like, how do I make sure that you don't, you know, like, access to systems you should not have access to? In- including your team permissions, like, you're gonna give your team access to your AI agents. Like, you're gonna have to define, like, what other teammates can ask from your AI agents and so forth. And then there's the cost management, and again, here it's similar to, to being a human manager. It's like you sort of manage payroll and, and raises. You don't have to give a, uh, you don't have to give a raise to your, to your agents, but you do have a lot of control over how much your agents cost you.Depending on how much context you feed them, depending on what model you use, like, you get to decide basically the, the IQ of your agents. It's like, okay, y-you know, a support agent, like, your, your work is not that complicated. I'm gonna power you by, like, a pretty cheap, simple, fast model. But you, like, you make million-dollar decisions. We don't have agents making million-dollar decisions yet, but we do have agents making, like, thousands of dollars decisions, like, pretty routinely. And so it's like you, I, I don't mind spending two bucks on tokens just to make sure you make the right $10,000 decision. So that, that's the difference. It's like ChatGPT, you have a single agent. It's, like, a very simple tool for, like, individual use. This is a work tool, almost a creative tool. It's, like, a very powerful tool that you use to orchestrate an entire army of AI agents.
- AGAakash Gupta
And I think on the cost component, people will intuitively understand, like, there are cheaper and more expensive models. I think maybe what they don't have an intuitive grasp on is how much context increases the cost. So how do you think about that?
- FCFlo Crivello
Yeah. Well, you, you do-- we basically charge you for the tokens that, that, that the models are costing us. The cost is linear. It's like the more tokens you, you add, like, the more, the more it costs you. Um, I wish I had a clearer answer to give on this one. Frankly, there is not much of a way to tell, to know how much an agent is going to, to cost apart from doing it, and we are in the process of building cost guardrails, so you are going to be able to configure your agents. And again, that goes back to the difference between, like, ChatGPT and Claude. Like, we're gonna give you, like, very, very granular knobs and, and buttons and switches that you can use to tweak your AI agent organization. And so you are going to be able to set up these guardrails for your AI agents and be like, "Hey, if you're finding yourself spending more than 10 cents on any given task, check in with me."
- AGAakash Gupta
Nice. So you can add in some guardrails to kinda limit your cost there.
- FCFlo Crivello
Yeah.
- 44:01 – 52:59
Competition & company building: Lindy vs Make/Zapier, lean teams with agents, and the Soham hiring story
- AGAakash Gupta
Within these tool platforms like Lindy, you guys now have some competitors. I think notably a lot of people are using, like, NAN or Agents.ai or Make.com. What are the differences between Lindy and those platforms?
- FCFlo Crivello
I think we're just easier to use. Like, a lot of those are, like, targeting developers. Like, from day one, our vision has been is the AI employee, and the AI employee has two big axes of performance. There's the capability of the AI employee, like, what can it do? And then there's the ease of use of the AI employee, which is how easy is it to get it to do this thing? And our vision is to just keep driving maximally on these two axes, um, with emphasis on the ease of use axis. Like, in the limit, we really do intend this to be a drop-in replacement for a human worker. Like, we want you to just be able to perhaps jump on, like, a literal call with your AI employee and be like, "Bob, welcome to the company. Share your screen with me." That's what we just released Computer Use for. "Share your screen with me. You are a support agent. Like, here are your intercom credentials. Go ahead and, and, and here is our knowledge base on Notion or whatever. Just answer support tickets and check in with me if you have any questions."
- AGAakash Gupta
Makes sense. And the other tool that I think people have been using a little bit for this and has really good penetration is Zapier. How should you think about Lindy versus Zapier?
- FCFlo Crivello
Yeah. We are more AI native. I think there's a lot of other tools out there that are more, like, about building workflows. We, we don't build workflows, we build AI agents. I think they just happen to sort of overlap a little bit, but we build agents, and we give you very advanced toolings and, tooling and controls to build very powerful AI agents.
- AGAakash Gupta
And with that very powerful control, how big has Lindy become today?
- FCFlo Crivello
We-- We're doing well. You know, we, we have more than 5x'd over the last five or six months. We're, we're growing fast. It's, it's going well.
- AGAakash Gupta
That's insane, 5x in six months. You wouldn't have heard that in any other era [chuckles] besides the AI era. How big is your team now?
- FCFlo Crivello
Well, just with over 40.
- AGAakash Gupta
Over 40 folks. And how have you structured that team?
- FCFlo Crivello
Uh, so it's a typical structure. It's like we've got engineering, product, and design on the one hand, and go-to-market on the other hand. Um, and the, the EPD teams are structured along, like, different product axes, and the, the go-to-market teams are, like, uh, marketing, sales, and customer success.
- AGAakash Gupta
Okay. Yeah, I always ask that because I think in the, these AI native companies like yourselves, they're running, like, particularly lean it feels like. Like, what is the typical scope for a EPD team right now?
- FCFlo Crivello
Yeah. We are running very lean. Like, our investors do tell us, like, "Wow," like, you know, there is this metric that's called, like, the burn multiple, which is, like, how much do you burn per dollar that you add? And we, our burn multiple is extremely low, like, very, very low. Um, we actually have a salesperson, uh, at Lindy who, she just posted the screenshots the other day. I forgot how many-- I think, like, it, it literally said she's got, like, a team of, like, her AI agents w-when you sign up to Lindy, you receive an email at the end of each week that's like, "This is what your AI agents did this week, and this is the equivalent of X full-time employees." And her email says 40 full-time employees, which actually checks out. Like, she is extraordinarily productive. It's insane. And by the way, she's making a lot of money. If you're a salesperson, hit us up. Like, salespeople make a lot of money. Yeah, and, and everyone has been wondering, like, how does she do it? Like, it's crazy. It's just she, she's got her fingers in every place. Like, well, she turns out, like, she's created a lot of AI agents. Um, um, the, the EPD team, we've got a lot of AI agents that in particular do code review. And so what we have, we, we have, like, one AI agent that's becoming irreplaceable that reads code review instructions from Google Doc. And we have another AI agent that joins our postmortem reviews every week. And so we have these postmortems where, like, like, there was an outage, there was a really big issue in the product. What, what happened in the code, right? And, like, does-- We, we try to debug. Again, that's, like, the hard side of management. We try to debug the system, try to understand how to do better next time. And, well, like, uh, one of the things that we do, we do multiple systemic updates when, when, when an issue happens. But, like, one of the things we do is that we have this agent updates that agent. We have the agent that sits with us in the postmortem review-
- AGAakash Gupta
Wow
- FCFlo Crivello
... updates the code review agent to be like, "Hey-This pattern has caused issues before. Like please watch out for this in future code reviews.
- AGAakash Gupta
So you can-- it's agents all the way down. [chuckles]
- FCFlo Crivello
Very much so. Absolutely. Yeah.
- AGAakash Gupta
Talking about the team, you hired perhaps the most famous engineer right now, Soham Parekh. This guy made waves, if you haven't heard of him, for working like five jobs. Tell us the story of hiring him.
- FCFlo Crivello
He was really good. He interviewed really well. When we hired him, he was supposed to move to San Francisco. Sure enough, within a week of joining, he started making excuses. He was like, "Like something happened with my flight. I couldn't come. I'll come later." And yeah, I mean, like I woke up one morning, and I saw that conversation on Slack, which was like a member of the team who had shared a tweet that was like, "Wait a minute, like is, is this our Soham?" So like there is this tweet that went viral that's like, "Hey, everyone, watch out. There is this guy called Soham Parekh that, uh, scams founders effectively." He just joins startups, and he joins a lot of them at the same time, and he just does the bare minimum to get fired, to, to, to get fired as late as possible. Uh, and he just like stacks paychecks like that. And, and so we were like, "Wait, is it like the Soham we hired like a week or two ago?" And, and it was him. So we just, we just fired him immediately, like within the half hour, he was, he was off-boarded. And then what ensued was like a, an amazing day for memes. [laughs] It was like the memes almost made it worth it.
- AGAakash Gupta
So I don't, I don't get how this guy is passing so many interviews and like getting so many jobs. How did he land the job?
- FCFlo Crivello
He's good. I think that's the sad part about it. Perhaps it's because he's got so much training that like he's done so much interviewing over the last few years. But like truth to be told, he, he crushed it. He, he crushed it. He crushed it at interviews. Like what-- I, I remember very distinctly one of the interview reports went, "This is the best I have seen any candidate." And we interview a lot. Like we've done literally thousands of interviews. This is the best candidate I have seen, like in terms of the performance at this particular exercise. So he's, he's good.
- AGAakash Gupta
So he has some skills, [chuckles] but maybe just isn't figuring out how to manage those skills. It feels like to me, with the huge paychecks going out in AI, if he had actually just focused on being good at his job, he could be making more.
- FCFlo Crivello
Yep.
- AGAakash Gupta
Crazy. What's your advice, you know, to other founders to avoid hiring a Soham?
- FCFlo Crivello
Trust your gut, and don't hire job hoppers. I blame myself because I remember very distinctly when I jumped on, like I'm still like the final step in every interview at, at Lindy, and I intend to be for a very long time. Literally, so I jump on these calls, and I know I'm the final interview, and the first thing I do is I open the person's LinkedIn, and I remember being angry when I saw, when I jumped on the call, 'cause I was like, why am I even, why is this, why are we interviewing this guy? Why did we waste the team's time and my time interviewing this guy? Because his LinkedIn shows like a very clear pattern of job hopping.
- AGAakash Gupta
Mm.
- FCFlo Crivello
And literally, the way I opened the call was like, "Soham, why can't you keep a job?" [chuckles] Uh, it, it was literally my first sentence in the call. Like, "Why can't you keep a job, man? Like, what's going on?" And I, I, I, I fully expected to just hang up. I was like, "Hey, it's not gonna be a fit. Bye-bye." Right? And he's convincing. I blame myself. I shouldn't have let myself be convinced, but like he, he had like a lot of really good excuses. He was like, "Oh, you know, this one, the company went down, and like this one, I had visa issues," and so forth. I also blame myself doubly because multiple of these exper- of these experiences, it, it's really inexcusable, frankly, for, for, for me, because, uh, at least two of these experiences were at p- at, at places where like I knew the founder, and so I should have, I don't know, I, I should have just shot them a text message. "Hey, how was your experience working with this guy?" And I usually do it. So we skip some steps, and we, we paid the consequences. So don't skip steps, and, and, and trust your gut and don't hire job hoppers.
- AGAakash Gupta
Yeah. So do back channel references, it sounds like is one thing. You already had the network with those people, so making sure that you have that as a step in your process.
- FCFlo Crivello
Yeah. Totally.
- AGAakash Gupta
Okay. The back channel reference wins again. Well, every time I talk about the back channel reference, all the PMs get really mad at me, [laughs] but from the founder perspective, you have to do it.
- FCFlo Crivello
Yeah. Why, why do people get mad? What, what do they have to hide?
- 52:59 – 1:07:41
Founding Lindy early, pivot lessons, and why ‘strategy is emergent’
- AGAakash Gupta
Yep. Yeah. We need a Glassdoor for that. So you were a product manager turned AI founder. I'm like, that's why I was like so excited to have you on because a lot of my listeners are PMs who aspire to start an AI startup. And one of the craziest things is that you really saw this AI agent wave years before other people. So can you break that down for us? What is the founding story of Lindy? How did you figure out, "I'm gonna pivot my life to AI agents," before just about everybody else?
- FCFlo Crivello
Yeah. I, I hate saying that, you know. It's like the whole, like I was in AI before it was cool, but I was into AI before it was cool, and I've been, I've been AGI pilled, and I have the record. Like you can look at my Twitter about it. I've been tweeting about AGI and, and the, the, quote-unquote, "singularity" for, for a very long time. So I've been, I've been very into AI for a long time, and I, I clearly remember like in 2016 or so, I, I was at like some AI research conference, and I remember ask-- I was the guy asking people about-- At the time, people didn't talk a, as much about AGI. They talked a lot about the singularity, and I was like, "The singularity," and all of that stuff. And, and literally people would laugh at you. It was like, it was like it was not an idea taken seriously even by AI researchers. I think people underestimate the extent to which a large part of OpenAI's success can be attributed to their, at the time, being the only ones taking the idea of AGI seriously.Certainly Google did not take it seriously, and Google was like the main, the main AI research hub. So always was very into AI, very passionate about it. I remember when GPT-2 came out, I was like, "This is--," I was so excited. And at the time, and I actually blame myself, like I, if I, if I really took it that seriously, I should have oriented my career around it. So I corrected my mistake belatedly. So when GPT-2 came out, I was like, "Wow," like this is, this is on the path to AGI, like this is happening. And then GPT-3 and, uh, came out, and then GPT-4. And for me, the tipping point was GPT-3.5 and the API came out, and that coincided with my previous startup not going so well. So my previous startup was selling, uh, a virtual office for remote teams. So it went really well at the beginning of the pandemic when everybody went remote, and then when everybody returned to the office, like the, the, the growth flattened. And that coincided with GPT-3.5's API coming out. And we, we just got sucked in, basically. Like we started building with GPT-3.5, and the first thing we built was like a meeting recorder because we, we had that meeting solution. And the second thing we built was making the meeting recorder, recorder update your Salesforce after the meeting. Because we had a sales team that went to us like, "Ah," like, "We hate it. We spend so much time updating the Salesforce. Can you get this thing to update the Salesforce for us?" And we're like, "Sure, absolutely." And basically at some point, like little by little, we build it for Salesforce, and then they're like, "Ah, I wish I could customize the fields that it updates." And so we build that, and then we've got customers that are like, "I wish it could also update HubSpot." And so we build that, and at some point it clicked and we're like, "Wait a minute. We can get CLLM to actually make the API call for us." That was around twenty twenty-two, and that coincided with LangChain coming out. At the time, it was like, it was not a company yet. It was just like a GitHub repository that was, that was going viral. And, and so that's when it clicked where we were like, "Wait a minute," like these LLMs are not just about generating copy and text. They can actually do stuff. So that's when this vision of the AI employee came together, and at, at some point we just decided to go, uh, all in and, and, and to, and to pivot to it.
- AGAakash Gupta
Wow. So it's a mix of responding to a variety of customer requests and then figuring out, okay, what if we abstract this one layer fur-further to build a full platform? Is that kind of the key insight that you had?
- FCFlo Crivello
I would say the key insight is even more meta, which is like, just get started, and I think your, your, your plan unfurls with time. I think people underestimate the extent to which strategy is emergent. And, you know, you, you-- I, I, I, I really think, like I forget who it is who said like, um, in- action produces information. And I think that the best way to think is to do. And so I think that's the advice. People ask me sometimes like, "Oh, you know, how should I go about like being a founder or like starting a company?" And literally the best and only advice I have is just do it. Like there is no, like don't overthink it. Don't do a grand plan. Like the only thing I know about your grand plan is that it's wrong. And so you, you-- It's true, like my grand plan was wrong. Everyone, everyone's grand plan is wrong. I think people greatly overestimate their own ability to plan, at least in granular detail. I think what you can do, however, is you can do a lot of stuff and, and you can, you can let a long-term blurry vision emerge. And I think it is fine for the vision to be blurry. It's like AI employee. It's just two words, right? It's like that's, that's the vision. That's what we're going two words. And then it- it's fine if, if the plan and the path to the AI employee unfurls with time.
- AGAakash Gupta
And I think the pivoting part is also a really interesting insight. There are basically two major pivots there, right? The first pivot from your first startup to this new startup, and then the second pivot from these general solut- these specific solutions to a more generalized tool. And I personally sometimes face this pivot challenge myself, you know. Like I have a pretty nice business here within product management. Should I pivot fully to AI, for instance? How do you personally convince yourself, or how did you in these cases convince yourself that, okay, the pivot makes sense?
- FCFlo Crivello
I think there's two meta heuristics here that you can use. One is frankly, you know it in your gut. And I think sometimes I feel people struggle to listen to their gut, and maybe because the word gut itself has become so ambiguous. You know, an alternative for it would be like, listen to what you find yourself being sucked in and what you find yourself being like genuinely interested in. Like, what do you do in your free time? Like, what do you wish someone else did? I love that question. It's like, it's, it's, it's ask yourself like, "What's the best startup idea that you think someone else should be working on?" Well, that's probably the one that you should be working on. That's number one. Number two, I think status quo bias is, is underrated. I think it is one of the strongest psychological biases we have, and I think it stems from our ancestral environment being an environment of scarcity. There was no abundance in our ancestral environment, so it was very much like a case of, you know, the, the bird in hand. It was like, just keep what you have, 'cause there's, there's very-- there's so very little around you. And I think this is, this is-- there is such a tremendous mismatch between that ancestral environment and our modern environment, which is in, in fact an environment of, of, of abundance, of tremendous abundance. And so I think that you should make a deliberate attempt to counterbalance that status quo bias you have. And I think that if you find yourself even considering a drastic change, despite that, that, that very strong instinct you have for the status quo bias, it probably means it's already too late. Probably means this is a change you should have done six or twelve months ago. And so I have found every major change I've made, I have found that I, I wish I, I, I'd done it six or twelve months earlier. And it's not just hindsight twenty/twenty. I think it's literally like when I think of the information I had six or twelve months earlierEvery time I'm like, "Yep, I could have made the decision six or 12 months earlier." And so I think it's number one, follow your gut, follow, follow your, your instinct and your, what you find interesting. And number two, make the big changes much faster than, than you think you should.
- AGAakash Gupta
Yeah. I think everybody would resonate with that, right? When we make a big change, we always look back and say, "Why didn't I do that six months earlier?" [laughs] And so recognizing that when you're considering a new change, I think that's a huge unlock. Now that you are an AI founder, would you advise all these PMs who are aspiring to be AI founders to become AI founders?
- FCFlo Crivello
I think if you want to be a founder, you should be a founder. Absolutely. Totally. I, I regret waiting as long as I, as I did. I, I stupidly let myself be held back by, like, visa and, like, savings and, like, a bunch of, of, of, of secondary concerns. I think it's, uh, Sam Altman who calls it the, the default life plan. Like, don't do the default life plan. Just go, go after what you want to do. Absolutely. I, I, I find it to be so meaningful. And if you don't do it, you'll never know. You'll never know whether you could have done it.
- AGAakash Gupta
So if somebody does decide to do it, what would be your advice looking back at your earlier version of yourself before you became a founder?
- FCFlo Crivello
I wish I'd known how little I knew. There was, there was certainly a, a fair amount of Dunning-Kruger. You know, you could also argue if, if you, if you know everything you don't know, I, I think it's, uh, Paul Valéry, he's like a French philosopher, who said, "How many things one must ignore in order to act?" I think, I think there is that, that naiveté that, that perhaps is, is, is a requirement to get started. I think there is something to be said about the YC advice. It-- at this point, it sounds a little bit like the drink your soup thing. But it's like, look, talk to customers, build product. That's, that's literally all that matters. Just give, give the market what it wants. And, and, and pretty much everything else is a distraction. People way overthink this shit. Like, all the networking events and the dinners and the conferences and, like, frankly, the investor coffees, like, you can skip all of it. You should skip all of it. It's, it's people, people fool themselves into thinking they're doing something useful, but actually they're just wasting their time. And, and frankly, it's a form of procrastination because they don't want to do the hard thing and, and useful and important thing, which is talk to your customers and build product for them. That's, that's all that matters.
- AGAakash Gupta
Yeah. I feel like there are so many wantrepreneurs out there who are, like, living this [laughs] wantrepreneur lifestyle. You really need to just get out there and get building. Was there anything that really helped you just get over that and say no to all these other activities that are, were enticing you, and stay focused on building?
- FCFlo Crivello
I spent a fair amount of time reflecting and, like, looking back. Like, I, I set aside a couple of hours every week and, like, a couple of days every, every year where, like, I literally just sit down and think. And I, I look back at, like, the mistakes I made, and I look back at, like, the things that mattered. And so I'm like, okay, at any given point in time, there is one thing you should be focused on. There is only one question that matters for you and your company. And so I like looking back at, like, six months, 12 months earlier and, like, it's, it's, it's funny because we live in a sort of inverse fog where it's like the present is a lot less clear than the past. So you can look at the past and you can be like, okay, six or 12 months earlier, like, what was the thing that mattered? And if you do that thinking in writing, then you can engage in this level of, of meta thinking. So you can look back and reread what you were thinking at the time and see what you were missing. And you can also, once you have identified that in hindsight that this thing was the question that mattered, right? And this is what I was thinking about. And when I look at my calendar, this is how I was spending my time. And oh, fuck, like, that week I spent, like, three or four hours going to a bunch of, like, random, like, AI events and, like, dinners and all of that stuff. Like, this time would have been much better spent focusing on this thing instead. And so I think you do this often enough, and at some point it just, it just, y- you just internalize it. You're like, yeah, like, all of that stuff is, is, is a waste of time. In general, whatever is the most important question that matters most for you right now, these things are not going to make you make progress on any of these questions.
- AGAakash Gupta
Yep. And there's just an infinite [laughs] amount of them, especially if you are in San Francisco. It's like you're looking at your calendar, you're, like, having so much FOMO. But building fundamentally is the core of it all, and it's not always a straight line. At least that's what I've experienced. There are some serious lows as a founder. What has been your hardest moment since founding Lindy?
- FCFlo Crivello
The pivot from, from TeamFlow to Lindy and, and, and in particular, actually we don't even call it... There was one day where, like, we fired, like, two-thirds of the team, like the go-to-market team. Because we were like, "Okay, we're moving away entirely from this old business to this new one, and we don't need a go-to-market team. We're gonna be building, and we, we need to be lean and small again." But actually we don't say that was the, the, the hardest moment. I think the hardest moment were the couple of months before that. And I think in any pivot you traverse this, what I call, like, the valley of death. And it's, it's basically you're, you're in between two mountains, right? You're in between the two ideas. And that valley of death is so confusing and so unclear and so rife for intense emotional conflict in the team. Because it's, it's such a time of uncertainty and, like, there is no right answer. It's really confusing. It's really ambiguous. And everyone's got an opinion. And, and the stakes are high, right? It's, like, people's jobs at stake and, and, like, the future of the company. And, and yeah, I mean, like, it's, it's, it's a time when, like, people are demoralized. It's a time when people quit. It's a time when people fight you, and it's a time when, when, when people question you as a leader, and it's, it's painful. And, and i- in a way, I know it sounds cheesy to be like, "Oh, I'm grateful for the hardship." Like, fuck that. I, I, I wish the hardship had never happened. But truth to be told, I think, I think entrepreneurship is, is, is a tremendous vehicle for personal growth. And I, I, I, I did grow through that period. And I, I think one of the things I have found a lot of entrepreneurs to share, perhaps counterintuitively, is that they actually tend to be surprisingly humble. Sure, right, they, they, they have a big ego, ego and stuff, but, like, they, they, they, they have accepted so many times that they were wrong. They have screwed up so many times that in a way, in a way you-You learn perhaps to decouple your self-worth from how your company is doing, because otherwise you can't make it. You, you learn to, like, not tie your self-worth to, like, your ARR growth or anything like that, like month to month to month, and, and, and you learn to, to... I think there, there, there's some level of faith. It's like, "Okay, we're in trouble. There's, like, a problem right now that I need to handle, and it's, it's fine. It's part of it, and it's not a reflection of, of me." It's just-- Well, maybe it is. Some people decide. [laughs] It's like, it's, it's, it's most of the times it's not a reflection of me. It's like we're gonna, we're gonna see it through, we're gonna work on this problem, and after this problem, there's gonna be another one, and then another one, and that's the journey. And you can't feel bad every time you have a problem. You have to rewire yourself to override the instinctive negative reaction you have when these problems come up and replace them by, like, almost a reaction of excitement, right? Of, like, gearing yourself up for the challenge.
- AGAakash Gupta
Yes. You-- If you are not ready to deal with getting punched in the face every morning when you wake up, you're [laughs] not gonna like being a founder. That is just the reality of it, at least for me. It's like every day, "Okay, now this person is mad at me. Now this thing I completely made the wrong decision on." So when people are thinking about becoming a founder, what they're really always honed in on is, like, what should I be building? And I think you probably have the world's most unique insight on this business model that everyone wants to build these days, which
- 1:07:41 – 1:16:32
AI agent agencies & enterprise transformation: AI ‘czars’, centers of excellence, and the road to autonomous companies
- AGAakash Gupta
is helping companies set up their AI agents, these AI agent agencies. What do you think of that as a business model?
- FCFlo Crivello
Yeah. I think it's huge, and we are actually working with a lot of those and, and some of them are making really good money. It's pretty- Sometimes I'm like, "Fuck, like, we are in the wrong business." Yeah, I mean, I, I, I think it's a really big one. I think companies are very eager to, like, transform themselves. I think it's very similar to, like, digital transformation, right? When, like, computers came about and you had all of those, like, old companies. Your, like, Procter & Gamble or, like, General Motors and all of a sudden, you've been running on, like, fax machines at best, and all of a sudden you need to entirely rethink yourself and migrate to, like, well, at first to, like, just software in general and, like, you know, like, having data centers and all that stuff, and then the cloud. Like, it's just a totally different way of thinking, and I think every company is going through it again, but it's, like, AI transformation. And they're very eager to go through it because they're excited, but they're also afraid, and they're confused, and they don't know what to do and how to do it and what solutions to adopt. So there is a lot of demand right now for this kind of service, absolutely.
- AGAakash Gupta
So should every company be using AI agents?
- FCFlo Crivello
Yeah, just in the same way that every company should use computers. Like, it doesn't matter if you're selling software or if you're selling bottled water, like, you, you, you need to be using computer and AI agents, absolutely.
- AGAakash Gupta
And if you're a really hesitant company, I don't know, you, you're really focused on HIPAA compliance in healthcare, or you're really focused on governmental compliance in fintech, what is the first area you should be thinking about putting AI agents in your business?
- FCFlo Crivello
Well, I, I, I think the compliance concerns are valid, and there's a lot of AI-- Like, most AI agent platforms at this point, like, you will find one that checks your boxes. That's w- w- By the way, we are HIPAA compliant, we're SOC 2 compliant, and so forth. So you, you can find these platforms that, that check your boxes. Um, I, I will say, though, there is a lot of alpha right now. There is an opportunity. It's almost like an arbitrage opportunity because there is this window where it's like if you're the first ones to adopt these innovations, the gap between the practice that, that companies do, like, the, the, the way the average company operates right now and the frontier of, like, the best way a company can operate has never been so big. And so that's, that's a time for pioneers, and that's a time for people who are ready to rush to the frontier and to sometimes take, take shortcuts. So we are seeing a lot of our customers be like, "Yep, we're gonna, we're just gonna, like, blast through the internal walls to, like, implement these AI agents because we cannot wait."
- AGAakash Gupta
Is there any advice around you could hire an agency like we just talked about? Is there anything that you've seen from your customers where, I don't know, they created a AI agent center of excellence or something like that, that a nugget people can hold onto that will help them propagate AI agents through their organization?
- FCFlo Crivello
Yeah. That, that's certainly a, a pattern we're seeing where... And i- very often it comes from the CEO and, and it comes from someone in the company who is particularly AI-built. And so the CEO very often takes that person aside and, and goes like, "Hey, do you wanna be, like, the AI czar in the company?" Now, I think one important thing to, to note here is, like, there is no surer way to make sure that something does not happen, uh, than to make it someone's halftime job. Uh, you've gotta, you've gotta make it, you've gotta make it the, the full-time job. So w- w- the companies we see be most successful in their AI transformation are the companies that, like, pull. And it's painful because very often that person who's AI-built tends to be good, right? They, they, they're on top of their shit, they stay up to date and all of that. And so you, you gotta pull that person from the important thing they do, and it's a good person doing an important thing, and you've gotta have them sort of go into the unknown and shepherd the AI transformation inside your company.
- AGAakash Gupta
You've said your goal is for Lindy to autonomously operate companies [laughs] in 10 to 20 years. What are, what is the gaps right now from present to that future?
- FCFlo Crivello
Yeah. I, did I say 10 to 20 years? I think I, I used to a very long time ago. I think my, my current timelines are much, much, much shorter than that. The gaps are... Well, the models obviously are, like, not good enough right now. I think in particular for, like, long time horizon tasks, like, they lose coherence really quickly, and I think it's underrated the extent to which people always talk about the context window, but I think there is such a thing as the effective use of your context window. Like, increasingly, we basically have not infinite context windows, but, like, a couple million tokens is, is a lot of context, right? But we actually see dramatic performance degradation the more your context grows. SoThat is, that is very much an unsolved issue. Uh, so I think the models are still, are still very much improving. I think memory, in particular, is, is one very active ar- area of research, and I think it's, it's currently the most important one. And I'm, I'm, I'm very bullish. Uh, I think, I think we're gonna see huge developments in the next year as far as memory is concerned coming from the labs.
- AGAakash Gupta
Yeah, the context one is the one I personally struggle with the most. It's like the, when you create that giant prompt or you create that giant editor workflow, all of a sudden the performance starts to decrease. And so that seems like a technical roadblock that the foundational model companies will have to solve, right?
- FCFlo Crivello
Yeah, and there, there are workarounds to that. So, like, when you build agents in Lindy, for example, like, we do give you a step that's like clear context, and you can pass a prompt to that step that's like, "Hey, before you just nuke the entire context, what do you wanna extract from it? Like, what, what should we keep for, like, what do you know about the rest of your agent's existence that's going to be, that's going to require extra, like, the context that is already generated?" And eventually, we're gonna build some of that behavior in by default. So there are workarounds to it, but yeah, I think, like, mostly this needs to be handled by the model better.
- AGAakash Gupta
So we've been talking about AI-native companies. I wanna talk a little bit about big tech. What big tech companies are just totally missing the boat on AI agents?
- FCFlo Crivello
Frankly, I've been disappointed as a whole from, about the, the incumbents. Like, it blows my mind, and I'm sure there's, like, excellent regulatory reasons or, like, like, legal reasons why they can't do it, but, like, it blows my mind that none of the main ebo- e-book platforms right now have implemented AI. Like, the most obvious AI use case for me, like, I'm, I'm reading a book, I wanna talk all the time. Like, at this point, like, my, my workflow is, like, I read a book and I have my phone next to me, and I'm talking to ChatGPT, like, nonstop. So, um, I, I, I think it's the entire inc- like, look, the fact that Google Home and, like, Alexa and so forth are still not running on the new generation, like, that just blows my mind. Like, what are they doing? And I know what they're doing. Basically, it's like the, the, they don't wanna nuke their code base 'cause they've spent so long writing these millions of lines of code, and they don't wanna just remove it all and replace it by, like, this one black box that's called an LLM. Like, I, I understand that. But I, I think the whole field is, is screwing up. I think, like, the, the companies I am currently most bearish on are the, are the, the BPOs, like the business processes, process outsourcing. Like, these companies who have, like, floors of, like, support agents in, like, the Philippines or India or whatever, they're basically selling tokens. So I think those, those guys are in trouble unless they, they transform themselves very deeply, very fast.
- AGAakash Gupta
Yeah, the Telus of the worlds or the TCS. So if there is somebody listening at one of those companies, how can they save themselves?
- FCFlo Crivello
I think it's like textbook innovator's dilemma. I think you've got to accept that you are going to invest a lot of your time and a lot of your money in the technology that is worse than the status quo, frankly. Like, it's, it's, I'm sure it's just not as good as humans yet. And, and, and that actually also will come in conflict with your existing business. So it's like it's gonna have, it's gonna open very uncomfortable conversations, both internally and externally. Like, internally, people are gonna be like, "Oh, we're gonna be out of a job. Like, what's going on?" Like, and why are we, like, resources are tight. Like, why are we setting all our resources to this thing? And externally, like, you're gonna have to be able to install the question from your customers, like, which ones should we use? Like, should we use your agent thing, or should we use your, your, your human thing? So that's, that's the painful resolution you have to, to make every day with deep conviction.
- AGAakash Gupta
Disrupt yourself or be disrupted.
- FCFlo Crivello
Yeah.
- AGAakash Gupta
I could go on for another 90 minutes, but we are at time here. If people want to find you online, where should they go?
- FCFlo Crivello
I'm active on Twitter and on LinkedIn, apparently. But, uh, altimor on Twitter, and again, we are hiring, so please shoot me emails.
- AGAakash Gupta
Amazing. Flo, thank you so much for your time. Thanks for being on the podcast.
- FCFlo Crivello
Thanks, Aakash.
- AGAakash Gupta
So if you wanna learn more about how to shift to this way of working, check out our full conversation on Apple or Spotify Podcasts. And if you want the actual documents that we showed, the tools and frameworks and public links, be sure to check out my newsletter post with all of the details. Finally, thank you so much for watching. It would really mean a lot if you could make sure you are subscribed on YouTube, following on Apple or Spotify Podcasts, and leave us a review on those platforms. That really helps grow the podcast and support our work so that we can do bigger and better productions. I'll see you in the next one.
Episode duration: 1:16:42
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Transcript of episode fbeNnNh1LGY
