The AI Wave Everyone's Missing (Crew AI Agents)

When I created Creai as an open-source project, we gained significant traction in the open-source community. We started trending on GitHub, and people began discussing it on Twitter. I would attend meetups and unexpectedly run into individuals from Oracle who would say, "Oh, we're actually using Creai." I thought to myself, "Damn, what is happening?" Agents are the word of the moment.

Everyone's saying this, right? But we have been working on this for over a year. AI agents will just provide more opportunities, right? Well, you're seeing companies—huge numbers of them—deploying these AI agents. The companies that are doing the best are the ones that are actually promoting their employees to manage these agents.

We use agents for a lot of tasks. What are some of the use cases you're utilizing at Crew internally? For the most straightforward tasks, such as preparing for meetings, we have agents that automatically research the company, the person, and anything else that might be relevant. Then, we have more complex applications. For instance, every pull request is currently reviewed by an AI agent. We're also using them for documentation and support tickets. Essentially, we are leveraging agents in various aspects, from coding to sales, marketing, and back-office operations.

What were they doing to generate 600 leads a day? Honestly, the way I look at this market is that it is going to be huge. I mean, a huge market. AI agents are not going back to the bottom. If anything, more and more companies are going to adopt them.

All right, Joe, back into the bottom. If anything, more and more companies are going to adopt that. All right, Joe, welcome to the podcast! Hey there! Thank you for having me. I'm very excited to be here, honestly.

Thanks for coming on. I remember when we were talking before, you mentioned there was a funny story related to your name and AI agents. So, honestly, we might as well start there. Yes, so this is what happened.

Early in the day, the creation of Cree AI as a company felt almost organic. When we started as an open-source project, it truly was an open-source initiative. I needed it because I had built a few agents and became so hooked on them that I wanted to create more. Like any other engineer out there, I didn't want to write the same code over and over again. So, I thought, "Alright, how can I package this?" But the problem is that it started to gain so much traction.

And then I started to have all these companies reach out to me and say, "Hey, we want to deploy this in production." So I had no other option but to turn this into a company. The first thing I needed was marketing, so I set up agents to handle marketing, scheduling leads, and calls. I granted them permission to schedule things in my calendar, but I never blocked off my lunch time. As a result, I ended up with two weeks of craziness back-to-back. Because my name is not easy for any Portuguese speaker to pronounce, these 15-minute calls often spent five minutes just trying to pronounce my name.

And I couldn't cover enough, so I started to just say, "Hey, my name is Joe. Let's get to it," just to make those extra minute or two. I had the whole questionnaire to ask questions for this customer to gather feedback. It was frustrating spending like two or three minutes on that. So yeah, that's what I go by now: I go by Joe.

That's hilarious. But also, hold up. How is it that you accidentally booked your calendar for two weeks of meetings? What were the agents doing? Well, the agents I initially worked with were focused on creating marketing content.

So, they would help me post on social networks, mostly LinkedIn. I would still provide high-level ideas, maybe one-liners about topics I wanted to post about. They would help me identify the key points. These agents would then research what others are saying about those topics and review my history, including my resume. I even documented my life experiences and created embeddings from that information. The agents would analyze my life experiences to find different angles that aligned with what I wanted to convey, generating a significant amount of inbound interest. Initially, I received around 10 leads a day, which eventually grew to 50, 100, and then 200, reaching up to 600 leads a day – it was insane! I needed to connect with these people, so I built another set of agents to enrich the leads, prioritize them, and attempt to schedule calls for the most important ones.

And that's when things went crazy. This agent was trying to schedule meetings, and it got a little hectic. I still have screenshots of it, and I'm going to ask for those later. But also, what were they doing to generate 600 leads a day? Was it scraping LinkedIn, or was it through posts, or what? No, no, no, that is inbound. That was crazy! I mean, 600 leads were later; during that period, it was maybe 100 or 200. But yeah, things just started to pick up. Right now, agents are the buzzword of the moment, and understandably so. We've been working on this for over a year, so we saw the trend heating up along the way. We recognized the increasing demand for leads, but back then, it was around 100 a day. The problem was we already had thousands of people in the pipeline, and these agents were essentially trying to chip away at that pipeline and schedule all these calls. It was an interesting experience. That's intriguing, and I'm guessing that before you were the founder of Crew AI, you did a stint at Clearbit, where you led many of the AI engineering efforts. You know, VCs often talk about wanting to invest in second-time founders or in founders who have done something incredible in their last role in the same market they're entering to start a new company. It sounds like that aligns with your persona as a founder when you set out to build the initial product.

How many of those leads came from your last role? Not that many, honestly. So, I think, like honestly, I had done other business in the past. There was one that was venture-backed, but that was kind of like 2014, so that was a long time ago—nine years ago, back then when I was doing Crew AI.

And that was it. I tried a bunch of different things. I tried hardware startups, and I promised myself never to do hardware startups again. It's so hard—so hard— and there are so many extra layers of complexity. So yes, to everyone doing hardware startups out there, my feelings are with you; I'm thinking of you all. If you're considering building a hardware company, don’t. I mean, it's hard.

It's very, let's say, rewarding because you have something physical, and there's something special about that. If you're a builder or a creator, you can't get that from software alone. But the thing is, taxes alone create a major problem along with all the components. Additionally, where you’re going to bring these components from and the port they are going to come through can affect the taxes you pay on them. So, there are a lot of extra layers of complexity, at least on the global level that I was trying to navigate.

But Clearbit was amazing. I was at Clearbit for five years, leading their main enterprise product and now the AI project. When I joined, I think we were at a little under $10 million ARR. That product went from zero customers to becoming the major revenue driver, almost five times that amount.

And it was an incredible experience! I got to meet a bunch of amazing people, and luckily I'm glad to still work with some of them now at Crew AI, and that has been very exciting. When I left, one of the initial use cases that I thought of was that AI agents could definitely help with data enrichment, so that was very much top of mind for me. Did you build an MVP before you started fundraising, or did you write a memo and send it out to investors? No, it was actually because I had some experience with open-source projects in the past. I spent about ten years in the Ruby on Rails community, and I used to lead a significant project called ActiveRecord Serializer, which was part of the Rails API. I did a lot of conference speaking at Ruby and Rails conferences during that time. Through that experience, when I created Crew AI as an open-source project, we gained a lot of traction in the open-source community. That was what started to draw attention from investors. We began to trend on GitHub, and people started talking about it on Twitter. I would also go to meetups and run into folks from Oracle who would say, "Oh, we're actually using Crew AI," and I thought, "Wow, what is happening?" It was very organic and incredibly rewarding. I was fortunate enough to have VCs reach out to me, and I had to decide whether to jump into this full-time or not. By then, Crew AI was a fully-fledged open-source project supporting thousands of use cases. Got it, that makes sense. You guys announced your $18 million Series A two months ago—congratulations on that! What was your timeline for fundraising? I have to say it was very fast. We did a pre-seed round, and then, within about two months, our investor approached us and said, "You know what? We want to preempt your seed."

We see the traction that you started gaining, and it coincided with our work on an enterprise version. I thought about it, decided to move forward, and we did, which was great. We then launched a beta program for the same enterprise product, and things began to go crazy. Other VCs started reaching out, indicating that they thought we were ready for an A round. I believed that it would be beneficial, as it would allow us to scale several aspects of the business. Whenever you're considering raising a round, there are pros and cons to weigh, and you have to think it through carefully. However, we found a strategic way to approach it that prioritized some of the things I believe are crucial for the company, and they still are. So, we said, "All right, let's do it." I have to say, it was a great decision.

Since then, we have been able to leverage not only the funding but also many of the connections that come with it to their fullest potential. It has been a transformative experience. I think going the venture capital route with Crew was definitely the right decision. Awesome! That's good to hear. After you received the initial funding, what was the first thing you did to go to market?

Because it's also different, right, when you're building something open source compared to when you're trying to build a closed source, straight-up B2B SaaS tool or B2B enterprise tool. Yeah, I have to say, this was one of the concerns early on with investors, right? And it's true that it's not an easy model. There are companies that have succeeded in this space; for example, MongoDB is a great use case, as is HashiCorp. However, there are also companies that have struggled with it.

Docker was a company that struggled with that, and there are many others that have faced similar challenges in the past. So it's definitely something concerning, but I think at the end of the day, it's largely about the founder. Even though we have an open-source project, I am a very capitalist founder. I am definitely building CRI to ensure that it generates substantial revenue, and I believe that comes across.

So I think that it really kind of put them at ease a little bit. And the other thing as well is that people often think it’s either going to be a closed search or an open search, but I believe it’s more of a spectrum. Honestly, there are so many reasons not to box yourself in. You can still do a free tier, which is pretty good; people can still use it. You need to make the code available so they get the opportunity to use it in your product. I think there's a lot of leeway if you really think it through, and that has been working pretty well for us. Got it. So where did your first thousand users come from? I have to say, I don’t even know. Open source is so expansive; there are so many people using it from everywhere. I know that the first customer we closed was from Pakistan, the second one was from Australia, and then things took off from there. The growth was significant in the U.S., and then a lot of users from America in general as well. Europe is moving, but not as fast as I would expect. It’s a global reach.

So, how did those first few customers find out about your product? Was it Twitter? Was it through open-source communities? Open source, I would say, is our biggest distribution channel right now.

If you do it right, it can be such an unfair advantage. Again, you've got to tread carefully. A lot of the decisions you make in open source are not two-way doors. I think it adds an extra layer of trust to your building and relationships because now it's not only about the customers; it's about the users and the developers who are becoming invested as well. It's almost a commitment that you have to the community, and you must treat that seriously. If you don't, they won't take you seriously either. That makes sense. Now, how does someone go about becoming part of the open source community? Let’s say you have a developer who just dropped out of school or recently graduated. They know how to build things, and they have an idea for something they want to create. How do they go from being unheard of, perhaps with a private GitHub account or similar, to actually becoming part of the community?

I have to say, I still remember back in the day when open source played such a huge role for me. For me, it was all about the projects I was working on. The way it worked for me was that while I was working on a project, either for myself or for a customer, I would often need to use a library that lacked a specific feature I required. So, what I would do is go around and try to implement that feature myself.

In the process of doing so, it's amazing because if you think about it, you get some incredible people to review your code. I still remember becoming great friends with, for example, Steve Klabnik, who contributed significantly to Rust at Mozilla, and it was all because of open source. I had the opportunity to have him review my code, provide feedback, and point out where I was going completely wrong. I don't know if I would have had the chance to work with him otherwise. If you do it right—and it’s not necessarily for everyone—engaging with open source can open up doors for you. It led to many speaking opportunities at conferences, which really helped me. My conference speaking, coupled with my work in open source, opened doors to even better job opportunities than I previously had. I remember doing a significant talk at RailsConf in Atlanta, where I was launching a new version of my open source library on stage. People were very excited about it, and as soon as I did that, it felt like everything just took off.

People are very excited about it, and I even better did it. As soon as I got off the stage, someone approached me and said, "All right, we're taking you out to dinner; we have a job offer." I ended up going to work for that company, working on the same open-source project, but now with their implementation. It's so funny. I remember having a conversation with someone recently, and he said something that was amusing: "You know, if you try to build a company and do it publicly, you may or may not end up with a successful company or get investor interest, but you will definitely get job offers." I feel like right now, especially in a market where I have friends who are trying to get jobs and are kind of struggling, I've been telling people to take some time off—two or three months—and just build something while being very loud about it online. Best case scenario, you end up starting a company; worst case scenario, you end up with a job. So, it's a win-win. Yeah, exactly. I think, honestly, I want to be respectful about everyone's tough moments right now. We're in a weird economic stage as we come out of COVID; there has been a wave of layoffs. There was a clear cultural shift; it feels like everyone is on edge about everything. Have you heard about companies implementing peer management roles?

So everyone is trying to do more with the last few people available. That said, it's also a very exciting moment, right? If you think about it, everything happening with AI has taken us 20 years to evolve from the dot-com era to now. It might take at least another 20 years for us to see another wave like this. This creates many opportunities that people can definitely leverage.

And, you know, speaking of opportunities and hiring, building AI agents today feels like building SaaS or software products like 10 or 15 years ago. Back then, software made tasks easier to complete, whereas now, if you build AI agents, they can essentially do most of the tasks for you or even the entire task, depending on the parameters involved. You guys are very interestingly positioned because you're essentially building the picks and shovels. This means that instead of people having to build these agents themselves, they can build them on top of your platform. I think it's fascinating; I was discussing this with someone, and I even made a tweet about it. It's like when software started charging recurrently as a service, people began to refer to it as SaaS, meaning software as a service. Now, you still have SaaS, but it's essentially the opposite: things that used to be services are now being offered as software.

So, why don't we hire someone to do research, for example? You could have AI agents do it for you. It’s still software as a service (SaaS), so we flip it around. That's funny! Some people say that AI agents are going to eliminate jobs and that they are the worst thing in the world, and that we need to stop them. However, a more measured perspective is that AI agents will create more opportunities. Suddenly, you can get something eight-tenths of the way done, and then a human in the loop can help complete these tasks. Where do you fall on that spectrum? I have to say, we are seeing companies—hundreds of them—deploying these AI agents. Honestly, in the enterprise space, we have many hundreds of customers by now. The companies that are doing the best are the ones that are promoting their employees to manage these agents. It’s almost like, “Hey, you’re still responsible for reviewing the output. You’re still responsible for presenting it to ensure that it’s valid.”

Like you're still doing it. It's just like another tool in your toolkit. Even we at Crew are definitely over-indexing on agents; we use agents for a lot, and we do a lot of dogfooding ourselves. I can tell you it's not like we're spending any less money because of that. If anything, it's just making us a little more excited about all the other areas of the business where we can invest more now because of it. Okay, so then let's talk about that. What are some of the use cases you're utilizing at Crew internally? For example, from the most simple things, we automate preparations for meetings. With our busy schedules filled with meetings, we have agents that automatically research the company, the person, different angles, news—anything that might be relevant. They even generate ideas for potential use cases that we should be considering. So, when we jump into those meetings, we feel like instant experts on the other person. For instance, we might say, "Hey Omar, it's amazing that you opened a new factory in Frankfurt! It's great to see everything that you folks are doing with AI. I think there are potential use cases in your procurement and legal areas; let me talk a little bit about that."

So those are the use cases, and then we have more complex ones. Currently, every pull request is reviewed by an AI agent. Most of our documentation is written by AI agents. Many of our support tickets now have AI agents perform a first pass. Whenever we finish a meeting, we have AI agents that write up follow-ups based on transcripts and even draft emails and update HubSpot.

So we're using it for a lot of different aspects, from coding to sales, to marketing, to back office, and everything along the way. There are many ideas in there that I want to discuss. I guess the first one we can start with is simple, right? Let's say I'm doing this podcast. Right now, the flow is very straightforward.

I've done enough of these to where I can kind of jump in. I'm mostly interviewing people for a podcast, and right now the flow is very simple. I've conducted enough interviews to where I can easily engage with the guests. I'm mainly interviewing people I know, whether they are friends or individuals I have followed online, so I don't have to do that much research.

Just a few questions that I wrote down beforehand. Um, if I were to do a podcast with someone completely new, what does the workflow that I'm building look like? What am I telling it to do? Is it like, "Oh, go scrape Twitter"? Is it a LinkedIn scraper tool? What does the actual workflow look like? I have to say that early on, the way I found best to create these automations is to literally try to understand the process myself—how I would go about it. For example, if I were to research someone, yes, I would check Twitter, I would check LinkedIn, but I would also try to find similar people, look for competitors, and see what people are saying about this person or this business. So, I think about all those different angles. They go beyond just saying, "Hey, go online and scrape this person's profile and their last X amount of tweets or X amount of posts." It might involve trying to understand their industry, finding news about it, and seeing what big companies operate in that industry and what is happening with their stocks. I would also consider how political changes or news might impact this person. So, everything related to those factors creates all those different chunks and angles that you can then have agents use to write a complete report. You could ask, "Write me 10 hot take questions that I could ask this person based on all of this and other relevant information." The beauty of it is trying to think about what you would do to make this process super special. You will probably do it a couple of times, allowing you to validate the process, and then you can transform that into a proper agentic workflow. Now, when I'm creating an agent on Crew, am I using just plain text?

Does it use natural language processing, or how does it work? It's up to you, actually. If you're using the open source version, then yes, you're working with code. However, the code is quite straightforward, allowing many people, even those who have never coded before, to use it effectively. This is because many of the agents and task definitions are actually configuration files.

So, they are YAML files, which are pretty easy for you to navigate. However, if you want to build completely with no code, you can access our enterprise platform, where we have a studio feature. In this studio, you can chat your way into an automation as you explain what you're trying to achieve, and it will automatically select tools for you, create agents, generate tasks, and provide you with the ability to modify them along the way. You can then download the task code or deploy it right away, turning it into an API. Now, shifting gears from research to go-to-market strategies: if I'm, let's say, the CMO of an AI workflows company, specifically Crew, and I am now trying to acquire more enterprise customers, it is very clear to me who our ideal customer profile (ICP) is.

It's like, okay, we're in it for the audience. It's very obvious who our ideal customer is. Now it's like, where do I go from there? In my head, I know this is who we're targeting. We're targeting—correct me if I'm wrong—CEOs, CMOs, and CTOs of large companies, whatever the actual Ideal Customer Profile (ICP) is.

And then where do I go from there? What I would do is, again, think about how you would act if you were a person, right? The first thing is that cold outreach works. Yes, it does. But if you have signals, it’s better. So what do signals look like?

Signals indicate that people are interested in that profile. They're liking your posts, interacting with you, commenting on your profile, and sending you messages. If I were trying to be effective, I would look for the Venn diagram of individuals who are actually engaging with me. There are other tools that can provide you with that information if you don't want to go through agents and manually map this out. Honestly, this could require regular coding, which may necessitate an agent to handle the initial mapping. What you could do is use tools like Common Room. For example, Common Room excels at monitoring overall interactions. Once you pick up that signal, you can then get agents to research that person further, diving into every angle such as what other posts they have liked related to you, your competitors, or others. This can help you create a narrative about how to approach them and what the story might be. In my last episode, which hasn’t been released yet, I spoke with Chris from Hanover. He utilizes a strategy he calls "inbound-outbound." At the end of each day, he reviews all the people who have liked his tweets, commented, or engaged with his profile in some way. He makes it a point to consistently post broad, high-level tweets that attract founders and investors in general.

And if someone fits his Ideal Customer Profile (ICP), he'll go and DM them a link to the tweet they engaged with or the LinkedIn post and say, "Hey, you engaged with this post. Here's my cold pitch." It's like two or three lines. That's how he's actually gained most of his customers. Now, he has a billion dollars in assets under management for his fund and his fund software platform.

And so it's like some version of that meets this, which is now using agents instead of manually doing it yourself. It sounds like you have a really good go-to-market strategy built out in just a day or two. I have to say, if anything, this is something I learned from Clearbit. Clearbit was all about marketing and sales; it was a marketing-driven and sales-driven software. If anything, I learned a lot from all the companies out there that are focused on sales and marketing. You learn that yes, it's about the product, yes, it's about the ideal customer profile (ICP), but mostly, it's about the timing. The right product and the right ICP at the wrong time turn into nothing, but at the right time, sometimes you can create opportunities that you would not even expect. So I think there are a lot of angles to consider. The ability to find that timing is something that many people can sometimes overlook.

And I got to say, there are companies out there that are very savvy about this. There are literally things that you would never think about tracking. For example, monitoring job postings can give you a lot of signals. A job posting from a company tells you what tags they are using, what new text they might be adopting, what tags they are dropping, and what market they might be targeting. This information could indicate what they might be trying to sell or what you could be telling them. At the end of the day, like business coding, it's our story, right? You have to tell a story. The one who tells the most convincing story has the potential to make that story a reality. Yes, absolutely! That's a very interesting point. Actually, I hadn't thought about hiring posts in that way.

Obviously, if you're building a recruitment company or an HR tech platform, then monitoring hiring posts makes sense. But even beyond that, being able to tell what tech stack a company is using and what is important to them at the moment is something a lot of people could definitely find useful. Exactly. For instance, if you see a significant blog post announcing the hiring of a principal engineer that mentions the need to manage millions of roles on PostgreSQL, Supabase would probably want to know about it. This could be a good indication, right?

Or if you're saying, "Hey, you're going to be processing millions of events," a Kafka provider is probably interested in that as well. They might be selling into that. So, there are a lot of signals coming from all those small details—from job postings to people changing jobs and many smaller things you might not even consider on a day-to-day basis. How often are you reaching out to someone, getting on a call, and discovering that they're using one of your competitors? It feels like you're in a very early market, and you guys are among the first to enter it.

Yeah, I have to say it's not very often. It definitely has been changing. It was not very often and depending on the geography as well. So, if it’s Australia, Europe, Latin America, or the U.S., it changes the distribution a little bit. But the market is definitely—honestly, the way that I see it—is going to be a huge market. I mean, a huge market. AI agents are not going back to the bottom, and if anything, more and more companies are going to adopt that. So if we believe that to be true, there is a consequence to that. No one's going to let a few companies take it all. All these other companies are going to be building on this, and we’re talking about everyone—not only startups but all the big players as well. They will be joining this market and taking their steps in it, so we’re going to have competition for sure. I believe the market's big enough that there won’t be one player that wins it all. That said, there will be one player that wins big, and I guess for most people now, if you ask them, they would say that we are the ones best positioned to do that. So we better execute. Awesome! I'm going to ask you that question I ask everyone when we're kind of at the end of the podcast, which is: if we were to do this podcast again in a year, what would you want to accomplish by then to consider the past year a success? Next year, well, I have to say I think 2025 is going to be the year of AI agents.

I would love to see how much of that we're going to capture. Many people are saying that 2025 is the year of AI agents. For me, next year will represent a significant portion of that market, and I believe that would be a great success. I think we're positioned to achieve that.

I think we're capturing it very well. But once that thing starts to really heat up this year, I think the one who scales better is probably going to be the one that wins most of it. I'm very curious to see how that's going to play out. Love to hear it. Awesome.

Thanks for coming on. No worries! Thank you for having me. And yes, it's great to see all of you who are watching. I appreciate you for hearing me out.

Catch you online.