Luke: [00:00:00] You’re listening to a new episode of The Brave Technologist, where we explore how emerging technologies like artificial intelligence, machine learning, and Web3 are influencing our lives, our choices, and our collective future. Soon, our team’s headed to London for the AI Summit, where we’ll have the chance to sit down with founders, builders, policymakers, and researchers to explore how cutting-edge technology is impacting the lives of everyday users.

We look forward to releasing these episodes to you later in June. This week on the podcast, you’re gonna meet Joe Mora, who’s the CEO of Crew AI, the team building solutions for enterprises to operate teams of AI agents that perform complex tasks autonomously, reliably, and with full control. Joe has over 20 years of experience with software engineering and specializes in building high-performing teams and implementing scalable AI and machine learning solutions.

In this episode, we discussed how to think about accountability and trust in a multi-agent AI system, how agents can potentially watch other agents to improve safety and [00:01:00] reliability, ways to use LLMs as judges to create quality assurance and building guardrails, and h- advice for founders and CEOs building emerging technologies.

And now for this week’s episode of The Brave Technologist

Joe, welcome to the Brave Technologist. How you doing today?

Joao: I’m doing great. Very excited about this.

Luke: Appreciate you making the time. This is a totally a topical, area to go into today, given how, much of the conversation is around agentic AI and people don’t really necessarily have a good sense of what it means or whatever.

But you all at CrewAI are doing some really cool stuff so CrewAI is built around this idea of multi-agent AI systems working together like a crew or a team. What’s a real world breakdown or frustration, made you realize that the single agent wasn’t going to cut it?

Joao: I gotta say, it’s so funny. When I started back on ‘23, I think people were… There was a lot of skepticism actually around multi-agent. I think right now a lot [00:02:00] of people already wrapped their minds around it and people are converging around the same idea. But, back in ‘23, I remember getting all sorts of pushback saying like, “Well, these LLMs are gonna hallucinate because that’s just inherently to the way that they work.

And if one LLM hallucinate 50% of time, like six LLMs are gonna hallucinate 95% of the time if you compound that.” And I remember I, I definitely got into trouble a little bit, like trying to answer that questions a few times over. But honestly, for me, in the back of the day, this is October ‘23, I start doing the code around July, give or take, and I was trying to create agents for myself.

That’s kinda like what I was trying to do. And I realized that if you think about those models, right, they are not very different from regular machine learning models, right? Mm-hmm. If you have any sorts of like regular models for doing weather forecasting, basically staking a few features, data that you know, and trying to come up with a prediction, the data that you don’t know.

And these LLMs are doing the same thing in, in very broad terms. The thing [00:03:00] is that the features, the data that they know, is kinda like everything that you have typed so far, and they’re trying to figure out what token comes next. Mm-hmm. So there is a direct correlation of like the datas that you– the, the tokens that you put in and the tokens that are gonna get out, hence why- Mm-hmm

prompting became so effective. Um, so if you do prompting by getting these agents to role play, usually you get better results. So the whole thing about multi-agents for me from the get-go was basically doing the context engineering so you have individual agents that are role-playing as different personas, and by that you’re tapping into the best space on the model to extract the most effective tokens for whatever you’re trying to do, right?

So if it’s a, an agent by, like passing by an engineer, you’re gonna tap into kinda like that engineering knowledge pa- space of the model that you actually can extract those tokens in a way that is more effective. So that’s how the multi-agent thing came together, was literally trying to get those models to role-play in effective manner.

Luke: Awesome. There’s no better test [00:04:00] subject than yourself with this stuff too, right? Like, it’s so cool that you’re kind of using this thing to solve a problem that you yourself were trying to deal with on your own. When we think about everyday life, right?

H-how would this concept of the crew and, this, group of agents, work for, my neighbor, right? Is there any examples of that yet or are we still too early for that?

Joao: No, there’s definitely some of that. I think especially with now things like OpenClaw, right?

I think a lot of people- Mm-hmm … are getting to experiment and experience this on the day-to-day. I would say for us, commercially at least, we’re more focused on larger enterprises. So we do a lot of businesses with companies like Johnson & Johnson or PepsiCo or AB InBev or Experian or even some government agencies.

So it, it’s funny because I think people can correlate with a lot of those businesses because those are businesses and brands that you know, right? Mm-hmm. And they’re not like tech-first companies, companies that are kinda like, they have these huge engineering teams. But they are savvy, and they’re trying to understand how to kinda like get value from AI, right?

They’re the ones that were promised [00:05:00] that would get all this value from AI. It’s like the Fortune 493, right? You remove the bank seven and, like, that, that cluster of companies. So we see- Right … a lot of all sorts of use cases. Think about back office automation, procurement automation. There’s some code generation automation as well.

So there’s a little bit of everything. But I think, a lot of people using agents on their day-to-day now, either through OpenClaw or Hermes agents or whatever it might be, or coworking or CloudCode, they can already get a, kinda like a look and feel. Now, imagine you being able to kinda like basically have this to become personalities and roles that are self-evolving, self-learning.

They can work together with each other, and you can bring them to work every day with you, and that’s kinda like what having your crew kinda like fused looks like.

Luke: The Brave Technologist is brought to you by the Brave Search API. Access billions of indexed web results from a simple API call with the Brave Search API.

Join the leading names in AI and tech using the Brave Search API to power agenic search, [00:06:00] keep LLMs current with real-time data, train foundational models, and bring the best of the web directly to the leading edge. Get started today at brave.com/api. So kind of like specialists each one of those examples basically, would be a different type of agent that has different types of, access or authority to do things.

Is that kinda how it works?

Joao: Yeah. It’s gonna be an element of access. There’s gonna be an element of control. There’s gonna be an element of skills. There’s gonna be an element of tools. And I think probably the most exciting is this idea of, self-learning and self-evolving, right? The more that you interact with them, the more they start to understand not only you, but kinda like your word model a little bit.

Kinda like what is happening, what are the people that work with you, what are your peers, what are your company, and all that. So yeah, there, there’s a lot of that going on. I think, like, on an individual level, you can have one single agent that’s helping you on the coding side, and on the other side you could have a dedicated team with a designer agent, a coder agent, a QA agent, a product agent, and [00:07:00] everything together.

Luke: That’s awesome. When we think about learning and obviously part of the learning process is mistakes that happen, right? There’s all this complexity, right? Delegating agents delegating to agents, automated decisions and and all this happening at scale.

How do you think about, accountability and auditability when, more of the human part is removed from the loop?

Joao: Yeah, I gotta say, that’s such a great point. I think, honestly, for most of the business out there, there needs to be someone accountable still, right? You cannot say- Mm-hmm

“Well, it was my agent, sorry,” right? Right. There’s gonna be someone that kinda, is invested and accountable for, if something goes wrong. We see a lot of, internal use cases. Usually y- they have a little less supervision. You still have some sort of human in the loop and all that. But a lot of the customer and kinda like the, outward-facing kinda like use cases, then that would definitely have some sort of human in the loop and kinda like more of a gating capacity or something like that.

But I also seen this idea- Policies are starting to [00:08:00] emerge where, yes, like we can give in more kinda like a freedom and, and more federation around building and running agents across the company, but we gotta have a way to enforce policies that are gonna be applied broadly across everything that is done.

So we are seeing some of that actually take place now, which is very exciting. I think, we’re still very early, but it’s great to see kinda like the industry now getting into kinda like what I would say the second order of magnitude problems where, all right, we have a few agents now, we’re seeing the benefit from that.

We need to make sure that we’re controlling them in a way that is effective.

Luke: That makes sense, like you said, it’s early and, I’m sure we’re still trying to kinda figure out too, like where the human needs to be in the loop and what the loop looks like, right?

Like necessarily for some of these things, right? Until we have, a lot more fit with this, right? But also I wonder what your take on this too, do you see agents maybe even helping as part of that accountability process where they’re like almost like I think you mentioned QA agent earlier, almost like [00:09:00] an, AI watching or an agent watching an agent kind of a role or type of function.

Is that something that you’re seeing companies doing? Is that something that just makes sense even from your point of view or w- what’s your take on that? I gotta

Joao: say it, it does. Beyond companies, I can even talk from our own experience at Crew AI. We definitely have a situation where we have agents writing code and agents then approving code, and then agents kinda doing debugging.

So you can find yourself or you can have a feature that kinda like an agent debug, another agent wrote the code, another agent review. We still gate a human review before anything goes into production internally ourselves, but definitely makes our lives way, way, way easier just to kinda like jump into that and knowing how to look and know that a lot of kinda like the simple things that I might have s- passed by was actually captured by another agent, so that whatever findings we’re gonna have is not gonna be any of the silly ones that you would expect some of the AI to make out of the gate.

So definitely think there is an element of using what people call LLM as a judge, right? Either LLM- Hmm … as a judge [00:10:00] in term of quality or LLM as a judge in terms of efficiency or in terms of adding guardrails. I think there’s another, there is a set of models even that are being fine-tuned specifically to be operating as judges.

I think Llama launched something called Llama Fire, basically to look for PII information and kinda like agent behavior and all that. So a- again, I think, I think it’s very early, so there’s a lot of techniques that people are exploring, but I think certain patterns are start to establish where people understand, like, all right- This actually works.

And I think this idea of LLM as a judge is definitely one of them.

Luke: It’s great to hear that you’re even seeing people starting to take, approaches around kind of refinement on this and approaching the models from that way too. And it brings me to the next question too.

When you see this next evolution of AI, do you think the big shift is toward better models or better coordination between- Mm … the models and the tools and the humans, or both? Or where are you thinking, for the road ahead?

Joao: I gotta say, that’s the beauty of it.

I think there’s so much upside in a bunch of different [00:11:00] fronts. So I gotta say, one of my frustration with the industry as a broader thing is that w- we haven’t seen fine-tuning become a thing at, at scale, mm, what honestly is unfortunate. I would bet like two years ago I was saying like, “Well, fine-tune’s gonna be huge,” but it, it didn’t pan out that way.

There’s definitely a few people doing it, but not at the scale that I was expecting. So I think there’s a lot of upside. I think people didn’t just realize that yet. But as we’re seeing kind of like in the news right now, everyone’s running out of their token budget for the year in like three months, I think like everyone’s now starting to kind of like, “Oh, maybe we should look at open source models.

Maybe we should look at how to fine-tune these models.” And I think there’s a lot of potential in there. There’s a lot of things that we could do in there that I think would be extremely relevant for a bunch of these customers. So that definitely for sure has potential. I think in terms of orchestration, there’s always new papers coming up, always new ideas that I think are proving to be interested.

I think there is definitely a point of diminishing returns where y- you gotta do certain [00:12:00] trade-offs, right? And it’s gonna depend a lot on the use cases. So you gotta choose how much determinism you want. You gotta choose how much self-learning and self-improvement you want. It’s one of those situations where you cannot have your cake and eat it too.

So depending on the use case, you might optimize for one or you might optimize for the other. But across the board, I definitely see potential to do improvement in so many areas. , It is insane. One that I’m particular interest is like fine-tune specific agentic models. I think that actually is remarkably interesting

Luke: Awesome.

Yeah, I think you’re right on the money too. People’s predictions from, even I feel like from two months ago are already completely different now. Yes. Stuff’s moving so fast, and it’s one of those things where I can’t really think of another instance, where you’ve had such top-down pressure, from, the Fortune 2000 C-teams down onto the rest of their team to start integrating this stuff.

There’s a whole new level of experimentation and unpredictability and, predictability too, like, where, these things are all kind of meeting the road at the same time, so it’s super interesting. When we think about agents and, I know we at Brave, we’re thinking [00:13:00] about this a lot, about, isolation, from certain parts of your online, life and what do you actually give agents access to, right?

How do you all look at that at Crew AI about, what AI gets the agents get access to as far as your data goes, and what don’t they, and what do they get authority to do? Who gets to decide that?

Joao: Yeah. So honestly, this is something that we spend a lot of time thinking. I think some of these problems are solved, but some of these problems are new. I think like things like OAuth, right, have existed for ages. I think they’re pretty good at like making sure that you give the right access of scope. I think especially if you’re associating an agent, someone else’s kinda like OAuth, that also allow you to change that agent access in runtime.

So like, hey, if it’s you, Luke, that is running it, it’s gonna have your access, your email, your credentials. If I’m running that same agent, it’s gonna be my access and my credentials. So there’s a few things in there that you can use that actually are pretty interesting, but there’s definitely kinda like a few different schools of thought.

There’s a lot of people thinking like, “Well, maybe these agents should [00:14:00] have their own identity,” right? “And they should have their own access and their own email and their own stuff.” What I actually think that is, uh, it, it is a valid way of thinking. I think it depends a lot on the company and kind of like the use cases as well.

So right now we support both. Like, we have agents that have their own email and that have their own sandboxes, so they can use that as if it was a computer to do whatever code they want to, and we also have agents that are operating through OAuth for individual users. I think people are very eager to bundle a bunch of like agentic applications and agentic systems as agents into one bucket, but there’s a lot of different flavors.

I think a lot of people are getting familiar with these conversational agents that are pretty good at ad hoc work, kinda like work where you care about the output, but the process is kinda like disposable, right? Like give me an amazing presentation or a spreadsheet. You care about it being good, you don’t care about how it gets there, right?

As long as it- Right … like looks good. But then there’s like completely opposite. There’s like these embedded agentic systems that are running automatically and [00:15:00] taking decisions and taking actions and that is like vastly different. Those you really care about the process to the point that you might wanna actually do a lot of deterministic behavior at some stages.

So again, there is a lot going on in there, but it’s fairly interesting

Luke: Yeah, definitely. Like six months ago, people were looking at this completely differently. It’s cool to see something like OpenCloud make it click for a lot of people all of a sudden when they start to see- Yeah

oh yeah, this is how agents can work from the system, right? And not necessarily like everything’s gotta happen through my chatbot or something like that. And I think as more and more people are getting, familiar with this, it’s becoming more and more clear for people too. But I think like when we look at like openness and safety,

Do you think that the move towards autonomous agents is at odds with open decentralized web principles or does it enable it

Joao: I think it depends a lot on the players, right? That’s the main thing. Mm-hmm. I think like there’s, one, it’s very noisy out there.

I bet that everyone that is listening to this, they are probably a little sick of [00:16:00] seeing the a- word agent out there, right? And if you are especially a leader, or like, and you have any buying power in an organization, I assure that you’re getting calls left and right, and everyone is selling you agents, right?

I think like you’re gonna have site builders selling you agents, you’re gonna have your CRM selling your agents, you’re gonna have your SAP selling you agents, you’re gonna have startups selling you a… It’s insane, right? So if you think about- Mm-hmm … like this is so noisy even for us that are kind of like in the forefront and have kind of like a good grip in the market, imagine really trying to be a leader into one company and trying to understand how to grab this bull by the horns, right?

And like understand, hey, how I actually use this thing. I think it is very, very confusing right now, but there’s a few things, a few principles of engineering that have brought us to this point that they think, that I think they still remain true, right? And I think this idea of like betting on open instead of closed, I think this is something that a lot of people are doing it [00:17:00] now.

So a good example is many of the CIOs into these large enterprises that we work with, they have been part of the on-prem to cloud migration. And honestly, , many of them have a bad taste in their mouth, because they, there were promises that everything would be unicorn and rainbows. And now, like they have these managed services bills that keep going up and up and up, and they’re trying to understand how they negotiate it, right?

It’s not they wanna migrate, but they need some sort of credible track. So a lot of these companies now are using Kubernetes, so they can make a point that they can move anywhere, right? And I think the CIOs now, they’re being way more smart about how fast Like this industry is evolving and the fact that they don’t feel like they’re gonna pick up winners just yet.

So I’m seeing they actually double down on idea of like, “I’m actually gonna use open source tools. I’m actually gonna use… Even if I’m paying for something, it needs to be something that’s extremely open and I’m not gonna get a vendor locked into.” So I think that has been interesting, [00:18:00] and then that also gets intertwined with something that you folks are very familiar with around the data, right?

And like how I make sure that whatever data that I’m using is not exposing me to extra risk, is data that I have control, is data that I know what’s coming from, and there’s an element of there that data is also becoming a major thing. I think everyone understands that as building gets commoditized more and more, data is getting a little bit of the spotlight, and data has a lot of gravity.

Whenever you have the data, applications will gradually start spinning around and close to it.

Luke: I think you’re totally right on, on both of those points, too. Especially when you start to see, some of the security side too, right? A lot of, vulnerabilities floating around and dependencies that are able to be caught with open source software,

And once word travels and that type of thing that you don’t really get, when things are fully closed off. Yeah, that’s been one thing that’s been really interesting to see about the AI space is, even if we went to early on with Hugging Face and a lot of these, open source models too, just like how much that’s actually had a big [00:19:00] deal and a big contribution, and it’s great.

If we can move everybody to open source, I think it’s just super helpful, especially with how powerful the technology is too. To you as a leader and a CEO, right? What part of your personality do you think helps you the most as a founder?

And have you ever had to dial it back because it was working against you?

Joao: Damn, that is a hard question. Well, I gotta say, yes, there’s parts of, me as a being that I think give me some sort of unfair advantage, and I think everyone has their own. It’s all about you being able to align that to the company and how you run the company.

I think for me, I always had been very much out there. I think that’s the best way to put it. So I have no problems with podcasts, speaking, classes, whatever it might be, and that was such a helpful thing from the early, early days in the company. Mm-hmm. And I look at the industry and a lot of the way that I, how I run the company sometimes is I wanna do what is [00:20:00] best for the company And I also, when I’m looking at the market, I wanna do what people are not doing.

So whenever I notice the pattern where, a lot of people in the industry is doing something, I almost like to have this urge in me where I wanna do the exact opposite. For whatever reason, I wanna hedge against that, and I wanna bet on, what we are building. So I noticed early in the company that all these AI companies, they didn’t have a face, right?

Hmm. These companies are always like, “Yes, you, you know the company,” but you don’t think about the founder necessarily when you’re thinking about them, or in the big labs and all that. So I made a point that I wanted to be intertwined with Crew AI. Mm-hmm. So I put myself in- out there as much as I could, and I think that was vastly beneficial for the company.

Allow us to close a lot of the early customers, people would look for us to advise. And a lot of times, because I was putting myself out there so much, if I say, “Hey, I’d love to take a meeting with your CIO,” the CIOs would actually take the meeting because “Oh, I watch a video from this guy.

I actually think that he has something to say.” So that did create a lot of [00:21:00] incentive, in the company. But that also got into a stage where it started hurting us a little bit, where the company cannot scale if I’m in the one-to-one, put my face out there. Right. Sure. So, like, we need to make sure that we have a marketing department, and not only that, but everyone in the team is playing marketing, right?

That was something that we had to change in the company where everyone was expecting me to be always be the one doing the launches and the promotion and the videos, and we had to grow out of that a little bit. So that is definitely has been something that has been, helpful to me and also something that we had to adapt as we grow.

Luke: I can imagine, it’s gotta be pretty tough to navigate as a CEO where you’ve got, this new technology that you know is going places, there’s all this pressure from the, market side of things, for implementing these things, so your customers are definitely feeling the pressure there.

At the same time, it’s changing so fast. It’s almost like fashion, but faster, right? Where you’re like, “Okay, now this season we’re talking agentic browsers. Now it’s OpenClaw. Now it’s these models. [00:22:00] Now it’s these ones.” Very much moving around but at the same time, to your previous point, there is something really great and authentic about having a founder that’s a human out there in this AI space where , the question of what’s real is kind of getting challenged a lot, right?

Yeah. So I think it’s, it’s gotta be tough to navigate. I can only imagine.

Joao: It definitely is. I wanna say that people ask this, to me sometimes. I wanna say that the challenges for the founder and CEO, they, they haven’t changed necessarily. The timeline just got extremely compressed.

Mm. So if you create a new product and, you could milk that for two years until you, had to create something new, now good luck getting eight weeks, right? So there’s a lot of that, and then it’s hard because that basically means that you’re in a constant sprint- Mm-hmm

and pushing your team to be in this constant sprint. At the same time, that you cannot stop running because you gotta keep innovating as much as you can, but you gotta keep finding– you, you gotta [00:23:00] find a way, right? You keep finding, trying to find some sort of advantage or shortcut or moat, whatever it might be that people are calling that can, can– they can get you a gap on that race, right?

Mm-hmm. You don’t need to pace yourself at that fast of, a pace, with the entire company going at 100 miles per hour non-stop. So there’s definitely some of that that is going on, but the challenges, I still think they’re mostly the same. Yes, building is getting commoditized, and that does means that other things get a little more of the waves distribution for, other things change.

So data is becoming way, way more important. Distribution’s becoming way, way more important. So all those things are still real. But in the end of the day, I think for the founders and CEO that might be listening to this or people that are thinking about starting your company, I would worry less about competition.

Yes, competition is important. You wanna track them. You wanna track the market. I would care about what is the thing that you think that you can do that only you can do, why you’re the only person that can do it and it’s gonna be [00:24:00] exceptional. And then you go out there and kill everyone else that is doing it.

Just like go for it, right? That’s awesome. You gotta have that. You gotta be in you, right? That’s the other thing. It’s fierce competition, you gotta be competitive, for sure.

Luke: Are there any, environments or kinda conditions that bring out your best thinking and how intentional are you, about protecting that space, as a CEO?

Joao: You got all the good questions there. Sorry, man. I, I … This is interesting,

Luke: though. More interviews I have with, AI CEOs, the more I wanna ask them these kinds of questions just because it’s just like you said, the competition’s fierce and, Yeah

We’re providing picks and shovels, right? But you guys are out there, really having to be on your toes, you know?

Joao: Yeah, I definitely strive in competitive environments, but I, usually they’re high-stress. Mm-hmm. So for example, I didn’t have this many gray hair when I start the company, for sure.

But again, it’s what I signed up for. Zero regrets. I love w- a lot of what I’m doing. But I definitely do a lot of great, great thinking when I’m very competitive. Mm-hmm. What turns out to be usually kinda it, it [00:25:00] can get dark for a hot second.

You’re like, “Oh, shit.” Like, “Damn, this is scrub, this is competitive and all that.” And then I start to make my way out of that, and I start to thinking about and you get more clarity, and then boom, you stumble into something. Like, all right, I think there’s something here, and then you start poking it. I think, the main thing, though, you just gotta focus, right?

That’s the main thing that I have learned. I wish we had, more focus early on in the company. I think we spread ourselves very thin, but we learned our lesson, and I think, right now focus is the name of the game. If anything, we are doing way, way, way better focus right now. And you gotta have some sort of conviction, right?

At the end of the day, if you’re running a startup, the one thing you have going your way is you could take big shots, right? Mm-hmm. And if they go south, people won’t necessarily care. Like, Microsoft cannot go out there and kinda, take a huge shot out of nowhere, right? They might take maybe- Right … one of those every two or three years.

But as a company, you can do it. You might not have the resources to do that many, but you can, and you can move that fast. [00:26:00] So I would say there gotta be an element of conviction where like, all right, this is what we’re gonna do, and like- You gotta go. You gotta– I think, especially on how competitive it is, I think you gotta bring that energy of “I’m gonna take my short or I’m gonna die trying,” and you just go for it.

That’s

Luke: awesome. No, I love it. I love it, man. It’s such a good, approach, too. The odds are so, so stacked, especially now, and I think that conviction is kind of what makes you not feel the same to everybody, too, when everything is starting to get more and more automated and feel the same, you know?

Do you have a belief today, that you probably would’ve argued against 10 years ago?

Joao: Pfft. 10 years ago, damn. I might have, for, like, five, six months ago. Go, go five or f-

Luke: five months ago, five years ago. Anything. All right. So

Joao: if I have a belief today that years ago I wouldn’t– Well, I mean, honestly, like, I, I started doing AI in 2017.

Um- Mm-hmm … I didn’t see LLM coming the way [00:27:00] that they did. I- Mm-hmm. That is definitely something that I didn’t see coming. Um, another thing, as LLM started, I was definitely in the gr- As LLM start- LLM started to get good at coding, I was definitely on the team that would think that this would be a very kinda it would be a more comparative effort between agents and humans in terms of, like, especially on the coding- Mm-hmm

Where there would be still a lot of human supervision. I’m now more bullish of the opposite. I think there’s gonna be a lot of, agents driving a lot of code on themselves. Interesting … I think humans will be still in review. But I think, the idea for– It’s Cursor versus Cloud Code. I mean, they’re– Like, Cursor was the thing, right?

I think a year ago. Everyone was using it. I think everyone mental model now, at least on the, at least on the Bay Area, like, bubble a little bit, has shifted into, Cloud Code, where you don’t- Mm-hmm … necessarily see all the code, and you gotta be, you gotta be more mindful about, “Oh, I actually wanna see the code that is written,” right?

Mm-hmm. Mm-hmm. I think that… Yes, a year ago, I would say [00:28:00] no way that people are just gonna delegate to agents and not even look at the code, and now I believe that, yeah, there’s gonna be a lot of people doing that. I haven’t made my peace yet if that is the right decision to do or the wrong decision to do, but I definitely have been challenging things where, for example, I thought about, well, agents are good at generating code, but people are gonna be maintaining it, so we better take good care of that.

Now I’m seeing- Mm-hmm … agents maintaining code, so you start to, you start to challenge everything else. You start to say, like, “Well, should we even be doing reviews, like, if the agent’s gonna be maintaining anyway,” right? Again, this is more of the forward-thinking stuff, right? It’s not to say that- Yeah.

Yeah, totally … you wanna go there. But I do find it interesting that, it’s making me challenge a lot of the principles that I have grown my career the last 20 years to believe in.

Luke: That’s one of the reasons why I wanted to ask the question in particular too, ‘cause, you know, startup founders have to kind of be adaptive, right?

And if you’re not changing up at least parts of your approach, you’re not adapting to what’s out there. So it’s super interesting. Assuming we kind of get the next decade of AI [00:29:00] right, w- what improves most in the everyday life of ordinary people from your point of view?

Joao: Well, I mean, I definitely– I’m getting a little bit… Again, to your point is I think as a CEO of a company that is doing agents, I get to be a very, very early adopter. I get to experience what I think like people experience like later, depending on where you are on the early adopter stage, might be years later.

I definitely think that having agents to help me on my day-to-day has been life-changing beyond work. I had some of those incredible like agentic experiences where my wife and I were planning a trip, and I just came to an agent, it’s like, “Hey, I’m planning a trip. Find somewhere for us to have our dog stay while we travel.”

And the agent decided to call the places. It called the places and came back with a document into a Google spreadsheet, and then I later got to watch and hear the call, and it was insane. Because it not only called and the people talk with [00:30:00] this agent, but he actually negotiated on my behalf and asked for a discount.

And I was like- That’s wild … “Man, like that, that is wild.” You don’t have success stories that work like just great like this every other day. Mm-hmm. But I think when you get a glimpse into this, you start to see what the future might look like, where you just delegate something to your agent, and I don’t think it’s just gonna make the decisions for you just yet, but it can bring like the digest.

Like, “Hey, here’s option A, B, and C. Here’s the pros, the cons, the cost, how much you’re able to save and everything,” and you get to make a conscious choice. I think some of that has been exciting. On the company, we’re also using that as a playground for trying like a bunch of cutting-edge stuff. So we actually have, many people don’t know this, but we are, we are working on a Create AI 2.0.

It’s gonna be a huge breaking change from 1.0, unfortunately, but it’s gonna be an amazing experience. And we are actually using it internally, right? So we have the few agents like on our Slack that is already running on this, and it is remarkable. And now [00:31:00] we have this one agent that has been Night and day for the company.

Last week, it actually did 42% of all the PRs in the company came from that agent. Wow. And we are now, we’re now considering creator first. So we work with a pod team structure in the engineering. Mm-hmm. We’re now trying to experiment with this very cutting-edge idea of, could we have an entire pod?

This is our agent tech pod. Wow. We put a product person, a designer person, an engineering person, like what will come from that? I, I don’t know, but we get the chance to kind of like use that as a playground, kind of like labs, and then we bring whatever learnings into the major framework.

Luke: So interesting.

There’s just these stories, like these parts of these stories like, the agent called on my behalf to all these- … places for me. Like these things where they just are different types of engagements or, having a pod, of agents.

It kind of opens your mind up to it from a different angle. It’s really interesting. Well, I mean, Jo, I really appreciate you making the time. It’s been a great conversation. Where can [00:32:00] people find out what you’re up to or anything- you’re putting out there, if they wanna go check it out after they hear this?

Joao: I mean, I am very good at shitposting on X, so if you wanna- … join into that, you might- Love it … wanna check it out. I go by, J-O-A-O-M-D-M-O-U-R-A. So it’s a mouthful, but it’s pretty sweet. We post a lot about the company, the progress in there. And I also do some LinkedIn if that is your cup of tea.

You can also find this me with the same, the same handle, but that would probably the easiest way, or follow Crew AI as well in both social networks.

Luke: Perfect. Perfect, man. Well, thank you again, Jo. I really appreciate you making the time. Love to have you back, too, to kinda check back in on how things are going in the future.

And, yeah, best of luck with everything, man. Thanks for coming by.

Joao: Thank you so much. Excited. I’d love to come back, and thank you for today, Luke, and thank you, everyone. I hope you enjoyed this. Catch you around. Excellent.

Luke: Thanks for listening to the Brave Technologist Podcast. To never miss an episode, make sure you hit follow in your podcast app.

If you haven’t already made the switch to the Brave browser, you can download it for free today at [00:33:00] brave.com and start using Brave Search, which enables you to search the web privately. Brave also shields you from the ads, trackers, and other creepy stuff following you across the web.