Doug Camplejohn
(00:01)
Hello everyone, this is Doug Camplejohn and in this week’s episode of Revenue Renegades, I’m excited to welcome the CEO and founder of Crew AI, Joe Moura. Joe, welcome to the show.
Joao (Joe) Moura
(00:12)
Hey there, thank you for having me. I’m excited to be here and to talk about agents.
Doug Camplejohn
(00:23)
I love founding stories. You were at Clearbit before and have a deep engineering background. What made you decide to take the leap and start your own company through AI?
Joao (Joe) Moura
(00:40)
I’m an engineer by trade and have been in engineering for 20 years. There have been many ups and downs along the way, but my curiosity always drove me to learn more. I was at Clearbit for five years until HubSpot’s acquisition. That was around the time ChatGPT took off and everyone started talking about it. Although no one was doing retrieval-augmented generation yet, we began experimenting with RAG pipelines and advanced embedding techniques. That experience inspired me with the idea of AI agents.
I started building AI agents on the side—even on my anniversary, despite my wife’s reluctance about my computer obsession. I would wake up early and work on these agents. They worked so well that after perfecting them, I realized I needed to develop more. That’s how our open‑source framework was born. Then at a Bay Area meetup, someone from Oracle said, “We’re actually using Crew AI in production. Can you help us?” That was a pivotal moment. If large companies were already using this in production, it wasn’t just an open‑source project—it was time to build a company.
Doug Camplejohn
(02:54)
It feels like a once‑in‑a‑lifetime period in technology and a unique opportunity for those building agents.
Doug Camplejohn
(03:04)
Great. Although our listeners have varying technical backgrounds, there is still some confusion about what actually constitutes an agent. Could you give us your definition?
Joao (Joe) Moura
(03:22)
Agents must have agency—that is, the AI must control the flow of an application. In traditional automation, everything is strongly typed: the input data, the transformation, and the output are predetermined. With modern LLMs, you can make ad hoc API calls that generate content, which is impressive but only agent‑assisted automation. The real innovation is when the model dynamically determines an application’s flow at runtime. In short, true agents require agency.
Doug Camplejohn
(04:34)
I love that definition. Without naming names, Mr. Benioff has clearly embraced the agent concept. I wouldn’t be surprised if he rebrands his company—launching products like Agent Force and Agent Exchange.
Doug Camplejohn
(04:53)
Additionally, Dharmesh at HubSpot is working with Agent.ai to build what he calls a “LinkedIn for agents.” In your view, which areas will agents impact the most initially?
Joao (Joe) Moura
(05:09)
At the outset, it’s evident that leading figures such as Benioff, Dharmesh, and Jensen believe this will become a trillion‑dollar industry. This implies two things. First, no single entity will dominate; every company will have its own spin‑off agents. Second, while there won’t be a one‑size‑fits‑all winner, some companies will win big. I believe Crew AI is well‑positioned in this space. I also see companies like Agent Force, ServiceNow, and SAP developing agents within their ecosystems to facilitate complex interactions.
Our strategy at Crew AI is to take a universal approach—our agents can communicate with those from Salesforce, ServiceNow, SAP, and more. I also look forward to insights from upcoming Gartner research that will educate executives on the evolving market. Overall, some companies are excelling, while others are still finding their footing.
Doug Camplejohn
(09:01)
I’ve always viewed this as a matter of global versus local optimization. While significant energy is spent on developing agent models that function solely within a single ecosystem—akin to hiring a digital worker with a limited pool—ultimately, adopting a universal approach where agents work broadly makes much more sense.
Joao (Joe) Moura
(09:40)
We are working directly with these teams. We’re engaging with ServiceNow, AgentForce, and SAP to achieve interoperability among agents, ensuring customers receive value. Ultimately, customers want results and efficient solutions.
Doug Camplejohn
(10:07)
When you assign work to an agent that has some level of autonomy, is there a specific syntax you consider? For example, how do you define the resources, conditions, and expected outputs?
Joao (Joe) Moura
(10:48)
There are additional considerations. Take agentic analytics, for example—agents operate atop a data lake, abstracting its complexity. They not only understand, in real time, the tables, schemas, and data types, but they can also write queries, execute and correct them, extract data, draw correlations, and even plot charts. This level of automation, once unimaginable within a reasonable timeframe, demonstrates the true power of AI agents.
However, this complexity introduces questions about scoping. What level of access should a user have when running an agent? Should permissions differ based on who triggers the agent? In consumer use cases, the entry barrier is lower; in enterprise settings, robust data protection and airtight security are crucial.
Doug Camplejohn
(12:46)
We focus a lot on sales use cases here on Revenue Renegades. I loved your point about analytics that can answer complex queries in real time—tasks that might otherwise take a revenue operations team days or weeks. What other sales use cases are emerging?
Joao (Joe) Moura
(13:15)
There are many. At Crew AI, we run around 150 agents, which is remarkable. Without these agents, we’d need a team three times larger. In sales, our agents handle various tasks—meeting preparation, for example. They research companies, key individuals, industries, and recent news. Beyond basic research, they integrate insights from our business to uncover relevant use cases, identify major relationships, and find similar companies in our databases. This equips account executives with comprehensive, timely intelligence—all in a fraction of the time traditional prep would take.
Joao (Joe) Moura
(14:41)
We also use this information for targeted marketing. For instance, if you, Doug, sign up as a lead and we infer specific use cases for your industry, your email campaigns become highly targeted—not only to your persona and company but also to your specific challenges. When you later engage with our product, demos and templates are tailored based on these insights.
Doug Camplejohn
(15:22)
That’s fascinating. Regarding the agents running internally at Crew AI, did you build them all yourself, or did you leverage a mix of internally developed agents and those built by others?
Joao (Joe) Moura
(15:35)
It’s a mix, but everything is developed internally. I built the meeting preparation agent myself because I needed help managing back‑to‑back meetings. Meanwhile, our sales and rev‑ops teams have developed agents to review NDAs and agreements—a necessity given the volume of enterprise NDAs.
For large companies, dealing with organizations that have over a hundred thousand employees means it’s often easier for them to use their own paperwork than for you to provide yours. This requires thorough NDA review. Instead of sending every NDA to lawyers, one of our sales engineers created agents that automatically review NDAs. If no red or yellow flags appear, the NDA is signed immediately.
Doug Camplejohn
(16:55)
How technical do people need to be to build agents on the Crew AI platform?
Joao (Joe) Moura
(17:02)
In the long term, our vision is for no‑code development to be the only option needed. For this industry to reach its full potential, it must be accessible to everyone. It’s similar to launching an e‑commerce store with Shopify—easy to start, but complex features might still require engineering support. Currently, whether we target a CIO/CDIO or a business unit, the most successful teams usually have a technical sponsor—one or two engineers partnering with our team to build the initial use cases while we empower them to do more.
Doug Camplejohn
(18:19)
How important do you see the marketplace aspect—where non‑technical users can assemble or purchase agents and related solutions? For instance, Dharmesh at Agent.ai has introduced a company research bot that seems very popular in the marketplace.
Joao (Joe) Moura
(18:58)
I believe a marketplace will eventually be essential; it’s simply a matter of timing. Consider custom GPTs from earlier days—OpenAI was well‑positioned, yet the concept didn’t take off. I expect a marketplace will emerge with the right audience at the right time.
For full disclosure, Dharmesh is one of our biggest angel investors, and I work closely with him. While his agent AI approach is heavily influenced by HubSpot’s model, I haven’t yet seen marketplaces become the primary adoption driver. I’m confident that will change once all preconditions are met.
Doug Camplejohn
(20:12)
A key topic in this agentic future is software pricing and its implications for labor replacement. In customer service, outcome‑based pricing models are emerging—charging per result. What do you see as the likely pricing models for these services, and is it appropriate to view them as replacing labor—costing a fraction compared to a customer service rep or an SDR?
Joao (Joe) Moura
(21:12)
This is a new industry—no one has been doing this for 20 years because it didn’t exist until recently. Companies are focusing on aligning their pricing as closely as possible with the value they deliver. However, many overlook the enterprise perspective: charging purely based on value or outcome can make budgeting unpredictable. Although paying only for what you get sounds smart, enterprise clients prefer predictable costs. I’ve seen cases where companies signed multimillion‑dollar contracts without utilizing the full capacity. In contrast, pay‑as‑you‑go or value‑based pricing, which allows for better forecasting and reporting, tends to work better with investors who require certainty in financial disclosures.
Doug Camplejohn
(23:09)
Outcome‑based pricing seems to work well for customer service, but I haven’t seen a successful model for sales yet. Have you?
Joao (Joe) Moura
(23:25)
I haven’t seen a robust sales pricing model yet, although some companies are experimenting—some even trying to replicate the role of a salesperson with revenue‑share models. Ultimately, experimentation is key. Companies must focus on delivering value and adjust their pricing incrementally. If you haven’t changed your price in a year, something might be amiss. I advise companies to consider the buyer’s perspective when setting prices.
Doug Camplejohn
(24:36)
I recall a call with you and Rob Bailey where his calendar was packed with 15‑ or 30‑minute meetings. How do you balance rapid growth and the many ongoing tasks? Based on your experience, what key principles do you follow as a CEO to stay focused on your goals?
Joao (Joe) Moura
(25:11)
Our approach mirrors that of many successful startups—we embrace a degree of chaos while adhering to a set of core principles. As long as decisions align with these principles, flexibility is allowed, and there’s no place for blame.
We also understand that what works one week might not work the next. As a growing startup transitioning to a growth stage, our key advantages are speed and the capacity for bold moves—qualities that larger companies often lack. We double down on our strengths; while we address weaknesses secondarily, our focus remains on what propels us forward. Often, a small signal in the midst of daily operations prompts us to explore a new idea with a small team, which eventually pays off. This agility is one of the benefits of being a lean startup, and I encourage others to adopt a similar approach.
Doug Camplejohn
(27:10)
Can you share the Crew AI principles?
Joao (Joe) Moura
(27:14)
Our core principles include:
• Move Fast and Be Bold: We take pride in rapid, decisive action without assigning blame. Constructive feedback is welcome as long as it comes from a genuine place of respect.
• Hold No Punches: We launch what we believe in without delay rather than waiting for perfection.
• Foster Community and Ownership: We communicate proactively, empower every team member by explaining our mission and methods, and embrace ownership to make quick, bold decisions.
These guidelines are part of our internal documentation and are shared with everyone who joins the company.
Doug Camplejohn
(28:31)
Given the rapid pace of change, what safeguards do you believe are necessary to prevent over‑dependence or unintended consequences in implementing these systems?
Joao (Joe) Moura
(28:52)
When considering governance, several layers must be addressed. At the foundational level, it’s essential to monitor logs, prompts, and system traces. At a higher level, one must ensure the quality of AI outputs—mitigating hallucinations and verifying tool reliability. Additionally, robust authentication, access control, and fingerprinting are critical for traceability.
There’s also the matter of data localization and privacy; data must be sanitized to remove personally identifiable information, especially for customers in regions with strict data residency requirements. Some organizations even require that data remain on local servers. These safeguards sometimes push companies back toward on‑premises solutions. In fact, we now offer Crew AI bundled with HPE, NVIDIA, and Dell—evidence of a trend toward reconsidering cloud dependence.
Doug Camplejohn
(31:06)
That’s remarkable. The pace of development is astonishing—it feels like it’s doubling every 18 days rather than every 18 months. Looking ahead 18 to 24 months, what do you envision for the future of agents and AI?
Joao (Joe) Moura
(31:38)
In 18 months, I expect Crew AI alone will support over a thousand agents. Progressive enterprises will likely deploy tens, or even hundreds of thousands of agents, resulting in substantial efficiency gains and fostering widespread innovation.
For example, we’re working with a major CPG company across three countries. In one pilot, they achieved an 80% efficiency gain by automating processes that formerly required costly data scientists. As these gains compound, companies won’t necessarily need to downsize; instead, they can reassign talent to more strategic roles and further fuel innovation. I believe automation will drive both adoption and continuous improvement.
Doug Camplejohn
(33:06)
Those are great examples. With rapid efficiency gains, organizations may need fewer employees. I recently heard discussions about a “one‑person billion‑dollar startup” concept. How do you see this model evolving in the enterprise space where companies focus on reducing headcount and achieving cost savings?
Joao (Joe) Moura
(34:01)
Ultimately, budgets are driven by personnel costs. Automation may not always lead to layoffs but can lead to restructuring—shifting employees from routine tasks to more strategic roles. For example, instead of hiring someone to review NDAs, an agent can perform that role. In some cases, this might mean hiring an engineer to innovate further rather than simply replacing a person. Company structures will evolve, and while headcounts in certain areas might decrease, the overall effect should be positive.
Doug Camplejohn
(35:07)
When considering your own growth, how do you evaluate when to hire a new person versus relying on agents to handle the task?
Joao (Joe) Moura
(35:33)
When a need arises, we ask ourselves three questions: First, can we build an agent to address it? If not, can we hire someone to build that agent? And finally, if all else fails, should we hire a person? We often leverage our open‑source community to develop agents when we reach capacity, which helps alleviate the workload.
Doug Camplejohn
(36:12)
This has been a fascinating conversation. You’re clearly on the leading edge of the agentic movement. I appreciate how you’ve separated the signal from the noise—especially when so many claim “everything is AI.” To wrap up, a quick rapid‑fire question: In your personal life, if you could wave a magic wand and have AI permanently take over one task, what would it be?
Joao (Joe) Moura
(37:05)
I wish AI could manage my overflowing emails and social media inboxes. Although tools exist for that, I’d love to see something built with Crew AI for a more tailored solution. On the other hand, one task I have automated—and which I’m very proud of—is coding. When someone opens a pull request for Crew AI, one agent writes the code, another reviews it, and then the first implements the revisions. Most pull requests are merged immediately, which has been incredibly valuable.
Doug Camplejohn
(38:02)
That’s amazing. What is something you’re passionate about that might surprise people?
Joao (Joe) Moura
(38:10)
Surprisingly, I’m passionate about Japanese anime. Most people wouldn’t expect that from me.
Doug Camplejohn
(38:37)
Japan is one of my favorite countries, and I share your fascination with its culture. In your personal routine, what is one thing you cannot live without?
Joao (Joe) Moura
(38:57)
I always carry my computer and iPad with me—even though I once said an iPad wasn’t for me, I now rely on it daily. I also use an Ergodox keyboard—a split keyboard designed to prevent wrist strain after long coding sessions. I can’t imagine working without it; although it’s eight years old and needs renewal, it still works like a charm.
Joao (Joe) Moura
(39:54)
I can’t imagine being without them.
Doug Camplejohn
(40:02)
That’s wonderful. Finally, how can listeners keep in touch with you and support the mission of Crew AI?
Joao (Joe) Moura
(40:14)
Please check us out on Axe. Visit Axe.com/J-O-A-O-M-D-M-O-U-R-A for my personal page where we build publicly. I ask and answer questions there. You can also connect with me on LinkedIn using the handle J-O-A-O-M-D-M-O-U-R-A. Alternatively, visit our website to learn more about our team.
Doug Camplejohn
(41:02)
Thank you, Joe. I really appreciate your time and insights. Have a great day.
Joao (Joe) Moura
(41:09)
Thank you for having me, Doug. Thank you, everyone. Have an amazing day.
