Person
1 episodes · 22 quoted claims
Moura says a person from Oracle told him that Oracle was using CrewAI in production when CrewAI was still only an open-source project.
“I remember this person approaching me and saying, like, Hey, we are Oracle, we're using we're using CrewAI in production.”
Moura says more than 50 million agents ran on CrewAI in January alone.
“And January alone was, uh, over 50 million agents. So that was, yeah, that was insane.”
Moura says the largest crew he remembers seeing on CrewAI had 21 agents.
“I remember seeing at once a like a crew that had 21 agents on it, and that, for me, was was impressive.”
Moura says a media company uses CrewAI agents to automatically edit live game footage, including cutting, captions and sound, for social media.
“when we see kind of like a media company using CrewAI agents to automatically edit files.”
Moura says a Fortune 500 customer that began with simple use cases now uses agents to adjust prices in marketplaces and monitor competitors.
“and now they have a lot of their pricing flows automated. So they adjust prices using agents on the runtime, uh, in marketplaces and other spots”
Moura says Gartner uses the term 'agent washing' for companies that label existing products as agentic.
“we're talking with some folks from, uh, Gartner, and I think they're referring to this as kind of like agent washing, like, um, companies that are stamping this is agentic now”
Moura says some companies, including LinkedIn, are experimenting with seat-based pricing for agents.
“I mean, we're seeing some companies experimenting with that. I think LinkedIn is doing something, uh, in that in that lines.”
Moura says a customer's IRS tax form automation involved a roughly 60-page form with a 720-page instruction manual.
“it was a form that was around 60 pages long, but fear not, it comes with instruction manual, and the instruction has 720 pages long.”
Moura says CrewAI lets users set a hard time limit, such as 60 seconds, after which the agent is asked for its best answer.
“You can say like, Hey, you can only run for like 60 seconds. And then what happens is as it gets to that 60 seconds, we do a final call saying like, Hey, all right, you're done, your best answer right now.”
Moura says an agent deployed on the CrewAI platform automatically becomes a production-grade API with load balancing, auto scaling and SSL.
“That automatically becomes an API, and that's production graded. So I'm talking, uh, load balancer, auto scaling, SSL, like all those different things.”
Moura says CrewAI agents have short-term, long-term and entity memory, with a fourth user memory type added recently.
“we do have all the agents in CrewAI, open source in an enterprise. They have short term memory, long term memory, and entity memory.”
Moura says CrewAI's short-term memory is a shared space for agents that is cleared after every run.
“But that gets reset in every run. So you do a run and then boom, everything gets cleaned up and a new run happens.”
Moura says CrewAI's long-term memory stores learnings across multiple runs, using the expected output defined for each task as a comparison point.
“Now, the long term memory is where they store learnings that are coming from multiple executions.”
Moura says CrewAI removed its LangChain dependency after it was overriding about 90 percent of one imported class, which made version updates require heavy rebasing.
“So it got to a point in time where we were overriding like 90 of the code that we were importing, and that was one specific piece, the class that we were importing.”
Moura says the company Composio offers over 300 tools that can be used with CrewAI.
“I think they have like over 300 tools that you can also use with CrewAI.”
Moura says finance and insurance companies often self-host language models so that no data leaves their environment.
“finance and insurances, they usually want to self-host their models because they wanna run the whole thing in kind of like a airtight container”
Moura says reasoning models excel at math because problems with clear right answers suit reinforcement learning, unlike nuanced business decisions.
“what we're seeing these models like really excel, especially cognition models, is on the math problems is because they have a hard, right answer, right? So they can do reinforcement learning based on that”
Moura says Crew Studio starts as a chat interface, then becomes a tabular view of agents and tasks, and finally a node view.
“we have a what I call a gradually evolving UI. So you start with a chat interface, similar to, uh, custom GPTs back in the day.”
Moura says Cursor supports custom per-project rules that are embedded into prompts, and that his team uses them.
“is that you can add custom cursor rules, and I think people don't realize that, and you can do that per project.”
Moura says CrewAI uses its own agent crews to review every pull request across its open-source and closed-source code.
“we have crews reviewing every single pull request across the company, open source and closed source.”
Moura says CrewAI runs agents on new sign-ups to research them, then uses the results to tailor CRM outreach and customize the product.
“So as these customers onboard the platform and they put their names and emails, we kick off agents that research them”
Moura predicts that small, highly capable models will trend in 2025, outperforming some current 70-billion-parameter models, and that fine-tuning will make a comeback.
“I think that's gonna be a trend this year where we're gonna have smaller models that are very, very capable, like better than like 70 billion parameters models nowadays.”