Inside Ramp, the $32B Company Where AI Agents Run Everything | Geoff Charles

**Geoff Charles** (0:00)
If you're not using Claude Code this year, no matter what your role is, you're probably underperforming compared to others on the company. PMs often pride themselves on the spec, the perfect spec. They have to understand that it's actually AI that's reading the spec now versus engineers. 50% of RAMPS code is built by AI, and that's 50% up from 30% in December. It'll probably be 80% by March.

**Peter Yang** (0:19)
And this is not just like a front-end prototype, right?

**Geoff Charles** (0:21)
This is the real product, back-end, front-end, and I have a PR, and I can just submit it to the engineer team. PMs are shipping tons using inspect. And so are designers, so are operators, so are like account managers and salespeople are also getting activated. My job is to automate my job, and all our jobs is to automate our jobs.

**Peter Yang** (0:40)
All right, everyone, my guest today is Geoff, a CPU of Ramp. And Ramp is one of the fastest growing companies ever, and probably the most AI-native company that I know outside of the big labs. So last year, Geoff and team shipped over 500 features and hit over a billion dollars in revenue, all with around 25 P.F.s. So yeah, really excited to talk to Geoff today. And welcome, Geoff.

**Geoff Charles** (1:03)
Super excited to be here. Thanks for having me, Peter.

**Peter Yang** (1:05)
Awesome, man. So, you know, I've worked at a lot of big tech companies, but like, can you give us a quick overview of how Ramp ships features, like from idea to launch?

**Geoff Charles** (1:14)
Yeah, I'll skip the basics and just jump into the fact that it's a crazy time right now. And the way that we are building has always been around velocity and the way that you move fast is by leveraging tools, and AI is just an incredible accelerant to everything that we do. And I hope during this call that I'll be able to share a few of the ways that we've leveraged AI to accelerate, to inspire folks and help amplify the learnings. I also expect that a lot of the things that we're going to talk about today are going to be outdated, even by the time that you even share this recording. So I'm excited for it. But yeah, I mean, the product development process hasn't dramatically changed in terms of root principles, right? It's about understanding customer pain point, about identifying the right solution, about building the solution, and then testing and iterating. And I think AI just lowered the cost of each of these sections dramatically. The cost of code is basically down to almost zero, apart from the tokens. And so PMs just need to be actually writing the specs for the agents rather than the engineers themselves. And I think that's a complete shift in terms of how we go about it.

**Peter Yang** (2:29)
Yeah. So basically the PMs will make the product first pretty much by themselves, right? Or make the prototype at least, and get some validation before doing anything else. Yeah.

**Geoff Charles** (2:40)
PMs often pride themselves on the spec, the perfect spec, and they have to understand that it's actually AI that's reading the spec now versus engineers. And so the spec itself is basically the output of a prompt, and then the output of the spec is the product. So at the end of the day, it's just prompt to product, back to prompt, back to product. And yeah, we are essentially collaborating on an actual product itself and a prototype. I would even call it a prototype. It's actually a working product rather than the actual spec itself.

**Peter Yang** (3:12)
Yeah, I always suspect that engineers don't read my specs carefully. So I always try to keep my specs to like less than two pages to begin with, because no one wants to read this shit. But yeah, the AI agent will actually thoroughly read it.
So that's a good thing. Okay, so before we even get to the spec though, first, like you said, you have to understand the customer, understand the product problem. And how do you guys work with AI to figure out what to build or what the customer pain point is?

**Geoff Charles** (3:37)
Yeah, so the advantage that we have is that we have 50,000 plus customers on ramp and growing super, super fast. We have over a million end users. And so that gives us a ton of signal. We also have a ton of people on sales, on support, on account management. And so those are all touch points that we can leverage to understand kind of what the problems are and what opportunities are and what we should be focusing on. The question is around like, how do you actually sift through all this noise? And that's where like a large language model is fantastic. So the first thing we invested in is what we call voice of the customer. And typically it was a person that we hired that tried to do all this work themselves. Now it's basically an agent. And that agent is essentially able to sift through all our GONG recordings, all our Salesforce notes, all in-app surveys, all support tickets, all in-app chats, any email that is being sent to account managers, and essentially gather all that context, as well as our Snowflake database and our analytics, and help answer any question that product managers have around their persona, their pain points, their workflows, and the gaps of their products.