← Behind the Craft
Behind the Craft
Inside Ramp, the $32B Company Where AI Agents Run Everything | Geoff Charles
00:44:02
Loading summary
Transcript84 lines
- [00:00]
Jeff
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 like the spec, the perfect spec. They have to understand that it's actually AI that's reading the spec now versus engineers. 50% of ramp's code is built by AI and that's 50% up from 30% in December. It'll probably be 80% by March.
- [00:20]
Peter
And this is not just like a front end prototype, right?
- [00:22]
Jeff
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 sales people are also getting activated. My job is to automate my job and all our jobs is to automate our jobs.
- [00:40]
Peter
All right, everyone, my guest today is Jeff, CPO 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 Jeff and team shipped over 500 features and hit over a billion dollars in revenue, all with around 25pm so yeah, really excited to talk to Jeff today and welcome Jeff.
- [01:03]
Jeff
Super excited to be here. Thanks for having me, Peter.
- [01:05]
Peter
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?
- [01:14]
Jeff
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 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 I also expect that a lot of the things that we're going to talk about today are going to be outdated even by the time that you even share this recording.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 excited for it. But yeah, I mean, the product development process hasn't dramatically changed in terms of 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. You know, the cost of code is basically down to almost zero apart from the tokens and. 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.
- [02:29]
Peter
Yeah, so basically the PMs will make the product first, pretty much by themselves, or make the prototype at least and get some validation before doing anything else.
- [02:39]
Jeff
Yeah, I mean, PMs often pride themselves on the spec, the perfect spec. And um, they have to understand that it's actually AI that's reading the spec now versus engineers. And so, yeah, the spec itself is, is, 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.
- [03:13]
Peter
Yeah, I always suspect that engineers don't read my specs carefully. So I always try to keep my specs to 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 problem. And how do you guys work with AI to figure out what to build or what the customer pain point is?
- [03:37]
Jeff
Yeah, so the advantage that we have is that, you know, we have, 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, and that's where 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, the pain points, their workflows and the gaps of their products. So happy to jump into that, but that's a huge thing that we've invested in.
- [04:45]
Peter
Yeah. Are you able to give a live demo of that or show us how that works?
- [04:50]
Jeff
Let me share one version of that. So this is our voice of the customer tool. And as you can see, you can ask any question from, for, on this, on this, on this bot. And this bot will literally go through any type of question. So you know, for this demo I asked you know, what's feedback that people have on our procurement product.
- [05:08]
Peter
Right.
- [05:09]
Jeff
And you can, you can see that the sources. So do you want me to look through support tickets, chat logs, sales research, feature requests, et cetera? I said, okay, let's just go through support ticket and chat logs. It literally went through 90 days of support tickets and chat logs and identified the actual key topics that we needed to focus on as well as links to the underlying assets for me to double click into. So you know, purchase order management, approval for routing chat, you know, chat understanding with ramp assist exports, currency constraints. I mean this is like this was done in you know, from 38 to 40, so about eight minutes and something that would have taken eight days for a human to actually do across the entire volume.
- [05:50]
Peter
Yeah, I mean this is basically like, kind of prioritizes your roadmap for you are it like has number of support tickets and everything.
- [05:57]
Jeff
It at least helps you identify with a ton of context the problems that your, your customers are facing with and enables you to go deeper. So then you can, you know, it's essentially a conversation, right? Imagine you have this like full blown analyst. How do you continue prompting the analysts to go deeper? So now it's like, okay, I want to go super deep on this specific problem case. Bring customer quotes, bring me some like log rocket sessions, bring me, you know, customer IDs that I can go and research to create a email that I can go to that I can use, draft that in my Gmail account for me to actually automatically send to set customer to book meetings on my behalf. All of these things are basically prompts and this agent actually has all the connectivity to be able to do these things.
- [06:35]
Peter
And I love how the user interface is just like a Slack channel or like I guess you can DM the agent too if you want.
- [06:41]
Jeff
Yeah, 100%. We've seen like Slack being a great place to actually host these things because that's essentially what you would do with a human, right? You would, you would slack your product operator with or slack a team channel being like go do these things. It's a Very natural way of doing work and personalizing these agents as essentially your coworkers.
- [06:59]
Peter
Oh, wow. Okay, okay.
- [07:00]
Granola Host
This episode is brought to you by Granola. If you're in back to back meetings, you know how much work it is to take notes, live and clean them up afterwards. That's why I love Granola, the best AI meeting notes app in the market.
- [07:13]
Peter
Here's how I use it.
- [07:14]
Granola Host
Granola automatically takes notes during a meeting, and I can add my own notes too. After the meeting ends, I use a granola recipe to extract clear takeaways and next steps in the exact format that I want. Then I can just share notes directly in Slack with my colleagues or even get Granola to share their notes automatically. Honestly, of all the AI apps that I use, Granola is the one that saves me the most time. Try it now at Granola AI Peter, and use the code Peter to sign up and get three months free. That's Granola AI. Now back to our episode.
- [07:47]
Peter
So that's a qualitative piece that you showed me. What about the metrics and the data piece? Do you just give people access to pull data themselves or.
- [07:58]
Jeff
Yeah, so there's. The space is moving very, very quickly. So six months ago, we launched our own bot for data analysis. I'll give a quick. A quick view of this. And this is now outdated and I'll tell you why. So we launched what we call Ramp Research. So it's funny, like, before you would, you would ask a data analyst or you would try to do it yourself with, you know, looker Hex is getting pretty good at, like, creating your own prompts, et cetera. But it was still, you know, fairly, fairly a lot of work to, like, get an answer to a question, right? And now it's like, hey, I have a question. Give me the answer. So now we have Ramp Research that essentially, I mean, the use cases are insane. And actually, by making it easier for people to ask questions about data, you actually, you actually increase the number of people who actually ask questions about data and you actually become more data centric as a company. So, you know, what's an example? You know, let's say that you have an automated email campaign and you want to understand the performance. What's the open rate of automated emails that customers sent?
- [09:01]
Peter
Boom.
- [09:01]
Jeff
You know, Ramp Research understands our entire database and understands all the schemas and understands what you're trying to do and. And automatically generates the actual interpretation of said results. And this is used so much. And this was in two minutes, right, by literally everyone. So salespeople trying to find what are our customers in Milwaukee support people trying to figure out the common use cases of XYZ product marketers trying to figure out the performances of their campaigns, et cetera. But I shared that this was outdated in the sense that we now basically have moved to Snowflake CLI plus CLAUDE plus skills. So essentially we've now moved to using CLAUDE code. We have our own database of skills that we've developed. So we have a data analyst skill that essentially fully understands our database and understands how we go about approaching a data analytics problem and the best practices of that. And then we essentially can now prompt Claude to say, hey, build me a full report of the performance of our procurement product and identify the top reasons why people opt in the top blockers in our funnel and draft with me 10 different growth ideas that we could be running. And CLAUDE will actually generate a full HTML report fully baked into our data that is directly actionable.
- [10:27]
Peter
Yeah. Okay, so it's not just Q and A anymore, it's actually doing work for you. Right? That's why it's better than the thing.
- [10:33]
Jeff
Yeah. I mean, at the end of the day, like, it's funny, like, you, you, you know, you ask a question, but you have a goal. Right. So sometimes you ask the question and you get the result. But you should just tell AI what your goal is and you'll actually be surprised at the questions that the AI can actually ask themselves to get to the goal.
- [10:49]
Peter
And have you given the entire company access to cloud? Cloud code? Or is it like just engineers or
- [10:54]
Jeff
everybody can use it, Anyone can use it? Right. And in fact, like, you know, we'll get into this around, like, how you actually become more AI driven as a company. But 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. And so it's certainly not a product for engineers. It is absolutely a product for builders. And, you know, we're talking a lot about cloud code right now. Like, you know, Opus 4546, like big launches and a big, big movement in the last, like three months. I mean, you just saw the anthropic $30 billion raise, but I expect the tools to continue evolving by the time we meet Next, the next 90 days, it might actually be completely different. Right. And so it's less about forcing people to use one tool. It's about giving people full access to any tool they want to share openly what people are using and then get people to Adopt to get to the aha moment. And then. But yeah, we don't want to be dogmatic, but we want to radically empower and also have full visibility on what people are doing.
- [11:59]
Peter
Yeah, let's talk about it later, man, because I think so many companies still don't get this. They're like, oh, what's the cost of this? What's the ROI of this? Why should I give some salesperson this thing? Just don't get it, man. We'll talk about the hater. Yeah, so let's keep going, let's keep going down the product development process. So now you have all this crazy feedback coming and the quant, I mean people talk about product sense like it's some mythical thing, but I think it's just like how much product feedback are you getting? How much are you embedded in the feedback loops and the data every single day? And then you kind of just develop that as a second nature kind of thing. But after you have that, how do you actually. You mentioned that you don't actually respect anymore. How do you define the solution? Just make the product right off the bat or.
- [12:45]
Jeff
Yeah, so there's the problem identification, then there's like the actual like writing of requirements and we have, you know, our own cloud skills for that. Right. So you know, Claude has full access to our notion. Notion has the full context on all our Personas and all the research we've done that's all automatically transcribed and aggregated. And then we have skills in Claude that is like your, your product spec skill. And we've prompt, we've designed it so that it's a conversation based approach. So just like you have maybe like a manager or a peer reviewer on your spec. Claude will actually like interact with you and ask you clarifying questions. So for example, you know, what's the main goal here? What are the main. Here are the trade offs, like should we trade off this or that? Like, have you thought about this? Like, what is the intersection with that? It has all the context about what we're trying to do because it has all of the other projects that we're building and all of that is in notion. And so it helps you basically refine and get to an end state. But yes, we don't really talk about the spec itself. That's just a step in the process. We talk about the actual product and so happy to share a little bit of like how fast we move in terms of prototypes and show you kind of what we're talking about here.
- [13:57]
Peter
Yeah, yeah, please show us the skill and Everything else.
- [14:00]
Jeff
Yeah. So let's go through like an example of this skill and then we'll talk about the actual how we build. So this is an example of cloth scale. Folks should be pretty well versed in this world. Yeah. So product shaping, defining the role, push for simplicity. Surface trade offs, surface questions, key definition of the problem. Looking up all the data that we have access to, do the actual research, look at different competitors, customer evidences. It has links to all these different things. Synthesize the completion, Help me shape this question. So you know, you know, present the synthesis, ask pointing questions around, you know, the forcing decisions, all the different principles that we have and then like related skills. So like this is like a skill that you, that we load up and this, this was actually just built by, by one of my PMs. Other PMs actually have their own skills. We're trying to figure out how we actually get to one strong skill as part of the evaluation process. But this is one of the examples I want to share.
- [15:09]
Peter
So you basically just be like, hey, I want to build some expense tracking feature. This thing actually drives the conversation, right? You actually drive the conversation with you. Yeah, exactly.
- [15:19]
Jeff
And then let's go into the build. 50% of ramp's code is built by AI and that's 50% up from 30% in December. It'll probably be 80% by March. And you know, it's, it's not inconceivable for it to be like 90 to 100%. Yeah, like we've hit, we've hit coding escape velocity and it's a brave new world out there. The question becomes like, how do you make it easy for a non builder to engage with, you know, code? Because it's, it's obviously pretty intimidating. So we invested a lot in building our own visual on top of, you know, any large language model and to radically accelerate like how builders can build and how even like PMs can build. I mean if you have like infinite coders at your disposal, you know, you are actually the bottleneck and you actually need to like start moving faster. So I'll give, I'll give you an example. Let's say that you actually want to, you have a lot of feedback from customers saying, hey, I need more visibility on what needs my attention and I need to understand kind of what's overdue, what's on track and like what's upcoming. So I need to understand like my accounts payable cash flow.
- [16:38]
Peter
Okay.
- [16:38]
Jeff
Yeah, yeah. So all I need to do is I will go in and say, okay, Please build me a report on top of this table that has four metrics. My overdue bills, my upcoming bills. 0, 30 days, 30 to 90 days and the total amount outstanding that we'll need to pay. Okay, so this is obviously a shitty spec, this is just for demo purposes. But the, but Inspect will go and it will actually implement this product and it will understand the task it needs to do, it'll actually plan, it'll understand the code base you've already directed it exactly to where you actually need. And it also has access to our design component library. So I don't need to teach it to design what a metric should look like, what a module should look like, what a click should look like. It has all these components baked in and so it's actually able to just reuse a lot of our existing code to build this thing. And I actually did this yesterday for this demo purposes. So yeah, you know, here's like where it gets to see if this is working. Boom. Wow. So now you have on top of the bills table, your entire, your entire metrics, what's past due, what's coming up in the total to pay. And this took five minutes.
- [18:03]
Peter
And this is not just like a front end prototype, right? This is like the real product or
- [18:07]
Jeff
this is, this is the real product. Back end, front end. And I mean a lot of this is front end code though, because I didn't need to create more endpoints. Like we already have all these endpoints, but Inspect is able to do both front end and back end.
- [18:22]
Peter
Dude, that's because I've been using like, you know, Google AI Studio and stuff just to make prototypes. But that's like pure front end code is. It's not. You can't actually push it through prod, but.
- [18:31]
Jeff
And it doesn't have context on your code base. It doesn't have, you need to, you need. It doesn't look the same as your product. No, I mean here I can, I can literally now I can go in and you know, I have a PR and I can just submit it to the engineering team. And we have automatic PR review processes where you know, a double digit percentage of our PRs are automatically approved. So PMs are shipping tons using Inspect and so are designers, so are operators, so are like, you know, to some extent like account managers and salespeople are also getting activated on this piece. So it's just a massive accelerant. And the number one users are also engineers, engineers using Inspect. And here's the other crazy thing about this technology that I want to Share. Um, yeah. So, you know, oftentimes you, you, you have feedback. So we, we, we, we love feedback. We obsess over like customer feedback. We have, you know, tons of stock channels where people are just constantly posting things. Right. It's very overwhelming once you have, you know, the number of customers that we have. You know, this is an example of, of a, a UX channel and you know, basic thing, right? Hey, like Treasury's a product should probably be case sensitive, right? Yeah. Add Inspect in the webro side nav. Change the following sentence. PR merged. This is just one, and this is a very easy thing. But any question that you have, say you're an engineer and say, where do I get started? There's an escalation, a problem. Like anytime there's an escalation on ramp, AI takes the first step. It understands exactly what happened, it understands where in the code base it, it creates the actual PR and oftentimes it ships it. Same thing with, with support tickets. Anytime there's a support ticket that comes in or someone is confused, we have Inspect basically run through that and, and, and recommend changes and have the PR up and ready for like the PM or the product operator, even the engineer to review and ship it. Like the, the speed at which we can move with these, some of these things is like radical. And this is AI.
- [20:38]
Peter
The first, first pass.
- [20:39]
Jeff
Yeah, yeah, first pass everything. Yep.
- [20:41]
Peter
Let me push back on a little bit. If everyone in the company is shipping these PRs all day, how are you going to keep the product cohesive or like keep the quality bar higher?
- [20:50]
Jeff
A lot of the PRs themselves are like quality of life improvements. We also, you know, within Inspect we have an understanding of like, complexity. And so we do have a process by which we review things based on like the sheer amount of complexity that it has. And it does route to the right person based on, you know, whether this is like a big change on the product side or on the engineering side, et cetera. But we haven't yet gone to a big problem. We also have a pretty robust release process. So once the PR is merged, we will slowly roll it out and before any major changes happen on the product that goes to the rest of our customers, we have an automated process by which I get involved or the directors of product get involved as well.
- [21:31]
Peter
Okay, so you have the typical first, everyone in the company plays with it and then if nothing breaks, you could beta users to play with it and then you will. To all users.
- [21:41]
Jeff
Yeah, exactly. So we have DocFooding. Alpha is like your customers that Are as part of your research group, Beta is anyone that opted into the beta tier. We have about 10% of our customer base that's in the beta tier. So you can launch very, very quickly to beta and then you track analytics on that. And then to go from Beta to gea, we basically require for large announcements, not really this naming convention or anything like that. For any large feature that is material to the customer, we basically have a review process that's fully automated. So because everything is in our databases, we have another bot, Ramp Releases, that creates a ramp release report. It pulls all the information of the context, it pulls a preview of the actual product that we can use. It pulls from our snowflake databases the impact that this feature has had. It pulls from any Slack channel a summary of all the work that was done and it basically synthesizes all the things that it also can do work. So to do release, you basically need a Help center article. Well, it gets written automatically. You probably need internal enablement of what this feature is, how do you use it, why do we build it? It writes that automatically. You can also post it in Slack. So yeah, that's a little bit of how, how we speed up that process.
- [23:02]
Peter
And when you review these larger features, are you reviewing the actual product or are you reviewing. Because a lot of companies are just like the PM writes some sort of document and then it goes through multiple rounds of reviews and then you approve it and then they finally go build a product. But I don't think that's how it works at Ramp.
- [23:20]
Jeff
Right? Yeah. I mean the question is, what is my role in all of this now? And I think in the past my role was, well, I'm the best at the craft, or I'm the best at understanding what customers want, or I'm the best at understanding the data. And that's no longer true. You have a super intelligent platform that you can leverage. So yes, I will try my best to look at all the customer feedback, make sure that this is actually meeting the customer feedback. I will look at the metrics and call bullshit on this is not good enough or this is not big enough. That's something that is fairly subjective. I will go into the product and test it out and play with it and like really just hone in on like what's working, what's not working. But I think the higher level job for leaders now is based on your feedback what broke down in the, in the process, right? So if you caught a poor user experience, what broke down, what prompt failed, what skill failed, what design System failed because giving feedback to the person and so that they can just fix it. That's, that's, that's. Yeah, that's a one time band aid. Right. What you want to do is you want to figure out within the process what broke down and fix that process so the next time you never have that feedback again. Like a classic example for me is like I've asked, I've told the team 10 times. The, the call to action needs to be above the fold. Yeah, that's crazy. You want to, you want to, you want to just, you know, six years of A.B. testing, you want to increase conversion. It's a big button that's above the fold. That's it. And I've said that like 10 times or maybe 100 times. But now it's part of our design crit process, which is a fully automated process in and of itself. And so before it gets to me, within our FIGMA prototypes, those core concepts are actually fully integrated.
- [25:15]
Peter
Okay, got it. Okay. So you don't have to, yeah, you don't have to say the same thing over again. You can provide maybe like higher level feedback or something.
- [25:19]
Jeff
Yeah, my job is to automate my job and all our jobs is to automate our jobs. Yeah, we can talk about what happens next.
- [25:27]
Peter
But yeah, and how about the, how about that? Because another thing that sucks up a lot of time is like this annual planning process of like, oh, I spent like a month to figure out what we're going to build for the year or like for the next three. Three years.
- [25:38]
Jeff
Right.
- [25:38]
Peter
Like how do you guys manage that process? Or is there even like a, like how far do you go? How far out do you guys look on this stuff?
- [25:45]
Jeff
Honestly, three months.
- [25:47]
Peter
Okay.
- [25:49]
Jeff
We can only predict within three months now. And by the way, like within three months you can do what you can do in three years now. So like three months is actually a really long time. Yeah, you know, planning, Planning for me is, I think there's actually like three main objectives to planning. One is actually aligning on strategy, which is much, much more important. Like what problems are you focused on? What problems are you not focused on? And which customer segments are you going after? And, and how are you thinking that we're going to win long term? Like what is the end state for this thing? So it's about trade offs and I think like the conversation should really be about trade offs. The second thing that planning is good for is just having some level of commitment from the teams. Right. Some level of accountability. And the third is to have some Baseline for sales to know what's coming for them when they, when they, when they, when they talk to a customer and the customer asks, okay, like this is great, but you know, I have a lot more needs when it comes to our international exposure. And the sales team needs some basic assets. And so that's the third kind of pillar and that also is like fairly automated. So once the team kind of does their backlog and their plan in notion, we have an automatic process that creates one pagers and then it creates slides and content for the sales organization within our own branding guidelines. And then the sales team can essentially just look at a higher level view of our roadmap to be able to sell effectively against it.
- [27:22]
Peter
Wow, okay. And then you have all these like vision and like, you know how we're going to win stuff. Obviously AI can read it and if something changes, you can just add, ask AI to update it. Is that how it works?
- [27:33]
Jeff
I mean what I ask AI to do is to synthesize information. Like a lot of leadership time is about helicoptering between the nitty gritty problems and then the higher level strategy and roadmap and making sure that every level of the organization understands information at the bottom and information at the top. Like how you communicate to the CEO and the board is very different than how you communicate to the director is very different to how you communicate teams. And that LLMs are incredibly good at because so like the translation layer, right, When I'm at an all hands meeting versus when I'm at a team meeting versus when I'm in a boardroom, very, very different. And so I, I waste a lot less time on those things.
- [28:13]
Peter
Got it, Got it. Okay, great. Let's get to the key question then. I mean you just mentioned that your job is to automate your job. I'm sure all your PMs feel the same way. And so, so what's going to happen to the PM function? Do you think, do you think it's game over or like, yeah, what was going on?
- [28:28]
Jeff
You know it's funny, like I was surprised by like once you automate code, like a lot of people concluded that PMs are, it's like over for PMs. And I, and I was, I thought to myself, it's over for the engineer for most engineers. So maybe it's like a lot of engineers who are like, I'm going to be a PM now because the engineering function is, has changed a lot now obviously there's a ton of value for engineers because I think an engineer now is managing Hundreds of thousands of agents and they can actually scale their impact. But let's go back to the pm, the PM role. Like a lot of what a lot, there's a lot of bad PMs out there or badly trained PMs. I think that the way we've trained TMS in the past has been really, really bad. And we've trained them on stakeholder management, we've trained them on prioritization, we've trained them on communication, we've trained them on frameworks and those are all outdated because code is free. And so like all that matters now is are we going the right direction, how fast can we go and how do we remove bottlenecks and how do we build a system by which like we can accelerate and, and, and to do that I think PMs need to really rethink their skills. So like a lot of PMs join product management because it's a safe job. You know, they might not be good enough at the engineering tasks, they might not be good enough for the design tasks, but like they're really good at, you know, the, the, the consultant. I'm an ex consultant. Like that's why I joined the function. Like I understand that the customer, I can communicate to engineers and I can, like I can, I can really, I can, I can, I can somewhat facilitate decision making. The downside is that if you're a risk averse pm, you're not going to change your way. So I still see, you know, very high performing PMs who don't get it, who haven't yet adopted these, these core skills, who haven't changed the way that they're working because it's worked for them so, so far in their career they've been successful because of it. That is the biggest danger that I'm seeing. And so I, I think that the, the role of the PM is going to shift and I think it's going to shift in two directions. PMs are going to become much more builders, right, because code is free. So just like I showed like a product, right, that I, that I basically built in, in five minutes. It's going to require then like the iteration from the product very, very quickly. And so I think, I think the craft and the building is going to be like really, really essential versus the spec. Like you no longer have to write the spec anymore. You need to actually like be in the product itself. Now an engineer, a great product engineer can do that and a great product designer can do that. The other path for product is the business side. So you know what engineers are and designers often lack is, is an understanding of the context in which the business operates and what actually matters and how we're gonna win long term. So they're really, really good at, maybe they're really good at building really good products. And so give them that. And then the product team should be focused on like okay, but now that we have this really good product, like how are we competing, how are we positioning, how are we distributing, how are we monetizing, how are we actually using this to win and drive enterprise value? And I think that Even looking at OpenAI anthropic, the, the, the it's a, it's a, it's, it's a decision of strategy. Yeah, they have different strategies and, and that's actually where the PM should be, should be really, really focused is, is the underlying way that we're going to win and playing that GM mindset because, because they're gonna, you're gonna have a ton of builders that can, that can build great products, that can iterate on customer feedback that have all the context you've built that system. So now focus on like what actually no one can do, which is to make sure that the product you're building is going to have insane amount of value in the market and insane amount of money for your business.
- [32:07]
Peter
And like a lot of PMs are just like stuck like you mentioned, they're stuck in like cross functional alignment meetings all day, like back to back. So like, and I think it's like a company culture kind of thing too, right? Like do you make sure your PMs actually have time to build or is it sort of like do you have to get alignment from 10 people to ship anything? No, it doesn't seem like, doesn't seem like that's the case.
- [32:25]
Jeff
Yeah, no, I mean we've designed the organization so that we, we do not have committees and we do not have sign offs. You just need to prove that you've added value and then you can go, go for the races. I will say that like it's, it's actually really, really important for PMs to carve out time to build. And I say this, not just PMs but like managers. I think that it's, it's a really tough time to be a manager right now because you're managing a team whose skill set needs to change and you might not actually have that skill set. So I think that right now like going back to IC mode is paramount and I've done this for myself where I Say like, hey guys, like I'm going to be way less meetings. I'm going to be way less in one on ones and I'm going to be like, I'm just going to be adopting AI tools. I'm going to be building and vibe coding and understanding what's working, what's not working so that I can be become more educated. Because it's just, this is just the beginning. I mean, I mean the sheer amount of changes that happened over even the last like three months is profound. And I think if you're stuck in meetings, you're not going to be effective. So definitely creating space for work. And honestly that's also where nights and weekends come in, which is like this is the year that you need to really prioritize learning and growth because no one's going to do that for you. So yeah, it's going to be a wild ride.
- [33:46]
Peter
And if doing it the old way, your company's going to die, Bay basically. Right. If it's the waterfall and all this kind of stuff, it's not going to survive. Yeah. Let's skip to talking about companies that are watching this. They want to become AI native, like ramp, like how you guys operate, how do they go about doing building systems and that kind of stuff. Yeah.
- [34:06]
Jeff
So there isn't one right way. But I'll share kind of what we've done and we've kind of built a framework around this. We think about being AI proficient in multiple levels. The bottom level is people who sometimes use ChatGPT. Right. We'll call them like the L0. Level one is like people who build their custom GPT. Maybe they build a notion agent, maybe they've built, they've used like cloud code to like do some of these things. Level Level 2 is people who are actually like fairly proficient. They, they have been able to build an app that, that automates part of their job. They have been able to commit code or feedback to other people's work. And then level three is like the fundamental, like systems builders. Okay. And our job is to get everyone in the organization up the ladder. And the way we do that is as follows. The people who are still in L0, they will most likely not be at the company because the fact is like you can, you can, you can tell them as much as possible. If you're not a self starter and you don't have that growth mindset, like it's gonna be very, very hard to train, to train you out. So, so that's L0 the L1s to get L2s, L2s into L3s and L3 is like basically like influence the rest of the organization. And the way we do that is we have a lot of public channels around people sharing what they've built. We've made it really, really easy for anyone to adopt these things. So we've removed any constraints around access, around tokens, around budgets. We have like the setup of those tools are extremely well done. So you have access to all the different MCPs, you have access to all the different skills. We even have like an internal repository of skills that people are deploying to, you can pull from those. And then we have a lot of culture around in all hands around showcasing non builders doing things. Our finance team building their own treasury management system, our legal team doing contract reviews, our marketing teams automating website creation to get people inspired. And then we have office hours that people can join to ask any questions to get set up. We have designated experts that people can just ping and their entire job is to get, to evangelize, to get you set up, to get you comfortable, to get you going. Those are like some of the principles there. And then the other piece is just like you know, hiring and performance management. So on the hiring front we now have an absolute requirement for anyone who joins the company to be somewhat proficient for these tools. There's just absolutely no excuses. And in the interview process we'll have basically a dedicated session for this where like they will either, I mean for the product team, I literally have a session where you're going to build, you're going to build a product. Like you're going to show me a product that you've built and you tell me exactly why you built it, how you built it and how it works. Like it is a full blown prototype. And then we also track usage of AI across the company. So you know we have, we vibe coded this, this, this product even within the team where we can see every other company and their full usage of tokens across Notion, AI, ChatGPT, Cloud Code, Cloud coworker, our inspect tools or any of the internal apps. And we can see kind of like who is actually pushing the bar to amplify and who's not and who we need to intervene on.
- [37:22]
Peter
Do you worry about like cost of running out of control or like the ROI is so clear that there's no, it's just fucking. Just give everyone access, let them do it.
- [37:29]
Jeff
Yeah, I mean I haven't done the ROI around like if you, let's say you have a person who's, who has $100,000 salary, how many, how many tokens should this person use? And there's debates right now around, you know, you know, productivity versus just like noise and you don't actually need these things. I think right now we need to invest the budget for people to discover and if we are not as efficient in that spend, that's okay. That's our competitive advantage. That's why we raised money, that's why we have a pretty good war chest. But I can safely say that, you know, we pay our employees a lot of money and the token consumption per employee is not even close to double digits. And I think it's, I think it's not unreasonable to think that it should be higher than your salary because like, if you have agents that are able to do 10 times more work than you, then why would you not pay them twice as much as you? And, and so I think that's like the way that we should be really framing it. But yeah, I would say like, we're not really worried about costs. We're mainly worried around we have like the next X months or X years where AI has not yet fully one shot at a ramp platform and we need to use that to our competitive advantage to move as fast as possible.
- [38:50]
Peter
Yeah, and I feel like a lot of the internal tools that you show me are also really good for ramp customers.
- [38:54]
Jeff
Right.
- [38:55]
Peter
You can just like, you know, make that available for ramp customers 100%. Okay, last question, man. So if I'm a PM or builder, how should I think about my career these days? Like the old climb the ladder to VP or whatever, Is that still going to work? Or how should I think about being employable still?
- [39:15]
Jeff
I would say, I think that where you should be optimizing is not management, it is being the best builder in the world. I would say that management is probably dead. There's always going to be value in someone giving you feedback and coaching and being your advocate and being a team leader. But now is not the time to build that skill set. Now is the time to be very, very proficient in this new technology and to radically improve the way that you use it. Um, and so I would say for, for, for all the PMs out there, you know, get really embedded in these tools and that's why engineers are so good at, at, you know, understanding what AI is capable of because they live and breathe it like they're, they're, that's the first knowledge work that has been, you know, mostly automated with, with, with coding agents, but it's coming for everyone else. I mean, it's going to come for PMs, it's going to come for designers, it's going to come from, for any, any white collar job. And so I would say just get very, very proficient at using these tools. And ultimately your career is about impact. And right now, the impact that you can have is to ship great products faster and move more metrics for customers into the business, and so create a lot of space to learn these things and have the beginner's mindset, the humility to understand that the way you're doing things is not the best way. And I think my job as a leader is just to get people to get to that aha moment. And Even my brightest PMs, I had to sit down with them and say, we're going to go through this workflow together. What have you done today? And I will show you a new way of doing it. And once you get that aha moment, that red pill, there's no coming back. You're like, oh, I get it now. And it'll also make you a better builder because the software you're building, if you're in B2B and even B2C, it is going to look radically different than what exists today. I mean, fundamentally, software is dead. It's all going to be like coworkers. And if you haven't used coworkers in your own job, you don't understand how like, that actually might look like your product a lot more than you think. Right.
- [41:26]
Peter
Yeah. You don't have to process.
- [41:27]
Jeff
That's exactly right. Like, Ramp itself is going to look much more like a finance coworker than it does, you know, tables and charts and workflows.
- [41:34]
Peter
Yeah. I found, like, it's all like, you know, I've been using Open Call everything. It's not like clis and like, you know, there's like, no one wants to touch buttons anymore. It's just like, let me, let me talk to my co worker and, you know, get him to do stuff for me. So that's basically it. Yeah. Cool.
- [41:50]
Jeff
And how do you build one of those great co workers? Domain level expertise is another one. Like, I think that in the past it was like, I'm going to talk to customers, I'm going to kind of understand the requirements and kind of build a product for them to do their job. Right. But if you're doing the job of your customers so that they can do other things, you need to, you need to actually be an expert or you need to build a system by which you can ingest that expertise. Right. So accounting, like you can build an accounting workflow where they have to go and code things, but if you're actually going to code on behalf of the accountant, you need to, you need to deeply understand that the philosophy or be able to extract that knowledge, like how do you download, you know, CPA and all the best practices and actually bake that into your product. So it's a very different way of thinking where fundamentally like a login in your product in the future, I think is going to be a failure. Right. And I think that's also how we think about it. We, we track the amount of time you spend in Ramp and how we can actually reduce that time as much as possible, which is, by the way, the opposite of how many PMs are trained. The, the, the Facebook and Netflix.
- [42:48]
Peter
Right.
- [42:49]
Jeff
The fangs of the world that are, that are mainly advertising businesses. Like, it is the opposite. And I think there's going to be reckoning for sure. But also a very exciting time. I mean, you know, I think it's very, very scary and a lot of people are alarmist and everyone should be paranoid, but man, it's an amazing time to be builder right now and especially a product manager where you have taste and vision. The time it takes to go from your taste and vision to a product is shorter than ever. And I think it's really, really, really exciting time to be a builder here.
- [43:16]
Peter
Yeah. And I think another thing you mentioned is just like setting up systems to dedicate all the bullshit work to AI, Right. So you can focus on stuff that you actually enjoy doing. Like that's a key part of it. So.
- [43:27]
Jeff
Yeah.
- [43:28]
Peter
All right, Jeff. Well, I mean, thanks for being an inspiration, man. Like, I, I, I, I think hopefully, well, hopefully every company can learn how to operate like Ramp.
- [43:37]
Jeff
Yeah, we're just getting started. There's also a lot of things that, you know, we're not doing well, that other companies are doing super, super well. I think, you know, part of me going on this, this talk is not to, to share that we've, that we have all of it figured out. Most of the things that you saw here are things that we built in the last months. So excited to keep the conversation going, excited to continue learning and really a privilege to be here today. Thanks a lot for having me. Yeah.
- [44:01]
Peter
Hi, Jeff.