Vibe Coding Is Easy, Scaling Is the Real Job

Lightly edited for clarity. Speakers: Ysi Gonzalez (Vantelira) and Rafael Angarita (SERHANT.).

This episode is powered by Vantelira. We go inside a scaling operation, fix the process, and build the system that keeps it running.

Introduction

Ysi: Hello and welcome to The Forward Deployed Operator. My guest today is Rafael Angarita.

Rafael is a marketing systems architect with over a decade building marketing infrastructure across luxury real estate, technology and media. He is currently Director of Integrated Marketing at SERHANT., where his team builds the system connecting brand strategy to execution for thousands of agents. We're going to get into that later on. If any of this sounds familiar, stay put — we're going to talk about it deeply.

Rafael, welcome to the show.

Rafael: Thank you so much, Ysi, for having me. Super excited to be here with you, and to hang out and really dive into some of these forward deployed operator rabbit holes that we could get into, and hopefully come out of it with some valuable insight for your audience and everyone listening.

How he ended up in marketing

Ysi: Thank you so much for that. The first thing I'm going to ask you: I know I made a short introduction, but do you want to talk to us about how you ended up in marketing, and then all of the way to SERHANT. and the position you're currently holding?

Rafael: Sure. I stumbled into marketing, or digital marketing, almost by mistake — or really simply luck. When I was back in college I went to school for business. I didn't really know at that point, you know, I was 19, 20 years old. I didn't really know what I wanted to do as a career. I knew that if I went to school — or I trusted that if I went to school, I got a degree and I prepared myself — an opportunity was going to come up, and I just needed to be ready and be able to identify that opportunity.

Through my last year of college I got an internship to work with a financial company in Miami. I grew up in Miami — I'm from Venezuela, but I grew up in Miami — and I worked for a financial company that was doing personal loans and business cash advances, managing their blog.

They did a ton of micro-site SEO. That was part of their strategy back in 2012. They had a network of organic content, an inbound lead capture machine through micro-sites and localized SEO. So I came in to help manage that network a little bit and support with some of their social media profiles as an intern, and that turned into a full-time job.

Realizing where the money actually was

Rafael: Then I was able to recognize there's a ton of money being made through online advertising. These companies are paying agencies thousands of dollars to manage their paid search, or paid social, or their lead gen campaigns. And there are definitely companies investing internally to have an internal team handle that rather than paying out to agencies. So: let me just really double down and consume as much content as possible to teach myself.

Through that journey, for the next eight years, I was really lucky to know exactly what I didn't know. So I was able to seek out that kind of knowledge around paid media management. And I'm naturally a very creative person, so I was able to bring in some of my interests around creativity, design, just overall good aesthetics, which helped a ton with direct response advertising.

That luck, I guess, of bringing that skill set — I was also able to work with really great people I could learn from. People who had been in the space of performance marketing for a while, and I was able to absorb as much information and key insights and different approaches as I could, which obviously I added to my toolkit.

Fifteen years, Miami to New York, and then SERHANT.

Rafael: Fast forward to now, 15 years into my career, I've been able to move from Miami to New York City and work with different growth-stage startups across online education, coaching, tutoring, direct-to-consumer brands in the healthcare space, insurtech companies that were able to help over a million people get insurance on a yearly basis with healthcare.com.

And now, for the past six years, I've been working with SERHANT., one of the fastest growing real estate brokerages in the country, if not the world. Really leaning into not just everything I've learned over my career, but also tapping into new ways to promote and build brand through organic content.

So, kind of blending the performance side of things — because we're very much data driven in everything that we're doing now — but also leaning into really compelling storytelling to build brand and provide a platform to everyone that works with the firm, to grow their business and achieve new milestones for their lives.

"I didn't hear any software engineering in there"

Ysi: It's a very interesting journey, and in everything I'm hearing I did not hear any software engineering, developer, or even process improvement. However, you've come to that.

Rafael: Yeah.

Ysi: So, the way that I found you — I actually sent an agent to investigate who else in the world has been talking about forward deployed operations. Who's building things from the inside of a company, improving the processes, building a solution for them? And you were at the top of the list.

And I thought, I need to talk to Rafael, because the interesting thing about this is not just that you built something — which I want to know more details about. It's that you're inside this large organization. You understand marketing, you're a subject matter expert on it, and all of a sudden now we have this crazy experience with AI that allows us to turn that into something else.

That's where I want to tap in a little bit. What happened, that you said: I need to do something, I need to start getting into AI? And how did that translate later on into what you've built so far?

The transformation began 18 to 24 months ago

Rafael: Sure. It's been definitely a journey of transformation, let's call it that. On paper I run integrated marketing for the firm I work for, and I've been in the marketing unit for many other companies over my career. But in practice, I think my transformation journey into a forward deployed operator began maybe 18 to 24 months ago or so.

In the role that I have right now, we are servicing — you mentioned a thousand agents, but it's actually double that amount. We're looking now at almost 2,000 agents across the country, and those are the clients. For my nine-to-five, I needed to find ways to scale my team without increasing cost.

So the transformation for me began 18 to 24 months ago, where I'm realizing the challenge I have in front of me. Being able to scale my team, and/or scale the output, was just the puzzle in front of me.

n8n, Zapier, and the node that changed the shape of the problem

Rafael: This is when n8n was really ramping up, becoming very popular amongst marketing automation conversations and blog posts and courses and all the things that were being sold to teach you that skill. I had already been leveraging tools like Zapier, or Make, or Integromat at that point — I think they still had that brand before they transformed into Make.

And n8n was the first workflow automation tool that allowed you to have a layer of, you know, bringing in an LLM node to start making decisions, and to have a logic point of optimization, or choosing. So I didn't necessarily have to map out every single step.

But learning n8n is not an easy thing. It's not something that just clicks, and it's so technical. There's so many fields, so many things to break. So I took — I want to say almost a year — of really leaning into n8n and trying to understand it, trying to master it.

Learning by deploying at a real enterprise

Rafael: And I was doing that by applying workflows and deploying workflows at a real enterprise. It wasn't just watching a course or watching a tutorial and making a dummy project and trying to see if something output the test you were trying to do. I was actually outputting stuff into the real world that had real consequences.

So having that layer of, let's call it pressure, really took me to that edge of: I have to figure this out.

Claude Code, then Cowork

Rafael: After about maybe six to ten months of learning that, optimizing that, deploying different types of automations and processes, Claude comes out with Cowork. Claude Code had already been out, but it never kind of came across my peripheral to even become curious about it. I always felt like it was something so far ahead of me, or like I needed to be a full developer and software engineer.

Once Cowork came out and it was like the user-friendly version of Code, I definitely went in and started testing it, and I learned about MCP connectors.

I leaned into everything that had been said about leveraging AI from 2022, when ChatGPT first came out: you need to prompt this system in a structured way that defines its role, defines the kind of output you're expecting, and really provides clear direction for what you want out of it. So I started experimenting with that same mindset using Claude Cowork, leveraging some of the MCP connectors.

"What if it breaks everything I've built?"

Rafael: At that point it was a little intimidating, because I was like: I'm giving access to these different tools and personal assets, from email to maybe my n8n account to other areas of our infrastructure. What if something goes wrong? What if I prompt it and I ask it to do the wrong thing, or it starts doing the wrong thing and it breaks everything I've been working on for the past two years? How do I create some kind of moat that protects me?

So I started with small wins with Claude Cowork, and little by little I started building confidence. I moved from the Claude app directly into setting up a proper project leveraging VS Code, and then the next thing was setting up a GitHub repository, and then now I'm pushing code to that repository, and then I'm connecting Vercel.

A simple automation that became a full application

Rafael: And all of a sudden I lift my head up. I was trying to set up what seemed like a simple automation — I was looking at one change that happened in one platform, that triggered another action, that sent out a notification. Now all of a sudden I have a full front-end application that I'm interacting with to trigger those automations and take action on my behalf.

At that point I was like, I'm too deep in to go untangle what I just did. Let's just really refine this, polish this, and start bringing a lot of the functions that my team is accountable for into this app, into this tool, to help with everything end to end. Content distribution — if you follow SERHANT., if you follow what we're doing, we are by far the real estate brokerage with the strongest digital presence across PR and social impressions. From my perspective either way, we are by far the ones leading the way.

And the tool that I built is able to contribute to that mission, and really enables my team to scale their output. I've been able to basically create the output of possibly two or three full-time employees through that application. Everything from content creation to content scheduling to content publishing, and kind of everything in between — from communications with our clients, to notifications to our team, to clear views of how everything is running according to our criteria.

Having seen how different machines work

Rafael: It's a very custom integrated tool, and I think the experience I've gained over the years working with different marketing teams and different industries means I've been able to see how different machines work. Whether it's a very much organic content machine like I have now with SERHANT., or something on the opposite side of that spectrum, which is a company that will deploy a million dollars a month on paid media — which is not necessarily accounting for organic content, because they're just deploying money and it's a pay-to-play situation.

I've been able to see and understand end to end how everything is working on the content delivery, the inbound lead flow. And then also, because of the roles I've had, I've been able to work with dev teams on a business intelligence dashboard — bringing databases together with data from Facebook, data from Google, data from the website, and blending that together to build out attribution and tracking. Building a website with connections and integrations with our CRM, and so forth.

I've been part of those projects that have a lifeline of, from kickoff to actual delivery in best case scenarios, three months. Depending on the scope, maybe six months of work.

The video game rush

Rafael: And once I started playing with these — and I don't want to say playing, but it felt like I was playing, because it was the same rush that a video game will give you. I'm playing a game, I achieve a new level, and you're like, yeah, I got better at this thing. Let me keep on playing the game and try this other thing, and try this other thing. And the output that comes out is a functioning tool, exactly the way that I imagined it.

Why writing PRDs is the skill that transferred

Rafael: I think the benefit of having gone through the process of actually working with a human team — where I needed to put together a product requirements document that outlined everything I was hoping to get, for let's just use a marketing dashboard with key insights and different metrics we're tracking. If it was custom, now we're looking at having a data warehouse, and the transformation of that data, and standardization of that data, and making sure everything is tracked and all connected. Again, that's months of work.

And because I had gone through that process, I knew exactly how to put a request together with the spec and exactly the setup that I needed. Then I realized: the more detail and the more time I spent putting that request together, and I just fed it to the AI agent I built — without realizing I was building an AI agent — the output was happening almost immediately.

I was able to iterate, and in a matter of a week or maybe two weeks I was putting together tools and resources that before would take six months of work with several humans involved in the process. Now I'm working directly with my AI agent and the key stakeholders within my team who are putting requests in. And because I understand the business and I understand their role, I'm able to ask them follow-up questions to really get the detail and the core essence of what we're trying to deliver. I work with them to build custom tools internally — for SERHANT. Studios precisely — and it has really changed the way that we are operating end to end. And all of that has happened in the last six months.

Understanding the work is what makes it translate

Ysi: And that's the key of what you're saying. You're describing understanding the processes and understanding the work exactly how it needs to be, to be able to translate it.

I've been having all these conversations about forward deployed operations, what it means and what it takes, and the same conclusion comes to mind: people doing the job, understanding it, and then moving towards okay, we need to improve it, compress it, accelerate the deliverables — but we also understand the technology that needs to connect to it.

It's just very inspiring to hear you say that. And I totally get the dopamine hits that we get when we accomplish one level of the game and then you want to get to the next one. It's just very exciting.

"Where do I start with AI?"

Ysi: Part of the mission of the show is this: I get a lot of questions all the time. Where do I start with AI? What's the training I should be getting? Where do I go?

And you describe it perfectly. You need to start and get curious and continue achieving levels in it, because there's no better way to learn something than putting the work in, putting the hours in, spending the time using it, understanding it, and failing at it.

Have you had any failures that you want to describe a little bit — something that went totally wrong and you had to redo it? What did you learn from that?

Give yourself a real goal

Rafael: Yeah, absolutely. I'll give you the failure now, but addressing the first part: you have to actually give yourself a goal. Anything. It could be as big as something that could be deployed internally for your actual nine-to-five job, or for a client, or whatever it might be, or for yourself for your own benefit. But the only way to actually learn is by actually doing — starting a project.

And mind you, you should spend some time consuming content to understand what some of the basics are. How do you set up a basic Claude Code project, or a Codex project? I'm right now using both. I'm using both models to audit each other, and I'm building with both at the same time. My projects now are optimized to be able to deploy both Claude and Codex.

Use the models to teach you how to use the models

Rafael: From there, I think a lot of people kind of miss the mark on leveraging the models — whether it's any of the models from Claude, or OpenAI, or even Perplexity, Gemini. I'm not necessarily spending too much time with Perplexity or Gemini besides tracking how many times any of our entities are being mentioned, or how much traffic we're getting from those models.

But using the models to teach you how to use the models — that's something I've done quite a bit of. You are literally saying, in the chat window: you are my coach. You need to break this down. I am not a software engineer. I'm not that technical. I understand what I'm trying to achieve, but I need you to break this down as if I know nothing and I am the least technical person ever, so that I can continue to educate myself and get better at what I'm doing.

That's something I have done a lot, and I'm still doing that today.

The failure: what the infrastructure actually needed

Rafael: Now, as far as a failure. When I started really diving in and I realized I was building an application, I started kind of understanding what is the infrastructure that is needed — the different tools, backend databases, and different connections I need to have for this to be actually functional.

I was kind of building it isolated, by myself, putting it together, getting insights from the people who were going to be using it eventually as far as what the thing needed to do and accomplish for them. But I was the only user that was logging into it.

As I'm building and I'm seeing the thing come to life, I'm making UI changes and I'm plugging in my Figma account and I'm mocking up exactly what the login journey is going to be, and all the buttons and all the UIs, and how one screen will go to the next screen through the different pipelines and workflows we have set up. Everything is building my confidence and I'm feeling like I have superpowers. Like I have a team that is all digital. It's almost like imaginary people, but they're actually executing and they're building things and they're functional, and I'm able to click the buttons and I'm triggering the things I want to trigger.

Four people log in, and the app crashes

Rafael: Eventually I got to a point where, okay, this is ready to be deployed to my team. My team is ready to be onboarded to this new software. So I'm putting the guys together and I'm setting up training sessions.

I started with a small group of four people. These four team members are going to be the first users. I want to train them to become the power users of this new tool we're setting up. Let's all log in. We're going to do a workshop. We're going to walk you through how to use it.

All four people log in. The app completely crashes. It just stops working. And I'm left in the middle of the conference room feeling a little embarrassed, just could not figure out what was wrong with it.

"Vibe coding is easy"

Rafael: But essentially, I think the lesson there is: we need to understand the real infrastructure. Vibe coding is easy. If you have a good imagination and you're able to elaborate exactly what the spec of the build is, these AI models — especially the frontier models — are going to be able to deliver that for you.

But you really need to get into the details of the infrastructure and the kind of usage that is going to be happening with the application or the tool you're building, so that it's able to support that kind of usage.

I think one of the biggest leaks we had was with how we integrated one of the task management tools we're using, and bringing information into that app.

Anyway, there was a lesson. It was a little bit of an embarrassing moment for my team, because I had been building up the hype about this cool tool that's going to completely change the way that we work and save us so much time. And in fact it is saving us time, after we addressed all the infrastructure challenges we had. But when I first rolled it out, I felt like, man, I just wasted six months of my life building this thing. It's not going to work.

From four users to forty

Rafael: We figured it out. It works. It got deployed. Now we went from four people — that initial group of users — to that tool being used by 40 people or so, and it's processing over 300, 400 requests a week.

It's something that if I were to pull that tool from the team right now, it would completely break the system. We would not be able to keep up with the demand we have. So I'm incredibly proud now. And it's not something I built on my own — that was such a collaborative effort.

"So what did I just become?"

Rafael: It kind of taught me how to really be — again, another transformation season — of realizing, oh, I went from trying to build this small automation that solved one specific small problem, to building an application that does so much for us within my team.

So what did I just become, right? After I deployed that and I was able to see the impact of it, I started digging into: my job is no longer the same that it used to be 12 months ago. I went from running integrated marketing for the company, and being very involved in organic content distribution and lead generation and demand generation, to now I'm deploying systems.

So did I just become some kind of engineer? I'm not trained as an engineer, but I know what it takes to document a clear SOP to not only train new staff members, but now I'm able to train my AI agent to get things done for us.

I landed on this new role, or this new skill set, that people are now talking about a ton online: I'm a forward deployed engineer slash operator.

Fifteen years of marketing with an AI layer on top

Rafael: And it kind of reminds me of back when I first started my career out of college and not really knowing exactly what I wanted to do as far as work. I was just going to be patient and wait for an opportunity, and once that opportunity is there I'm just going to run with it. That's how I stumbled on digital marketing, performance marketing, where I eventually developed a career.

But now the opportunity is to bring all of the experience and all the expertise I've been able to gather over the last 14, 15 years around marketing, positioning, strategy, channel mix, budget optimizations — all of that — and bring that to the forefront with an AI layer on top of it, and work with people to create custom solutions and scale their output. Which is a fun job, without a doubt.

Every system you fixed became someone else's win. The next one has your name on it. An eight-week live cohort. You solve a real problem inside a real company, ship a working AI system, and leave with the case study, the build and the title: Forward Deployed Operator. Join the waitlist at fdoschool.com. That is fdoschool.com.

Lily Bloom makes personalized wooden and acrylic keepsakes out of Montreal. It is my choice for any keepsakes I want to give to my friends, my family, my customers. First birthday photo boards, cake toppers, first day of school signs, anything in between and anything you need. Every piece is thoughtfully designed and crafted with love, care, and an eye for detail, making your celebrations and your gifts even more special. Find them on Etsy as Lily Bloom Shop. That's L-I-L-I-B-L-O-O-M-S-H-O-P. Lily Bloom Shop.

Packaging it as a service

Ysi: I browsed to your website, of course. And you're now packaging this as a service, right? If I were to call you — okay, for my business or for future ventures I need somebody to take care of the whole thing end to end, set me up with a kind of software like you did — is this what you're offering? Walk me through what that is, because I'm very curious.

I like to know about this type of project, because we're always needing help. And knowing that there are forward deployed operators for any subject matter expert out there, it's incredibly exciting.

Rafael: First stop needed to be a website. If you had a website, that really kind of made you stand out in the crowd. Now, or over the last 10 years or so, obviously a website becomes really a basic thing. It's just the simple business card that you need to have.

Now it's a lot about content creation, and how you can position yourself to really tell an authentic story and position yourself as a thought leader in the vertical that you're operating in.

And then from there things have evolved into: okay, you have a website. All right, you're creating content. What now? How can you really optimize the way that you're operating, the way that you are planning your content? Are you able to have a full-time job and be able to do content research, and nail down new ideas, and plan out your newsletter, and have a website that's pumping out content for people to be able to find you? And that's really how you found me.

He built the offer for himself first

Rafael: So — let me put something together for myself, because I've been wanting to scale and really start creating more compelling content online, to share some of the insights I've been able to gather over the years, and really position myself as a resource for folks.

I am setting up this offer, which again I'm setting up for myself, but it's now packaged up and ready to be implemented. It's easy. You want a website that does all of the things, we could definitely talk and set you up.

Look at lifestyle photography. If you don't have lifestyle photography, we could get a couple of shots of yourself and run my AI agent to create new lifestyle photography — AI photography with yourself, without changing you or your facial features at all. But we could create anything we imagine, and there's no longer a boundary.

So being able to deploy that and offer that to folks and make it so accessible is a game changer. From a personal brand, to a consulting business, to a service business — there's no limitation anymore.

The edge: knowing the tempo a growth-stage machine needs

Rafael: I think what really gives me an edge is that I've worked in growth-stage startups, and I've seen how the machine needs to be operating, and at what kind of tempo and output it needs, so that you're able to actually break through the noise and bring in the kind of demand you need to grow your revenue.

And that's something no AI is going to be able to give you. It's one of those things that only time and actual experience can. There's no other way. There's no shortcut for that.

Why the website worked

Ysi: And whatever you did with your website worked, because it popped up. I was like, whoa. I remember it was, I think, at the beginning of last month that you published an article about it, and like a week later I found it. So it didn't even have time — you probably know a lot more about this than I do — it didn't have time to index itself into Google and go through the whole process.

Rafael: Part of the product that I have is that I have Google Search Console plugged in. So my AI agent is optimizing my website to be as visible as possible, not only to search engines but also to LLMs.

And again, this is literally the past 15 years of working. One of the things I've been really lucky with is that throughout every single job I've ever held, I've never been pigeonholed to one particular discipline. And I've never wanted that.

Never being pigeonholed to one discipline

Rafael: I've always raised my hand to say, hey, I know I'm here to run our paid social engine and scale that channel from 500,000 to a million dollars a month, and I will achieve that. But I'm also kind of interested in building out these email journeys that are going to nurture the leads I'm bringing in. So I want to get involved in that. I want to map out how many emails, what is the copy, what is the overall design, how tracking is happening.

I want to be involved — one, because I'm curious, and two, because it's going to give me peace of mind that my work is actually being delivered on a strong foundation across the board. From tracking, to general SEO and website optimizations, to form optimizations.

There's so much that goes into setting up a proper contact form. What are the fields? How do we label them? How are we structuring them? Are we able to track typing speed on the form, to get a sense of how tech-forward this client is who is submitting my form? Is it going to be an uphill battle to get them to believe in technology as an offering, or in these kinds of services as an offering? Or is their typing speed telling me they already understand — they use the computer enough to understand that tech and some of these solutions have to be part of the mix.

Building the checklist into the tool

Rafael: So all these little things that I'm now able to — obviously you have a checklist, but building it into the tool, so that if you want that, we could just enable it. It's already wired in.

Now it's: what do you want to have as your tool? From SEO tracking, to running audits and identifying all the pages on your site and your sitemap. Are all the boxes checked, from titles to meta descriptions to internal linking? Is your website now submitted to Google? Is your website submitted to Bing? There are people that use Bing as a search engine still. There are a lot of people that won't change their default search engine when they get a new PC, and that's Microsoft's doing.

All of the things that I wish — if I could have a team and budget was never a concern, and I could just hire all the experts to come in and own every single piece of the marketing engine — that's what I have. That's what I built for myself.

CRM, pipeline, email, and the paid media roadmap

Rafael: I've been working with different folks, especially in the service industry, to deliver that and set them up with that kind of website and basically lead capture engine. And it handles email marketing as well.

I don't want to position it as the end all be all, but I'm building it in a way that could replace a lot of expensive subscriptions from HubSpot. It has a CRM. It tracks contacts. It tracks deals. It allows you to see your pipeline and have that follow-up journey via email.

If you want to deploy paid social campaigns specifically, we have a couple of other things in the roadmap to be able to deploy YouTube and paid search campaigns. And I want to bring in ChatGPT ads — although I'm not super convinced of ChatGPT ads, because they're only showing to the free tier account. So I'm not sure if that's the kind of audience I want to reach with my product anyway.

But all these different integrations, to have a proper team — and you don't have to build it. You don't have to learn AI. You just have to work with me, and I deploy that to your operations.

Ysi: That is exactly what a lot of people need. It's a great solution, and as I said, it's working. It is working. So it is amazing to hear. I'm very, very glad that we had this conversation today.

The question from the last guest

Ysi: We do have a tradition on the show where a previous guest left a question for you, and you're going to leave one for the next guest who is going to be in your seat.

The question they left for you — and I'm going to position it in a way, like, you've developed this software, you developed this solution. The question they left was: is there any process that you didn't just automate into the software, for example, but that you had to completely reinvent? Was there any of those workflows or automations where you said, this is not the right process, we need to scrap that and start again?

Agentic versus deterministic workflows

Rafael: Yeah. I realized that there are certain workflows that are either going to be agentic — where you're going to tap into that logic power, or harness that logic power on the models, to be able to make decisions and just figure it out. Like, there's an objective, I need this, get it done. Agentic workflows are meant for that.

But then there are certain other processes where we don't need logic. We just need a structured data set to come in, and everything is going to be deterministic. We know the steps that need to happen every single time, and as long as the data is structured in the proper way and things are not lost and everything is mapped correctly, our success rate is going to be close to 100% every single time.

It was the data structure, not the process

Rafael: So I think more than changing the process, it was changing our backend data structure. As far as how requests are coming in, and how they're mapped to the database, so that once that trigger happens everything is mapped and all the ingredients to make that output happen are where they need to be, labeled how they need to be labeled, and all the data is available.

That was part of the learning curve. At some point there was a human handling X process for content creation, or content publishing, or distribution, or scheduling. They have their process, they know the things that are not 100% where they need to be, and they are able to adjust and make decisions and figure out the way to get it done.

The workaround that turns 15 minutes into 25

Rafael: Which is fine, but oftentimes having to go around some of those corners will just make that task go from a 10, 15 minute task to now you're taking 25 minutes, because the thing is not set up properly.

So when deploying these deterministic workflows, or even for the agentic workflows that we have, it's about ensuring that the data is mapped correctly and everything is flowing properly, and that there's clear structure across the board from the very beginning of that process. If it's a process that takes five steps, great. If it's a process that takes 30 steps, it adds a little bit more pressure, because now there are more opportunities for things to break. So just ensuring that there are those safety nets, and again, that data integrity is there. I think that's super important.

Tearing it out and rebuilding it mapped correctly

Rafael: It just forced us to have to audit and almost retrain that muscle memory we had, where we knew this thing is not perfect, but let's just make sure that manually we're adding that information into the address field.

But now we're breaking everything up. We're tearing everything out. And now we're rebuilding it — and we had to rebuild it to be mapped correctly, and have all that information flowing correctly. That was a big chunk of the work when it came to building.

So I'm excited that it's there, and that's a lesson for the next project. I'm not even going to spend time before actually getting that mapping done. That just comes down to what we were talking about earlier: there's no way to learn these things without having to go through those trials and tribulations, essentially.

Ysi: I couldn't have asked for a better answer. I have a soft spot for data governance and process mapping. It's so important, and it makes life so much easier. It is the base.

The question he leaves for the next guest

Ysi: What would you leave as a question for the next guest that is going to be sitting in your seat? And to give you a little bit of context: the next person coming here runs a massive department. They move a lot of volume of data, and they are very innovation forward. I'll leave it at that. You can ask one or two questions.

Rafael: I think — is there a rubric for what your AI agent should handle and what should be handed off to an AI agent, versus something that should just stay with your human team? What is that rubric that helps make that decision?

And I ask that question because there's so much conversation right now with shifting everything towards an AI workflow, whether it's a deterministic or agentic workflow. Oftentimes I've been in that bundle where I would like everything to be automated and everything to just be running — if I don't have to be concerned about it, I would love for that to be reality.

But where does that leave the human impact? Depending on what your industry is, there's a lot of service, and customer service that needs to happen, and overall strategy, and just taste, that cannot be handed off to an AI agent.

But from a decision-making standpoint, and before you spend a ton of time trying to build something that should just stay with a human — like, you should not hand this off to any kind of automation workflow or AI agent, or whatever you're labeling your stack — what is that rubric to make that decision? I think that's super insightful, and I would love to hear that answer as well.

Ysi: I cannot wait to ask that question. Thank you so much, Rafael, for today. It has been my pleasure to hear your story, and to hear how we all connect, and that it doesn't matter the industry — forward deployed operations fits anywhere. Thank you again.

That's it for this one. Follow the show so the next episode finds you, and the newsletter goes out every Tuesday — the link is below. See you next time on The Forward Deployed Operator.