Our guest today is Romain Torres, co-founder of Arcads AI, a platform that turns text into hundreds of high-performing videos using AI-generated actors. In just a year, Arcads AI has skyrocketed to $7 million in annual recurring revenue, driven by viral user success stories β including one video that caught national TV attention and put the company on the map.
In this episode, Romain shares how Arcads AI is redefining the future of video creation β from scaling personalized storytelling with AI actors to competing in a space dominated by tech giants. We talk about building products that balance automation and creativity, how virality shaped their growth, and what it takes to stay ahead in the rapidly evolving world of AI video.
About Romain: LinkedIn | Twitter
Connect with Arcads: Webpage | LinkedIn
ABOUT ROCKETSHIP HQ: Website | LinkedIn | Newsletter | Youtube | Podcast Website
FULL TRANSCRIPT BELOW
Shamanth Rao
Iβm excited to welcome Romain Torres to Intelligent Artifice.
Welcome to the show, Romain.
Romain Torres
Thanks, Shamanth. Happy to be here.
Shamanth Rao
Yeah, Iβm thrilled to have you here β especially because weβve worked with your team for quite some time, since the early days of ARC Ads. Back then, the avatars were pretty janky compared to how photorealistic they are now β itβs kind of wild!
We actually worked together before Arcads, right? You were my client many years ago.
Romain Torres
Thatβs exactly right β good memory! I had actually forgotten that. And yes, before Arcads, we worked together.
Whatβs fascinating about ARC isnβt just the product β which is great β but also how the business operates. Itβs extremely agent-driven and highly automated. I find that aspect even more fascinating than the product itself.
The product is fantastic, donβt get me wrong β we use it ourselves β but your journey and the way you scaled are just as interesting.
Shamanth Rao
So, when you started Arcads, the original plan wasnβt to scale with AI agents. At what point did you realize automation was essential for scaling?
Romain Torres
The company was growing incredibly fast β it felt like piloting a rocket ship. We simply didnβt have the luxury to spend six weeks hiring for every new role. So building AI agents became both an obligation and a strategic decision.
Plus, by 2024β2025, things became possible that werenβt before β for example, automating customer support was nearly impossible before 2024.
Being in the AI industry, we saw new innovations emerge in real time, so we were quick to integrate them. Being an AI-native startup definitely helped.
Shamanth Rao
What I also find fascinating is that, as a founder, you actually built many of these agents yourself β you wrote the code, built the automations. How did you decide when it was worth your time as a CEO to build something versus focusing on traditional founder responsibilities like strategy or hiring?
Romain Torres
Iβve always been a builder at heart. Sometimes I have to remind myself that Iβm a founder, not just an engineer. But in moments of urgency, I just build things myself.
For example, one of our most successful internal AI uses is customer support automation. Initially, I handled all customer support myself β answering every message. Then one teammate took over full-time. But one day, a video created with our platform went viral on social media, and our support requests increased 10x overnight β from dozens to hundreds, even thousands daily.
Hiring a team overnight wasnβt possible, so we had to build an AI system that could handle at least 80% of support requests. Thatβs what we did β out of necessity. Now, AI handles almost everything. We still have one person overseeing support, but automation is the core.
Shamanth Rao
Thatβs incredible β you were essentially forced to automate because of growth. What was the very first agent you built?
Romain Torres
The customer support agent was one of the first. But another early one had a huge impact β our user growth agent.
As co-founder, Iβm responsible for growth. My co-founder focuses on the product β weβre essentially two co-CEOs. In growth, you typically test several channels β three to five β and double down on what works. Early on, one of our strongest channels was Meta Ads.
But ad success depends heavily on creative testing β the more creatives you test, the better. As a founder managing multiple priorities, I didnβt have time to manually brainstorm creative ideas. So I thought β why not build agents to do that?
These agents automatically scan competitorsβ ads, gather insights, and even generate new creative ideas based on whatβs working in the market. That saved hours every week and gave us a constant flow of new concepts to test.
Shamanth Rao
Thatβs brilliant β and a massive time saver. Could you explain how that creative research agent actually works under the hood?
Romain Torres
Sure. Think of ad creation as a three-step process:
- Generate ideas.
- Produce creatives.
- Test them in Meta Ads Manager.
The first step β ideation β usually involves researching competitors. Traditionally, marketers spend hours on Reddit or Facebook Ads Library studying what works.
Our agent automates that. It crawls the Facebook Ads Library, downloads competitorsβ ads, transcribes them, then feeds the text into an LLM like ChatGPT or Claude. The AI rephrases the ad copy to fit our productβs messaging and tone β giving us new ad scripts.
We then use those scripts to create videos with talking avatars on our platform.
Shamanth Rao
Thatβs fantastic. Weβve built something similar that scrapes Metaβs Ads Library and analyzes ad visuals with vision models. But if competitors have hundreds or thousands of ads, how do you decide which ones are performing well?
Romain Torres
Thatβs the challenge β you donβt know performance data. You can either try to infer it or just let volume do the work.
Initially, we tried to infer β we looked for recurring transcripts or identical ad copy across multiple uploads. If a competitor reused an ad 20 times, chances are it performed well. AI can identify those patterns.
But eventually, we shifted to a higher-volume approach β pushing as many ads as possible and letting Metaβs algorithm decide which perform best. More volume equals more data.
Shamanth Rao
That makes total sense. So youβd upload 50 creatives per ad set and let Meta allocate spend to the top performers?
Romain Torres
Exactly. I donβt manually budget each creative β I let Meta decide. Meta has the data and signals to optimize spend far better than we can. My job is to feed the machine great creative inputs.
Shamanth Rao
Got it. So with these agents, you can technically create thousands of ads. How do you decide when to stop?
Romain Torres
Ideally, you donβt stop β the more, the better. But in reality, constraints like budget and manpower limit us. Until recently, I was running ads myself while managing everything else, so Iβd manually cherry-pick creatives that felt promising.
Now that weβve hired a dedicated person for that, we can scale much further. But yes, humans tend to overestimate their ability to predict β the algorithm always knows better.
Shamanth Rao
Totally agree. Letβs talk about other automations youβve built β besides customer support and marketing.
Romain Torres
A big one solved communication gaps between our product and marketing teams. I handled customer calls β sometimes 15 a day β while my co-founder focused on design and development. I had all this customer insight in my head, but it wasnβt reaching him.
So, I built an agent that takes transcripts from AI note-takers during calls, summarizes them, and turns them into weekly insight reports for the product team.
It transformed how we share user feedback and prioritize features.
Shamanth Rao
Thatβs genius β leveraging call notes to bridge product and sales.
Now, when you say βagent,β some people argue thatβs just a fancy word for a script or Zapier workflow. How do you define it?
Romain Torres
Youβre right β what I call βagentsβ are often just workflows. Technically, an agent is autonomous β it can make decisions. A workflow follows a defined sequence.
But letβs be honest β βagentβ sounds way sexier than βworkflow.β It performs better on LinkedIn!
So yes, I use βagentβ as a marketing term. But functionally, most of what I build are AI-powered workflows.
Shamanth Rao
Thatβs fair β and itβs true that AI makes building workflows much easier now than it was two years ago.
Romain Torres
Exactly. Two years ago, I struggled with Zapier automations. Now, with AI, workflows can understand context and interact with humans.
Itβs like upgrading a dog that knows one command β βsitβ β to a human that can hold a conversation. The intelligence layer changes everything.
Shamanth Rao
Love that analogy. Were there any workflows that didnβt work out β any big failures?
Romain Torres
Yes β whenever you start building without a clear purpose, youβll fail.
People often ask me to help them βbuild an AI agentβ without a specific process in mind. That never works.
Automation succeeds only when youβre replacing a repetitive manual task with something measurable.
We also failed once trying to use Instagram Reelsβ view counts to predict high-performing Facebook ads β but Instagramβs API blocks that data. So sometimes, technical limitations get in the way.
Shamanth Rao
Absolutely. Iβve had similar experiences β failures mostly come from unclear goals. Documentation helps too. I even ask AI to take notes on what went wrong during builds for future reference.
Romain Torres
Thatβs smart β and yes, documenting errors and learnings helps a lot.
Shamanth Rao
Youβve mentioned that Arcads has an incredible revenue per employee. Was that intentional, or just a byproduct of using AI everywhere?
Romain Torres
A bit of both. AI accelerates everything. Adoption is faster than any previous tech wave β faster than the internet β because the infrastructure already exists. Everyone already has the tools and data.
At the same time, AI lets us automate instead of hire. Ten years ago, if you had a problem, your first question was βWho should I hire?β
Now itβs βCan I automate this?β
That dual effect β faster adoption and leaner operations β drives our high revenue per employee.
Shamanth Rao
That makes perfect sense. How do you foster an AI-first culture across your team so itβs not just you building everything?
Romain Torres
Culture starts with the founder. When people see me automate tasks myself, they naturally follow. Plus, our product is AI-driven, so AI is already part of everyoneβs daily conversations.
For more traditional companies, it might take structured training or bringing in experts. I actually believe in hiring consultants to teach these things β thatβs how I learned ads years ago, from people like you.
Shamanth Rao
I love that. Lastly β when you hire, what qualities do you look for in your team?
Romain Torres
One thing: people who go the extra mile.
There are three types of people β
- Those who donβt do what theyβre asked.
- Those who do exactly what theyβre asked.
- And those who do more than theyβre asked.
We only hire the third kind. Everything else can be taught β but that mindset canβt.
Shamanth Rao
Thatβs so true. And with AI evolving this fast, itβs crucial to have people who adapt and move quickly. Speaking of that, how do you think about long-term differentiation? AI moves so fast β new foundational models are released almost weekly by companies like OpenAI or Google. How does a company like Arcads stay ahead?
Romain Torres
Thatβs a great question. In AI, the only sustainable differentiation right now is speed. The first one to build something useful, to deliver value, wins.
If you look at ChatGPT β it wasnβt necessarily the best model at launch, but it was the fastest to market with a great interface and onboarding. Once users integrated it into their workflow, switching became harder.
For ARC Ads, speed and focus are everything. Weβre not a foundational model company β weβre an application layer. We aggregate models β Sora, Veo, our own custom ones β and make them usable for marketers and brands in one place.
Shamanth Rao
That makes sense β youβre effectively aggregating the best models for creative use cases.
Romain Torres
Exactly. For example, a company like Nike might need very specific ad assets that pull the best capabilities from multiple models. They donβt want ten different subscriptions β they want one platform that integrates all of it. Thatβs what we provide.
Our differentiation isnβt just technology β itβs speed, usability, and specialization.
Shamanth Rao
Thatβs powerful. And itβs clear how that positioning helps Arcads stand out. Before we wrap up, where can people find you and learn more about Arcads?
Romain Torres
You can check out our platform at arcads.ai and follow me on LinkedIn or Twitter (@romaintorres). I regularly share insights about using AI for ad creation β whether itβs for mobile apps, e-commerce, or performance marketing.
Shamanth Rao
Wonderful. Weβll link all of that in the show notes. Romain, thank you so much for joining us today on Intelligent Artifice.
Romain Torres
Thanks, Shamanth. It was a pleasure.
