When Álvaro Martínez Iges first got access to GPT-3 in 2022, he had to apply as a researcher. He wasn't doing cutting-edge NLP work — he was a product leader who had just left Yoopies, and something in the API's raw potential wouldn't let him sleep. A few months later, he built a WhatsApp prototype called Chattify, handed it to his mother, and within weeks it had gone viral across Spain. That product became Lucía — today one of the most widely downloaded consumer AI assistants in the world, with more than 80 million installs across its app and WhatsApp integration.
In this episode of the Nova Podcast, host Ramón Rodrigáñez sits down with Álvaro to unpack everything: the founding story, the technology bets that paid off, the competitive pressures from OpenAI and Google, and a vision of the future that deliberately makes AI invisible.
The Two Bets That Started Everything
Before writing a line of product code, Álvaro made two foundational hypotheses about the market — hypotheses that were contrarian in early 2023 but now look obvious in hindsight.
The first: foundation models will commoditise. While the industry was marvelling at GPT-4, Álvaro was already betting that competing on model quality would be a race to the bottom. Anyone without Altman-level capital shouldn't even try. The second: in a commoditised model world, what wins is brand and user experience. If the underlying intelligence becomes a utility, the layer that users actually touch — and trust — becomes the moat.
"We always said we weren't going to develop foundation models. The hypothesis was simple: the models will commoditise, and what will have enormous value is being easy to use and having a brand users recognise."
— Álvaro Martínez Iges, Co-Founder, Lucía
This positioning — what the industry calls the application layer — let Lucía move fast without a billion-dollar war chest. Where competitors like Inflection AI spent hundreds of millions training proprietary models, Lucía focused on the interface, the personality, and the distribution.
The Technical Edge Nobody Talks About
Lucía is often pitched as a consumer brand story. But Álvaro admits he hasn't given enough visibility to the genuine technical innovation underneath. Three areas stand out.
Model routing. From early on, Lucía built the ability to swap between any AI provider in minutes — not hours, not days. What started as a necessity (when OpenAI's servers were overloaded, Lucía switched to Meta models in real time) became a sophisticated cost-optimisation system. A simple "hello" is routed to a cheap model; a second-order differential equation triggers a reasoning-heavy one. OpenAI launched something similar this year under the name "automatic model selection" — Lucía had been doing it for two years.
Tone of voice and personality. Lucía has a distinct character built through intensive prompt engineering and fine-tuning work. Users can further personalise their experience through what the team calls a "bestie" — a custom persona. This emotional layer, what Mustafa Suleiman (now CEO of Microsoft AI) once called "EQ", is increasingly what separates useful AI products from forgettable ones. Lucía's approach to this predates it becoming an industry talking point.
Persistent memory and personalisation. Over time, Lucía builds a contextual model of each user — preferences, communication style, past topics. This accumulated context makes responses progressively better, and raises the switching cost for users who've been with the product long term. It's the same dynamic that makes it hard to leave an iPhone ecosystem: everything is already there.
Going Viral Without a Marketing Budget
Lucía's growth story is a masterclass in scrappy, high-leverage distribution. The early strategy centred on one insight: for a B2C product aimed at a general audience, mainstream media coverage is worth more than performance marketing.
In Spain, Álvaro personally traded podcast appearances for editorial coverage — recording five-hour YouTube video sessions in exchange for a three-minute review. The inflection point came when Lucía was featured on Antena 3's flagship news programme (Spain's primetime equivalent of the BBC News at Ten). Downloads spiked overnight. El Hormiguero — one of Spain's biggest TV shows — covered Lucía without naming it, accidentally creating a mystery that drove its own search wave.
The same playbook was then adapted for Brazil. The team localised the product properly (not just translating the app, but "dressing Lucía as a Carioca" — adapting slang, cultural references, and persona to feel genuinely Brazilian), hired a local PR agency for $5,000, and secured coverage on Globo TV within a week of launch. Brazil is now Lucía's largest market and, by Álvaro's estimate, represents around 80% of the company's current valuation.
"We thought we were available everywhere because Lucía spoke every language. But the website wasn't in their language, the personality wasn't local. For a general audience, that localisation was everything."
— Álvaro Martínez Iges
The Competitive Threat: Dumping, Not Technology
Lucía has already outlasted several better-funded competitors. Inflection AI — whose CEO had told Álvaro "I have 100 times more resources than you" — has since pivoted. Character.AI, which invested heavily in proprietary models and flashier features, generated fewer downloads. On pure product and distribution, Lucía punched well above its weight.
But the next wave of competition is different in kind. OpenAI and Google aren't just building better products — they're building distribution at a scale that makes free features economically trivial for them. If Google's Gemini decides to give away deep research to every Android user (Android has 80% market share in Latin America), Lucía can technically build the same feature but can't absorb the inference cost at scale without charging for it.
Álvaro calls this existential risk not a technology problem, but an economic one — akin to dumping. The mitigation strategy is to go deeper on localisation, to solve problems that OpenAI won't "go down to the mud" to address, and to build user relationships strong enough to survive the commoditisation of features.
Key Takeaways from This Episode
- Bet on the application layer. Building brand and UX on top of commoditising models is a defensible strategy — and has been validated by the market.
- Model routing is a real competitive advantage. Dynamically selecting the right model for query complexity reduces costs and improves quality simultaneously.
- Localisation is non-negotiable for mass-market AI. Real localisation — cultural, linguistic, and tonal — is what unlocks general audiences outside the early adopter bubble.
- PR beats paid acquisition at the start. Mainstream media coverage, traded for time and creativity, drove more growth than performance marketing ever could have.
- The monetisation frontier is transactions, not subscriptions. Conversational commerce — where users buy things through an AI assistant — represents the most durable revenue model for consumer AI.
- AI advertising works, but the format must evolve. CTRs in conversational AI ads are exceptional, but the industry is still in its "banner ad era" — the right format hasn't been invented yet.
Monetisation: From Banner Ads to Conversational Commerce
Lucía began experimenting with monetisation in early 2024 — by Álvaro's account, among the first AI consumer products to do so. The core insight is that users are present at every stage of the commercial funnel within a single conversation: discovery ("I want to start running, what do you recommend?"), comparison ("I pronate — which shoes work for me?"), and intent ("I want these specific trainers — buy them for me").
This makes Lucía a uniquely powerful advertising surface, provided the ads are contextually matched and clearly labelled. The CTRs, Álvaro notes, have been extraordinary — at launch they were high enough to crash advertiser websites. In Brazil, Lucía users can now complete purchases directly within the conversation, opening up affiliate and transaction-based revenue that didn't exist a year ago.
A paid tier also exists — not primarily for revenue, but to fund development of expensive features (like voice-to-voice conversation) that would otherwise be impossible to build and test at scale. If inference for free users costs fractions of a cent per query, voice costs orders of magnitude more. The paid plan creates a testbed.
The broader monetisation thesis tracks closely with the evolution of the web: just as banner ads gave way to search intent, and search intent to social targeting, conversational AI will develop its own native ad format. OpenAI's recently launched "sponsored conversation" format — where users can talk directly with an advertiser's AI agent — is an early glimpse of where this is going.
The "Boring AI" Vision
Álvaro's 5-year prediction for Lucía is deceptively simple: he wants it to be omnipresent but invisible. Not a chatbot you open when you need it — something that handles tasks in the background before you even think to ask.
He draws the analogy to electricity: when power became ubiquitous, we didn't stay with one light bulb. It permeated everything — devices, transport, communication. AI, he argues, will do the same. The products that win won't be the most impressive demos; they'll be the ones that quietly make everyday life better without demanding attention.
"My vision in five years is that Lucía is omnipresent in your life — helping you with things you don't even realise she's helping with. And the result is that the lives of the people we help are genuinely better."
— Álvaro Martínez Iges
This philosophy shapes product decisions today. Lucía has introduced "Tools" alongside its chat interface — dedicated modes for maths, for example — because four times more users reach for the dedicated tool than solve problems via open chat, simply because it's clearer what it does. The future of AI interaction isn't one long chat thread; it's ambient assistance that surfaces the right capability at the right moment.
Álvaro is fond of an example from his own home life: he built a small Claude Code agent over a weekend that automatically checks his shared family calendar before accepting any out-of-hours commitments. It runs silently. No interface. No notification. Just fewer arguments about who's picking up the kids. That, he says, is the product direction.
On Engineers, AGI, and First Principles
The conversation closes with two broader themes that Nova readers will find especially relevant.
On the future of software engineering: Álvaro estimates that over 90% of the lines of code written at Lucía last week were written by AI (primarily Claude Code). He's emphatic that this doesn't mean fewer engineering jobs — it means engineering becomes more strategic. The engineer's role shifts from writing code to designing architecture, managing AI-generated output, and identifying the right problems to solve. The barbell effect he describes: average engineers who adapt will become top performers; top performers will become exceptional; the tail that refuses to adapt will be left behind.
On AGI: Álvaro takes the pragmatist view that AGI — by its original Turing Test definition — has already arrived. What matters isn't the label; it's the rate of change. He shares an anecdote from a founder retreat where Bill Gates, asked what advice he'd give young people today, admitted for the first time in his life that he genuinely didn't know. Álvaro's interpretation: the pace of change has accelerated so much that forecasting is genuinely hard — but first principles still hold. Human agency and taste, he argues, are two qualities that AI won't replace and that we should be teaching children today.
For anyone building, investing in, or working alongside AI products, this episode is one of the most grounded and honest conversations available on what the space actually looks like from inside the arena — not the keynote stage.
Are you top talent building something meaningful?
Nova connects the world's most ambitious professionals — founders, operators, and investors — across 80+ countries.
