Innovate, Don't Optimize: Lovable's Growth Chief on Rewriting the Playbook for AI

Verna says Lovable launched in the third week of November 2024, though it existed earlier as GPT Engineer. It passed $200 million in annual recurring revenue before its first birthday, going from $100 million at the end of July to $200 million about four months later. The company has more than 8 million users who have tried the product and hundreds of thousands of paying subscribers.

She and host Lenny Rachitsky both stress that nobody should treat this as a normal yardstick. Part of the story is Lovable itself and part is a category that is moving unusually fast. In her view, hitting a million dollars in ARR in a year is still doing well.

On who is paying, Verna describes two big groups. Non-technical founders build apps from scratch, and some already monetize them, while employees build internal tools, prototypes and landing pages. Lovable monetizes the act of building. Many others use it for personal projects, such as a tutoring app that makes her son answer questions to earn screen time, or a game built to propose marriage.

On retention she is candid that she cannot share exact numbers, but says paid retention is on par with the B2B SaaS companies she has worked at, and net dollar retention is strong because builders buy more credits. The team is more focused on engagement retention than paid retention, and does not optimize for revenue per user at all. Revenue is an outcome of getting more people through the door.

In earlier roles Verna felt confident in about 80% of the patterns she brought, and she admits it started to feel like copy and paste. At Lovable the company was already growing before she arrived, and she sees something she cannot map to a known framework. It is a brand-new category that hit fast-moving waters almost instantly, which usually takes years or decades.

That speed comes with a downside: growth is only partly in the team's control. Demand is arriving and the job is to capture it, grease the wheels and avoid hitting a wall. She now thinks only 30-40% of her past learning transfers, mainly paid marketing, habitual retention and free-to-paid frameworks.

Optimizing existing user journeys is mostly not worth the time. With so many competitors launching vibe-coding products, she argues the way to stay ahead is reinventing the solution rather than tweaking the problem. Her time split flipped from roughly 5-10% innovation to 95% innovation and 5% optimization.

In practice the growth team ends up shipping product. It launched the Shopify integration so people can vibe code storefronts, built voice mode, and now works on agentic workflows and agent instructions. Verna says this work goes deeper into core product than a growth team would normally go.

Verna usually spends most of her time on activation, smoothing the path from setup to the aha moment to a habit loop. At Lovable the user simply prompts an agent, and that agent team thinks about the first generation day and night. That takes pressure off her, because activation is built into the company's DNA instead of being a growth-team project.

Rachitsky summarizes the shift: the impact comes from making the agent better, not micro-optimizing each step. Verna agrees. It does not matter whether it is a first or an nth generation, because an agent that understands intent and reasons better improves the whole lifecycle immediately.

She frames this as core product work. Previously people built deeper features or platform functionality, and now the core team is obsessed with the first experience, because that experience is the product.

One of Lovable's biggest strategies is building in public, paired with employee and founder socials. Verna says this is hard for large companies, but a small team has narrative control and enough trust to move fast. Big tier-one launches still happen, meant to step-change product-market fit, but the aim is to keep noise in the market by shipping every day or several times a day.

She says this noise doubles as a resurrection and re-engagement strategy. Users check social to see what Lovable shipped, and because the time from voicing an opinion to seeing it delivered is short, they feel heard. She calls it a retention approach she can back, and hopes it survives scaling.

Rachitsky compares it to Mirage, formerly Captions, whose CEO says they ship a marketable feature every week. Verna adds that velocity is the development team's top value, and engineers are treated as product engineers who must announce what they ship themselves. That means a little marketer in everyone, since a marketing team could not otherwise staff the pace.

Verna sees marketing changing on three fronts: channels, involvement in product and where marketers are hired. Positioning that once lasted years now lasts about three months, so product marketers focus on big narratives while product and engineering handle smaller launches. Five years ago she would have called SEO the organic strategy. Now she says it is social, meaning what the CEO, the team, creators and users post.

For founder and employee posting she names X and LinkedIn as the strongest channels for B2B. Posts cannot just be ChatGPT copy; they need personality and vulnerability, because in a crowded category people want to rally behind a team. The customer side is a word-of-mouth loop that only works if the product blows their socks off and makes them feel they have superpowers.

Lovable's name gives it an unfair advantage. Internally the rule is that if something is not lovable, it does not ship, and the fastest way to get a bug fixed is to say it is not lovable. Verna works with her designer on adding more love marks and distinctive interactions, and says brand work goes into the product itself since there is no brand marketing team yet.

Influencer marketing is ten times bigger for Lovable than paid social, which is expensive on payback period and still being optimized. Short videos work because nobody knows what vibe coding is, but after ten seconds of watching it they say it's new and want to try it. She adds that as the cost of building software falls, teams can invest in how it feels, so a designer should be among a startup's first hires.

Asked how Lovable avoids a Frankenstein product when everyone ships, Verna points first to who gets hired. She says the best talent is not defined by big-company logos but by passion, someone for whom the job is a hobby and the biggest opportunity of their life. High agency and high autonomy are also screened for: people who can launch something themselves without waiting for a specialist.

Supervision is light. There are shared goals and big launches, but much of the work is up to individual developers and marketers, and because velocity is so high a failed attempt is no big deal. Lovable is based in Stockholm and asks many people to relocate, though the hype around the product means candidates often say they would love to explore.

The company also uses paid work trials of a couple of days and probation periods, because the pace is not for everyone. Verna says that after ten days of vacation it felt like she had to onboard again. Lovable prioritizes people who do not look for clarity but can create it from chaos.

Rachitsky sums up the levers so far: build something lovable in a growing market. Verna adds that viability belongs to the 2010s and that minimum lovable product is the only thing that matters now. She also notes the feedback cycle has collapsed, since you can go from idea to a working product to user feedback within a day.

She gives a recent example: a product her team prototyped through vibe coding that took a couple of weeks, mostly to get videos and design right, and could have taken a day. The old path through user research, design sprints and engineering roadmaps is largely gone. Rachitsky says the LinkedIn launch looked like a full product, yet it was a prototype.

Verna's team includes a full-time vibe coder named Lazar, a former chief of staff who is self-taught and was an early user of the vibe-coding tools. He built Shopify Lovable templates and is teaching the team what the tool can do. She sees vibe coding becoming a skill on many job descriptions, and says such a person can be non-technical if they are really good.

Verna adds community to the list. Lovable's runs on Discord, has hundreds of thousands of members, community managers and an ambassador program, and it amplifies word of mouth, social posting and retention. Not everyone can build a community, but she suggests at least plugging into someone else's.

Giving the product away is another lever. Many traditional companies gate AI behind a paywall to protect margins, but Verna says a new category needs a free first wow moment. For Lovable that is the first preview after the first prompt, even though it is far from finished. Premium is just a baseline, and Lovable gives credits to every event and hackathon it can.

Internally, LLM costs for the free premium tier and giveaways are tracked as marketing cost and not as something to reduce for better margins. She admits AI companies sit around 40% margins, not 80-90%, and says treating AI as a cost center is when you are in trouble. If a user wants free credits for a hackathon at work, Lovable says take it, since that person is doing the marketing and activation for them.

She argues this does not wreck the economics. Lovable has about 100 people, up from 30 six months earlier, spends little on paid marketing (influencer spend is low double digits as a share of growth), has only a couple of sales staff and little demand-gen cost. Giving the product away is, in effect, a cheaper way to buy eyeballs than competing on Google ads, and VC money is for future development, not survival.

Verna says product-market fit used to evolve over years, with long blitz-scaling phases before the next horizon. Now the question returns every three months for two reasons. Model capabilities change with each new release, and consumer expectations change just as fast, so teams must bet on new functionality before the model catches up.

She calls it terrifying and confusing, since the team that finds product-market fit differs from the one that scales, and Lovable needs both at once. Every three months the company throttles scaling to reinvent, then scales again in short blitzes. Rachitsky points to OpenAI's recent code-red moment as evidence that even ChatGPT is not bulletproof, and Verna says she would once have called the idea of a $200M company losing fit in three months crazy.

She also worries about the adjacent-user idea. The core users are mostly pioneers excited by capabilities, and the latent majority is made of adjacent users. If everyone keeps chasing pioneers, the mainstream may never catch up, and the category could be seen as for techies or prototypes. She says she does not know the right answer.

Asked whether to join an AI company, Verna says they are hectic and unstable by nature, so the question is whether you can turn chaos into clarity. If so, you will absorb new skills fast, and she says she never used AI at Dropbox the way she does at Lovable. Her shorthand for AI is average intelligence that gives a base level, to which she adds human thinking and creativity.

If your strength is structure and deep specialty, she says not to join yet, since small AI companies demand generalists and can burn out the wrong personality. Rachitsky suggests there is no work-life balance there, and Verna pushes back. She has two kids and sets her own boundaries, and she says nobody anywhere really has balance.

She does not think in terms of balance but of in-the-moment choices, asking whether she would resent a decision later. That means sometimes telling CEO Anton she cannot make something because her child is sick, and protecting sleep, health, family and downtime. She says time off is part of doing her best work.

Inside Lovable, internal tools are built on Lovable, written specs come with a clickable prototype, and the company is planning a company-wide hackathon. Verna recreates a pricing page from a screenshot and sends it to engineers. She also uses ChatGPT's deep thinking mode for brainstorming, Granola for meeting summaries and Whisper Flow for voice input, and asks what AI can do first for every task.

She says building a prototype also finishes the ideation process: some ideas lose their magic once built, while others turn out to have more in them. It gives her a better way to communicate with engineers and calibrates her quickly.

Verna then turns to her post about women in tech. Reports show a gap in AI adoption between women and men, and she sees few women among AI CEOs or big acqui-hires. Despite Lovable's pink and purple branding, third-party data suggested at most 20% of its users are women. She stresses she is not blaming men; she worries many women are stuck in the latent majority.

Lovable's response is the She Builds initiative, a women-only hackathon with 48 hours of unlimited access. Participants often build hyper-local tools for elderly parents, kids, a church group or a basketball team, software that was never economical before. The next cohort starts December 15, with a larger one planned for International Women's Day.

Verna closes with hiring advice. She disputes the idea that new graduates have no jobs, saying AI-native graduates at Lovable are impressive and she learns from them, though she criticizes schools for not teaching AI. Ex-founders with high agency, she adds, are now in hot demand.