He Started With an Open-Source Project — 15 People, $10M ARR in 60 Days, $6.6B Valuation in 18 Months

Process

The Beginning: A Swedish Engineer's Open-Source Experiment

In 2023, Anton Osika sat in his Stockholm apartment and started what looked like an ordinary open-source project. He gave it a name so straightforward it was almost clunky: GPT-Engineer. The function was simple: users describe an app in natural language, AI generates the complete code.

This wasn't an original idea. 2023 was the year AI code generation exploded — GitHub Copilot, Cursor, Replit, and dozens of competitors were already fighting in this lane. But Anton made one decision that changed the entire competitive landscape: he made GPT-Engineer fully open-source. Not "partially open." Not "freemium." Not "free trial then pay." Completely open — anyone could download, modify, and use it.

The GitHub reaction exceeded everyone's expectations. Fifty-two thousand stars poured in. GPT-Engineer became one of the most-watched open-source AI projects of 2023. More importantly, those 52K stars weren't vanity metrics — they represented 52,000 early users actively testing the product, filing bug reports, and contributing improvement suggestions. Anton wasn't building a product alone — he had a free, full-time QA team.

Founder Backgrounds

Anton Osika (CEO), born 1990 in Sweden. BSc in Physics from Hong Kong University of Science and Technology, MSc in Engineering Physics from KTH Royal Institute of Technology. His career spans hardcore technical domains — he worked on CERN's ATLAS experiment, was an early engineer at Sana Labs ($500M valuation AI company), and co-founded Depict.ai (ML for e-commerce recommendations) as CTO. In 2023, his open-source GPT-Engineer project went viral. In 2025, he signed the Founders Pledge, committing to donate 50% of his net worth to meaningful causes.

Fabian Hedin (CTO), also Swedish. The two met in Stockholm's tech community. Hedin led the technical re-architecture of GPT-Engineer from an open-source CLI tool into Lovable's commercial visual platform. In 2025, he shared the KTH Innovation Award with Osika (judged by Spotify founder Daniel Ek).

Their combination is precisely complementary: Osika is the product visionary who knows what to build (validated by his track record at CERN, Sana Labs, and Depict.ai), while Hedin is the engineering architect who knows how to build it. This division of labor allowed them to achieve, with 15 people, what traditional companies couldn't do with 150.

Phase 1: From Open Source to Commercialization — GPT-Engineer Becomes Lovable

The open-source project gave Anton two of the scarcest startup resources: validated demand and a loyal early user base. When 52,000 developers voluntarily download and use your tool, you know they genuinely need it. But open-source has limits — you can only reach developers, not the vast population of people with ideas who can't write a line of code.

Anton realized that to turn this into a real business, he needed a commercial version — and a partner who could bring it to market. He found Fabian Hedin, another Swedish tech entrepreneur, who joined as CTO. Together they re-architected GPT-Engineer, adding a visual interface, one-click deployment, automatic error fixing — transforming it from a CLI tool for programmers into an AI app builder anyone could use.

In December 2024, they rebranded to Lovable and opened to the public. The name carried the product's core promise — "lovable," meaning the tool is so user-friendly it inspires genuine affection.

Phase 2: $10M ARR in 60 Days — With Only 15 People

Lovable's post-launch growth numbers silenced the entire SaaS industry.

First 4 weeks: $4M ARR. First 60 days: $10M ARR. Team size: 15 people.

Quick math: each employee contributed ~$667K in ARR. A typical mature SaaS company considers $100K-$200K per employee efficient. Lovable was 3-7× more capital-efficient than industry benchmarks. Not because they had superpowers — because the product itself was the growth engine. Every app built with Lovable became a living advertisement for Lovable. Every website or application generated by the platform could carry a "Built with Lovable" badge — a free, infinitely expanding acquisition channel.

More crucially, Lovable captured the cultural momentum of "vibe coding" — one of tech's hottest concepts in 2025, referring to the practice of describing intent to AI in natural language and letting AI generate the code. Lovable embedded itself into the definition of this concept: you don't need to know how to code. You just need to describe what you want. This positioning expanded their addressable market from tens of millions of developers to billions of people who can use language.

Phase 3: The Technical Moat — AI That Unsticks Itself

Lovable has one feature that stuns everyone who uses it: after generating code, the AI can find its own bugs, fix them, and redeploy — all without human intervention. Anton calls this capability "AI unsticking itself."

The traditional AI code generation workflow: user gives instruction → AI generates code → code throws errors → user debugs manually → user tells AI what's wrong → AI regenerates. The slowest step in this loop is "user debugs manually" — because the user might not be a programmer and can't even read the error messages. Lovable shattered this loop: AI auto-runs tests after generating code, auto-analyzes errors, auto-modifies code, and auto-redeploys. The user might perceive none of this — they just see "Building..." and then "Done, it's ready."

The significance of this technical breakthrough cannot be overstated. It transforms AI coding from "AI-assisted development" to "AI-autonomous development." The user shifts from "the person writing code" to "the person providing product requirements" — two fundamentally different types of people, with the latter being at least 100× more numerous.

Phase 4: Funding and Valuation — The Rocket from $1.8B to $6.6B

Lovable's fundraising velocity matched its revenue trajectory.

February 2025 — just over two months after launch: Series A, $200M at a $1.8B valuation, led by Accel. At this point Lovable's ARR was roughly $10M-$15M — meaning investors were willing to bet at over 100× ARR. In traditional SaaS investing logic, this is insane (10-15× ARR is normal). But in this AI wave, Accel was betting that Lovable could become the next billion-user product.

The bet paid off. By November 2025, Lovable's ARR reached $200M with nearly 8 million users. December 2025: Series B, $330M at a $6.6B valuation. CapitalG (Google's investment arm) and Menlo Ventures led the round, with participation from Khosla Ventures, Salesforce Ventures, and Databricks Ventures. Anton Osika — a Swedish engineer who was posting open-source projects on GitHub just two years earlier — became one of Europe's youngest self-made billionaires in history.

More notably, Lovable shattered the narrative that "Europe can't produce real AI companies." Before Lovable, virtually all AI unicorns clustered in Silicon Valley. Lovable proved: in the era of open-source + AI, geography is no longer a moat — the community is wherever the company is.

Phase 5: The Lovable Doctrine — Knowing What to Build Matters More Than Knowing How

In his Lenny's Newsletter interview, Anton said something that got widely quoted: "The biggest bottleneck is shifting from who can build to who knows what to build."

This is the core of Lovable's business philosophy. When AI drives the cost of code generation toward zero, scarcity is no longer technical ability — it's product taste, user understanding, and the ability to define problems clearly. A product manager who can't code but deeply understands user pain points is more valuable in this era than an engineer who can code but doesn't understand users.

Anton's advice for anyone wanting to succeed in this era is simple: "Spend a full week using AI tools to solve a real problem end-to-end. Become a top 1% user of AI tools." He's not giving motivational fluff. He did exactly this — two years ago, used AI to build GPT-Engineer. Two years ago, used AI to build Lovable. Then watched the world reorganize around his product.

Sources: Wikipedia · Lenny's Newsletter · TechCrunch · Forbes · Lovable.dev

Thinking

Lovable's case isn't just another "AI startup success story." It's a strategic textbook on sequence, leverage, and positioning.

Layer 1: Open-source isn't charity — it's the highest-ROI market validation. Anton open-sourced GPT-Engineer fully. On the surface, "giving it to the community for free." In reality, he traded code for three things via 52,000 GitHub stars: ① Demand validation — if nobody downloads, the direction is wrong. ② Free QA team — 52,000 users testing, filing bugs, suggesting improvements. ③ Brand equity — when Lovable commercialized, those 52K stars were the most powerful credential possible. In the AI era, open-source community is the most underrated GTM strategy. It's not "free first, figure out monetization later." It's "build trust and user base first, sell a better paid version second."

Layer 2: 15 people, $10M ARR — not because the tech is magical, but because the product IS the growth engine. Every app generated by Lovable can carry a "Built with Lovable" badge — free, infinite, self-replicating acquisition. Traditional SaaS needs sales and marketing teams to "find" customers. Lovable lets customers become the acquisition channel through using the product. This isn't growth hacking — it's growth baked into the product's DNA.

Layer 3: Anton's "bottleneck shift" thesis is the most important business insight of this era. When AI drives the cost of "making" toward zero, scarcity shifts from "who can build" to "who knows what to build." This means product managers, designers, and domain experts who deeply understand user pain — their value is skyrocketing. Pure execution engineers see their value declining. Anton himself is the best example: he wasn't the most brilliant AI engineer (GPT-Engineer's code wasn't technically superior to competitors). He was simply earlier than others at understanding what this product should look like, how it should be priced, and how to sell it to non-programmers.

Action

If you want to replicate Lovable's path, here's a four-step actionable framework:

1. Validate with open-source before building a commercial product. Before spending months on a commercial SaaS, build a minimal open-source tool. Ship it on GitHub. If you can't get 1,000 stars in 6 months — pivot. If you get 10,000 stars — you have something. Open-source is the lowest-cost, highest-leverage market validation available today. Don't worry about copycats — if your product can be easily copied, it had no moat to begin with.

2. Bake growth into the product — don't rely on a marketing team. "Built with Lovable" wasn't an afterthought marketing tactic — it was architected into the product from day one. What auto-distribution mechanism can you embed in your product? A branded badge on free-tier output? Default-public generated content? Invite mechanics tied to core experience? Every time a user uses your product, they should also acquire customers for you.

3. Position for non-programmers, not programmers. One of Anton's smartest decisions was transforming Lovable from a developer CLI tool into a visual platform for anyone. The market of programmers is tens of millions. The market of people who can use language is billions. Who is your product for? If it's for "people with a specific skill," can you redesign it for "people who can speak"?

4. Make AI unstick itself — don't make users debug for you. If your product relies on AI-generated output, examine: what does the user have to do when the AI output isn't right? If they need to manually troubleshoot, describe problems, and re-prompt — your product experience has 90% optimization remaining. The goal: when AI fails, the user does absolutely nothing. The system auto-retries, auto-fixes, auto-optimizes. This experience gap is the distance between your product being "usable" and being "lovable."