Anton Osika: From Particle Physics to $6B AI Startup

Table of content

Anton Osika went from classifying particle collisions at CERN to building one of the most popular open-source AI coding tools on GitHub. His project gpt-engineer hit 55,000 stars—and became the foundation for Lovable, a startup now valued at $6.6 billion.

The Physicist Who Codes

Osika studied Engineering Physics at KTH Royal Institute of Technology in Stockholm, graduating top of his class (4.9/5.0 GPA). At CERN, he worked on triggers—algorithms that decide which observations from petabyte-scale detector data are worth keeping.

That pattern-recognition problem turned out to be surprisingly relevant. AI coding is fundamentally about deciding what matters from a prompt, then generating the right sequence of outputs.

gpt-engineer: The Open-Source Experiment

In 2023, Osika released gpt-engineer, a CLI tool that lets you describe software in plain English and watch AI write the code. The repo exploded—55,000+ stars, one of the most popular code generation projects ever.

The core insight was simple: don’t just generate code snippets. Generate entire applications. Let the AI handle boilerplate, dependencies, file structure. Let the human focus on describing what they want.

From the README:

gpt-engineer lets you specify software in natural language, sit back and watch as an AI writes and executes the code, then ask the AI to implement improvements.

Lovable: Taking It Commercial

Osika took the gpt-engineer learnings and built Lovable (formerly gptengineer.app), a no-code AI builder for apps and websites. The team includes serial founders, competitive programmers, and physicists—people who care about shipping fast.

The growth has been absurd:

$100M ARR in July 2025 (8 months after launch)
$200M ARR by November 2025
8 million users by late 2025
$6.6B valuation after raising $330M in December 2025

Osika projects $1B ARR within the next 12 months.

Staying in Stockholm

When Lovable started gaining traction, investors told Osika the same thing they tell every European founder: move to Silicon Valley if you want to succeed.

He didn’t. Speaking at Slush 2025, Osika explained why:

“It was tempting, but I really resisted that. I can sit here now and say, ‘Look, guys, you can build a global AI company from this country.’ There is more available talent if you have a strong mission.”

Instead of chasing Bay Area talent, Lovable “flipped the script”—recruiting people from Notion and Gusto to work in Stockholm. The slower European market pace actually helped them stay focused rather than getting caught up in hype cycles.

The Pre-Lovable Journey

Before gpt-engineer, Osika built a track record:

Sana Labs (2017-2020): First employee and engineer. Built personalized learning systems serving hundreds of thousands of learners globally. The company raised $80M+.

Depict.ai (2020-2023): Co-founded and served as CTO. Applied ML to e-commerce product discovery. Raised $20M, went through Y Combinator twice.

Stockholm AI (2016-present): Co-founded the community, organized events for three years during its critical growth phase.

The Philosophy

Osika frames his work around a single goal: democratizing software creation.

From his LinkedIn: “I believe that democratising the ability to create software is one of the highest levers to unlock more creativity and human potential.”

He’s also a member of Founders Pledge, committing a percentage of personal proceeds to charitable causes. His GitHub bio mentions passion for “humanity’s future” and “humans future co-existance with AI.”

Core Approach

Natural language as interface: Describe what you want in plain English. The AI handles implementation details.

Community-driven iteration: The Lovable community constantly pushes the product forward. “They’re breaking some kind of world record right now,” Osika said at Slush.

Full-stack generation: Don’t just generate code—generate complete, deployable applications.

Open source foundation: gpt-engineer remains MIT-licensed and hackable. Lovable is the commercial evolution, but the experimentation platform stays open.

For Practitioners

If you’re building AI-augmented workflows:

Start with open experiments: gpt-engineer was an experiment that proved the concept. Ship something minimal, see what resonates.
Community compounds: Lovable’s users drive improvement faster than any internal team could. Build for community feedback loops.
Don’t relocate by default: Being outside Silicon Valley can be a feature, not a bug. Different pace, different talent pools, less groupthink.
Physics training transfers: Pattern recognition, working with messy data, deciding what to keep—these skills port directly to AI product work.

Links