A top early-stage VC who backed a $6.6 billion AI coding unicorn is now telling founders to stay away from the category entirely.
The Summary
- Jussi Salovaara, managing partner at Antler Asia overseeing $72M, says he won't invest in new vibe-coding startups despite his firm's Europe fund backing Lovable at early stage
- The irony: He watches his own portfolio founders use AI coding tools daily, but sees zero defensible moats for new entrants
- His litmus test: "Anthropic ships your product next week, what happens to you?"
- Instead, he's betting on founders with deep domain expertise who can't be commoditized by the next model update
The Signal
Salovaara's position is notable because he's not an AI skeptic. Antler's Europe fund was an early investor in Lovable, the Swedish vibe-coding company now valued at $6.6 billion. He sees his colleagues and portfolio founders using AI coding tools every day to ship faster. He knows the category works.
But he also knows the window has closed. When a VC says "the guys are already out there," he means the market has crowned its winners and the marginal returns on backing number 47 in the category have gone negative.
"It's been fantastic, but starting new vibe-coding companies right now feels like it doesn't make sense anymore."
His framework is simple: How does a startup defend against platform risk? The questions he asks founders cut to the core of what makes AI-native companies fragile. What happens when Anthropic or OpenAI ships a competing feature in their next API update? What happens when model costs spike 5x because compute gets tight? What happens when a million people clone your wrapper?
These aren't hypothetical risks. They're the actual conditions of building on rented land. Vibe-coding startups are infrastructure plays built on someone else's infrastructure. The differentiation window is measured in weeks, not years.
Key vulnerabilities facing new AI coding startups:
- Platform collapse: Foundation model providers can replicate features overnight
- Commodification: Low technical moats mean fierce competition on price and speed
- Cost volatility: Razor-thin margins evaporate if compute costs spike
Salovaara's pivot is toward domain expertise. He wants founders who know insurance, logistics, healthcare, construction so deeply that their AI implementation becomes defensible through workflow integration, regulatory knowledge, or customer lock-in that transcends the underlying model.
He also sees consolidation coming. The existing AI coding companies, Lovable, Cursor, Emergent, aren't going to let dozens of competitors chip away at their lead. They'll acquire smaller players to expand feature sets or eliminate competition.
The Implication
If you're building an AI coding tool in 2026, you're either late or you're building the wrong thing. The real opportunity isn't in making coding easier. It's in making domain-specific problems solvable by people who couldn't code before.
The agent economy doesn't need more wrappers around Claude or GPT-5. It needs founders who understand underwriting, supply chain optimization, or clinical trials well enough that their AI tools encode expertise, not just autocomplete. Salovaara's bet is that defensibility comes from knowing the problem space better than the model does.