Quote · All-In with Chamath, Jason, Sacks & Friedberg
Former Intel CEO on What Went Wrong, What's Next + Lovable CEO on the Real Promise of Vibe Coding
Where this was said
How Lovable is Bringing Down Builder Costs
At 36:26 · chapter starts 33:38
With competitors and foundation model labs threatening to make Lovable obsolete every six months, Osika walks through the strategy that has kept the company growing instead. Lovable routes every request to whichever model — commercial frontier or its own fine-tuned open-weight — is most suitable for that task. A Stockholm research team applies reinforcement learning specifically to mistakes frontier models make inside Lovable's agent harness, using the enormous signal from a million weekly projects to improve continuously. Critically, the team has never made a decision to use a cheaper model when it measurably performs worse for customers — margin optimization never overrides product quality. Jason floats the question of whether Lovable is profitable; Osika is careful, noting they monitor margins closely but prioritize intelligence investment. About 60% of lowest-tier subscribers hit their caps and top up — a sign of deep product-market fit. The episode closes with a philosophical exchange about parallel experimentation: when Jason admits his team built two separate intranets — one for the US and one for Japan — Osika draws on his time at CERN, where isolated teams work on the same particle accelerator independently to avoid anchoring bias, only sharing results at publication. Now that building costs approach zero, running duplicate software experiments is not waste — it's the optimal way to avoid local minima and find the best product.
Lovable is evolving beyond software creation into full business operations. In pre-release, users can access an AI co-founder that monitors their business overnight and delivers strategic recommendations each morning. With all your apps running on the platform, Lovable has access to all the data — usage, revenue, customers — to recommend optimizations without being asked. The product moat just got a lot deeper.
When a US nursing education company used Lovable to build custom scheduling, certification management, and admin tools, they didn't just save money — they replaced 10 separate software subscriptions and now save over $1 million per year. As building costs approach zero, every company will evaluate whether Salesforce, Slack, and HubSpot are still worth their price — or whether a bespoke alternative built in days beats them on fit and cost.
A company called Narsa replaced more than 10 internal tools with bespoke Lovable-built applications, saving over $1 million per year in software costs.
Lovable reached $500 million in annualized revenue by May 2026, just 20 months after launch, growing by roughly $100M every 6 months.
Lovable uses multiple frontier models and increasingly its own fine-tuned open-weight models, routing each request to whatever is most suitable. The team explicitly refuses to optimize for cost by using measurably worse models. A research team in Stockholm does post-training using reinforcement learning on the specific mistakes frontier models make inside Lovable's agent harness. The data flywheel from a million weekly projects is the core competitive moat.
