The native AI wave
Building & scaling AI-first companies
The native AI wave
Building & scaling AI-first companies
A remarkable panel featuring leaders from Mistral AI, Replit, and DeepJudge revealed how native AI companies are rewriting the rules of software company building. The discussion, highlighting companies reaching 10x the growth rates of traditional SaaS, offered crucial insights for both insurgents and incumbents, from the lens of an LLM, a horizontal AI application and a vertical AI provider:
Speed and scale redefined
The numbers defy conventional wisdom. According to Stripe data, the fastest-growing native AI companies reach $30 million ARR, five times faster than comparable SaaS companies. Replit exemplifies this trajectory - Michele Catasta revealed they've built 4 million applications through their AI agent in just nine months, with 200,000 deployed in production.
"We are still to fill the gap of 1 billion [users], but this technology has been around not for long," Catasta explained, highlighting how AI democratises software creation. Their agent can build full-stack applications from natural language, turning English into "the next coding language."
The key differentiators: customer value over technology
Surprisingly, the panel largely dismissed traditional technical moats. "I fundamentally don't believe in moats, especially technical ones," Catasta stated boldly. Instead, competitive advantage comes from rapid execution and creating customer value through comprehensive, convenient workflows.
Mistral AI's Marjorie Janiewicz emphasised that differentiation comes from customisation and control: If everybody is using the same models, then how are you going to create your own differentiation? Janiewicz explained that Mistral AI empowers organisations to tailor AI to their specific domains, combining curated data with deep process understanding.
DeepJudge's Yannic Kilcher reinforced this, noting that as each new generation of foundation models emerges, it "immediately erased a whole bunch of moats that existed before." The key is understanding and serving your specific user base better than anyone else.
Organisational DNA for the AI era
These companies share striking organisational characteristics. Replit operates with just 70 employees, two-thirds in engineering, shipping features almost daily. They've achieved over $1 million ARR per employee - a benchmark Catasta aims to double within the year. The management challenge has intensified rather than diminished. "Every engineer has way more responsibility, has much higher throughput than they had in the past," Catasta noted. Traditional 12-month roadmaps are impossible; Castasta operates on a 3-month horizon, with the team having a 6 week roadmap ahead.
Pricing models in flux
The panel revealed a fundamental rethinking of how AI products should be priced. Janiewicz highlighted the future opportunity: "As we power more and more agents that are delivering great value for specific use cases, and those agents have different levels of specialties, there is an opportunity to reinvent prices based on outcome. And we are thinking quite a bit about that right now." This shift from per-token pricing - to value-based pricing represents a major evolution in AI business models.
Yet the path to new pricing models is fraught with challenges. Catasta candidly shared Replit's journey: "We started with the obvious hypothesis. It will all be consumption outcome based." But reality intervened - customers found complex calculators confusing and bills unpredictable. The pragmatic solution? "It's important to make people understand that compute and intelligence comes at a price, and you will be charged accordingly." Replit now uses dynamic usage-based pricing, betting that product value will prove itself -users see what they can build compared to hiring traditional developers.
The incumbent relationship dance
Native AI companies view incumbents as both partners and competitors. Catasta shared a revealing anecdote, hearing a competitor on a recent podcast that has an alternative to Replit, yet ‘spent the weekend building with Replit.' This paradox - where the CEO of a competing company publicly endorses their product - illustrates the complex dynamics at play. "The fact that he goes in public and he talks about my product first of all makes me proud... but also makes me believe that all of us can find a way to specialise on a certain vertical."
Mistral AI's approach demonstrates strategic pragmatism in navigating these relationships. Janiewicz explained: "Our strategy has been to really expand our distribution channel as fast as possible. So today we're working with the hyperscalers as an example. In the end they can extend our distribution channel. They are a partner of ours, and it's helping us expand our reach." She acknowledged the delicate balance: "It is true that foundation models are evolving towards systems and product companies. So the line is starting to get blurred on many fronts."
Looking forward
The panel's message to incumbents was clear: fast-following has become easier than innovating, but requires unprecedented organisational agility. For entrepreneurs, the opportunity remains vast - these native AI companies are creating entirely new categories while operating at previously impossible efficiency levels. The native AI wave isn't just about technology; it's about reimagining how software companies are built, scaled, and operated in an intelligence-abundant world.
