From Edge to Core: The AI-Native Revolution

Rebuild around models: treat LLMs as programmable building blocks and redesign systems for probabilistic behavior.

Sep 29, 2025 6m 30s

Oji Udezu

Expert Insights

Oji Udezue underscores that transitioning to AI-infused software development isn't merely about integrating AI components; it demands a radical shift in architecture towards model-centric, probabilistic logic, a departure from deterministic thinking. He warns about the inherent risks but emphasizes the transformative potential.

Oji stresses that AI isn't just another API you can bolt onto legacy software. Instead, it's a new building material that requires rethinking the entire stack.

Hear him explain:

Why "sprinkling AI" into old codebases leads to failure.
How moving beyond deterministic rules to probabilistic models changes testing, pricing, and product design.
What it takes to build truly model-native applications like Formless, and why most AI startups miss this shift.

AI at the core is when you build with your model at the core, when like more than 50% of your code base is model based... Because what models are essentially is a large block of code you can direct." "It's like playing with a new tool. Imagine that something came out of the fourth dimension and someone said, play with this. You're like, what the hell? I can't even see it." "I realized very early on that prompts were code, right?

Monterail Team Analysis

Here’s how to win with AI in modern software:

Build model-native, not model-adjacent. Avoid "AI sprinkling" on legacy codebases.
Design for non-determinism. Evolve testing and QA to handle variable outputs.
Rethink architecture. Once models make up 50% or more of your code, traditional practices break down.
Personalize over standardize. Embrace variability to create tailored, human-like experiences.
Transition with intent. Chart a clear path between deterministic systems and AI-centric ones.
Plan for uncertainty. Adopt new strategies for pricing, deployment, and risk management.