The business challenge for AI-native applications

Inside what it looks like to build competitive AI solutions today

I went to Carnegie Mellon as a student, and recently returned to give a guest lecture on designing for AI. The talk covered key topics regarding AI’s biggest challenges in practice. One topic largely unknown for folks is the state of AI today.

This is a deep dive on the current AI competitive landscape, along with very tactical, fundamental business & design challenges for creating potentially industry-defining AI products. Here, I break this all down in a simple & approachable way.

For context, read more about why AI projects fail, and competitive advantages of AI.

What are AI-native applications?

ElevenLabs, Hebbia, Cursor, Sierra, Harvey, Luma, Typeface… are AI-native applications. All are examples where AI is integrated into the core of their products, and fundamentally built into the company.

Some companies are AI model providers. Examples include OpenAI, Anthropic, Google DeepMind, Mistral. They develop, train, and offer access to their own custom-built pre-trained models, APIs, etc. But they are also looking for other companies to use (buy) their AI models and turn them into actually useful products for customers.

Why? AI models are NOT generally useful by themselves. Think GPT-4o, o3-mini, and others. They require integration into…