[Global] Beyond AI Agent Demos: Rethinking AI Apps

The first AI agent demo is usually easy: give the model a task, connect a few tools, write a long system prompt, and watch it do something that looks close to work.

But turning that demo into real software is much harder.

Once an agent needs credentials, permissions, memory, persistent state, UI review, resumable workflows, testing, and distribution, the problem is no longer just about making the model smarter. It becomes a software architecture problem.

In this session, we will discuss why prompts, tools, and frameworks are not enough as the boundary layer for AI agents, and why a new app layer is emerging: one that is visible to humans, operable by agents, and structured enough for real work.

We will use Anna as a concrete example to explore this idea: how AI-native apps, tools, skills, memory, permissions, and execution can fit together inside an AI-native operating environment.

The goal is not to teach a framework step by step, but to share a product and architecture perspective on where agent software may be heading.