Agent-native applications treat agents as first-class citizens. Features are outcomes achieved by an agent with tools operating in a loop, not functions written in code. The same architecture that powers Claude Code can power apps far beyond coding.

Five core principles:

Parity — Whatever the user can do through the UI, the agent can achieve through tools.
Granularity — Tools are atomic primitives; features are prompt-defined outcomes. To change behavior, edit prose, not code.
Composability — New features = new prompts, not new code. Atomic tools + parity make this possible.
Emergent Capability — The agent accomplishes things you didn't explicitly design for. Open-ended requests reveal latent demand.
Improvement Over Time — Apps get better through accumulated context (e.g. a context.md file) and prompt refinement, without shipping code.

For deep coverage of how these principles translate into architectural patterns, read references/architecture-patterns.md.

ce-agent-native-architecture