What problem does it solve?
Agent-native architecture solves the mismatch between user-facing features and agent capabilities by ensuring agents can achieve the same outcomes users can through the UI. It eliminates orphan UI actions, brittle workflow-shaped tools, and context starvation so agents can reliably pursue outcomes with judgment.
Core Features & Use Cases
- Action Parity: Ensure every UI action has an agent-equivalent tool so agents never say "I can't" for something users can do.
- Primitive Tooling & Composability: Prefer atomic primitives (read_file, write_file, list_files, bash, complete_task) so agents compose new features via prompts.
- Execution Patterns: Explicit completion signaling, partial-completion checkpoints, context injection, and model-tier selection for robust agent loops.
- Shared Workspace & Files as Interface: Use a shared filesystem (context.md pattern, entity-scoped directories) for transparency and multi-device sync.
- Safe Self-Modification: Provide approval gates, git-based workflows, build verification, and rollback for any code or prompt changes agents propose.
- Use Cases: Autonomous refactors, automated research pipelines, "organize my files" agents, mobile background checkpoint/resume, and emergent feature discovery via prompt-driven composition.
Quick Start
Ask the agent to "Organize my Research folder by topic and create an index.md summarizing each topic" to see atomic tools, context injection, and completion signaling in action.
