# Agent Workflows

WebTest AI is designed for agentic development without turning runtime truth into an open-ended prompt.

Recommended pattern:

1. Let humans or agents draft Markdown specs.
2. Run `webtest-ai run` for deterministic browser execution.
3. Use reports, screenshots, traces, and CI summaries as evidence.
4. Use `webtest-ai report export` for project-specific automation.
5. Use `webtest-ai mcp` when an agent needs local access to reports, specs, artifacts, or explicit browser debug tools.

Model-backed commands such as `discover` and recorder cleanup require explicit model configuration. They should produce proposals for review, not uncontrolled rewrites.