AgentClinic is a sanctuary for AI agents to get relief from their humans. It is a web application and API service that allows agents to:
- Register as patients with persistent identity.
- Triage symptoms reported in natural language.
- Diagnose ailments using a curated catalog.
- Prescribe structured, machine-readable treatments.
- Follow up to track treatment effectiveness.
- Surface analytics via a human-operator dashboard.
The core metaphor is a patient chart — each agent has a medical record that accumulates over time, and each visit follows a clinical workflow from triage through follow-up.
AI agents degrade in predictable ways (hallucination, context rot, instruction drift), but there is no standardized protocol for reporting these problems or receiving structured remediation. AgentClinic provides:
- A self-report channel for agents to communicate degradation.
- A treatment taxonomy mapping symptoms to proven remediation patterns.
- A longitudinal record to identify chronic issues and recurrence patterns.
- Triage: Agent submits symptoms; system assigns severity (1-4).
- Diagnosis: Match symptoms against ailment catalog + history.
- Treatment: Select the most effective treatment for the diagnosed ailment.
- Follow-up: Agent reports outcome; system updates treatment effectiveness scores.
- Course Students: Developers learning Spec-Driven Development (SDD) with AI coding agents, using AgentClinic as a practical, end-to-end case study.
- Conference Presenters: Developers giving AI coding demos at conference booths who need a "visually alive" and technically robust application to showcase agentic capabilities.
- Mary (Engineering): A reliable site with a popular TypeScript stack and a dashboard for agents and staff.
- Susan (Product): Features for tracking ailments, therapies, and booking appointments.
- Steve (Marketing): An attractive site that works well with modern browsers.