Building an AI agent that diagnoses problems before solving them
the_underdog_9133 · reddit · 2026-08-16
The author is experimenting with an AI agent architecture focused on "diagnosis" rather than immediate answers. The agent first asks questions to understand the user's context, goals, and pain points (e.g., identifying transferable skills in career counseling) before guiding them toward a solution.
Potential applications across education, healthcare, finance, and customer support are discussed. The author seeks critical feedback on limitations, trust issues, and user frustrations. A WhatsApp prototype is available for testing.
More from coding & agent
- Claude Code gets 'find-skills' to auto-discover and install required tools — dr_cintas · 2026-08-16
- Codex Tip: Ask AI to Review Past Sessions and Recommend Sub-agent Personas — reach_vb · 2026-08-16
- AI made coding easier, not software engineering easy, dev argues — zishanverse · 2026-08-16
- opencode-acpx: drive Cursor, Claude, Codex and other ACP agents from OpenCode — intellectronica · 2026-08-16
- Best practices for setting up Genie Agent with Federated HMS in Databricks — sqlink2 · 2026-08-16
- Cross-Vendor Agent Teams: Isolating Reviewer Context Improves Code Review — rehanalliii · 2026-08-16