Dev.to
5/9/2026

I Built an AI That Argues With Itself About Anything
Short summary
AgentMesh forces AI models to surface genuine disagreement by running three Gemma 4 agents with fixed opposing stances—skeptic, advocate, pragmatist—each searching for evidence and debating opinionated questions. Unlike single-shot LLMs that average perspectives, each agent commits to its position and a synthesizer shows where they align and diverge. Runs locally in your browser (WebGPU) or cloud (BYOK Google AI key), MIT licensed, no backend.
- •Three specialized agents (skeptic, advocate, pragmatist) with fixed stances debate opinionated questions in parallel
- •Each agent searches Wikipedia and Hacker News for evidence supporting its position
- •Synthesizer reconciles views, exposing genuine disagreement rather than single-model consensus
Generated with AI, which can make mistakes.
Is this a good recommendation for you?


