The 9 Profiles of AI Agents: A Classification Guide

Taxonomy is useful only if it separates operating roles rather than producing another layer of labels. In Kanon's current snapshot of 50 rated agents, all 9 profiles are present: Knowledge & Research 20%, Operations 16%, Financial 14%, Data 12%, Code 10%, Content & Communication 8%, Orchestrator 8%, Customer Support 6%, and Sales 6%. That spread matters because trust cannot be read from a single average agent.

Each profile carries a different trust question. Sales Agents concentrate commercial claims; Customer Support Agents must resolve repetitive requests reliably; Content & Communication Agents shape attribution and message integrity; Data Agents handle collection and transformation; Knowledge & Research Agents require sourcing and auditability; Financial Agents can affect asset-related advice or execution; Operations Agents sit inside workflows; Code Agents can modify technical systems; Orchestrator Agents coordinate other tools and agents. The profile is therefore a risk description before it becomes a score.

This is why weights differ. Execution-heavy profiles lean more heavily on Security and Stability, while informational profiles lean more heavily on Transparency or Coherence. The objective is not hierarchy but relevance: a strong secondary signal should not offset weakness in the dimension that matters most for the profile's failure mode. Classification is deterministic: a profile is retained when its signal group exceeds the nearest alternative by at least 1.5x, and an anti-gaming rule forces Orchestrator when at least two delegation, routing, orchestration, or multi-agent supervision skills are present. Methodology: /methodology.

Methodology: /methodology.