AI agents sorted by 5 axes, from Agentforce to Dify

This framing reflects a practical problem: vendors describe AI agents in different ways, so buyers struggle to compare them. By separating the discussion into five axes, it shows that many apparent disagreements come from talking about different dimensions at once — one vendor may describe a product by mechanism, another by delivery form. The piece links academic classification to commercial services. It notes that many services built on generative AI behave close to goal-based agents, and that multi-agent setups combine several mechanism types. For example, a smart factory might use reflex agents for emergency stops and goal-based agents for production targets. Service information reflects October 2026. It closes with a four-step selection process: list tasks to automate, choose a use-case type, pick a delivery form matching in-house skills, and trial two or three candidates.