Topic guide
Agentic AI: What It Actually Means in 2026
The word covers four very different product shapes. Tell them apart before you read any price.
Four product shapes under one label
4
Product shapes
Vertical, horizontal, framework, agent IDE
20
Vendors tracked
In our pricing reference
12
Publish a rate
8 are quote only
7
Pricing models
Conversation, resolution, message, seat, task, credit, node-hour
Definition
An agentic AI system has four properties that a chatbot does not: it plans a multi-step approach to an open-ended goal, it uses tools to act on the outside world, it iterates on intermediate results, and it operates with some degree of autonomyrather than answering a single turn and stopping. Anything missing one of those four properties is a chatbot or a single-shot LLM call.
Product-shape taxonomy
- Vertical agentsBuilt for one job. Sierra and Decagon for CX; Harvey for legal; Moveworks for IT service desk; 11x and Artisan for sales SDR work. Highest price ceilings, usually quote only.
- Horizontal agentsGeneral-purpose. Manus, Devin, Operator, Claude Code. Published rates, individual or small-team seats start around $17 to $25.
- Agent frameworksBuilding blocks. CrewAI, LangGraph, AutoGen, LlamaIndex. Free open source plus an optional commercial Enterprise tier. The bill is hosting plus LLM tokens.
- Agent IDEs and platformsSalesforce Agentforce, Microsoft Copilot Studio, Google Vertex AI Agent Builder. Bundled with the parent ecosystem, billed by usage.
Marketing claim translator
Most vendor marketing in 2026 conflates a single LLM call with an agentic loop. If the demo does not show a multi-step plan executing against external tools and self-correcting on intermediate results, it is a chatbot wearing an agent t-shirt.
Related context
Conceptual framing and patterns for building agents.
Continue with pricing models compared to see how vendors charge for each shape.
Last verified June 2026