LS-Labs

Glossary

AI Agent vs Chatbot vs Workflow: What Changes in Practice?

An AI agent chooses steps and tools toward a goal. Compare it with a chatbot and a fixed workflow through a product-comparison task and completion criteria.

What is an AI agent?

An AI agent is a system that uses a model to choose steps and tools toward a goal, then adjusts its approach in response to the results.

Definitions vary. Anthropic distinguishes two architectures: an agent lets the model direct its process and tool use, while a workflow follows predefined code paths. A product can combine both behind a chat interface.

The same task through three approaches

Illustrative task: compare products from three shops and return prices with source links.

What determines the next step?
ApproachHow the task proceedsWhat needs checking
ChatbotResponds in a conversational interface; may work from information supplied by the userWhether it actually retrieved current product details or merely described them
Fixed workflowRuns coded steps such as fetch three supplied URLs, extract fields, then compareHow it handles a missing page or a field outside the expected format
AgentChooses searches and page visits, notices missing details and decides what to try nextWhether tool use, permissions and evidence meet the agreed completion conditions

These are behavior examples, not mutually exclusive product labels. A chatbot can expose an agent, and an agent can call a fixed workflow.

What an agent needs to act

An agent typically needs a goal, a model that chooses what to do, available tools and a loop that uses each result to guide the next step. Some systems also retain memory. A tool can provide search, browsing, file operations or an API connection, subject to its actual permissions.

MCP is one way a compatible application can connect to tools. A search API can provide candidate sources. Neither alone establishes that an entire task will finish correctly.

Define completion before running the task

For the product comparison, require identifiable products, sources, capture dates and explicit missing fields. If a price cannot be confirmed, the result should say so or ask for human review. One successful tool call is not completion of a multi-step comparison.

The word “agent” does not establish checkout or payment capability. Those operations depend on supported tools, platform access and authorization. This definition makes no claim that a particular shopping service is currently available.

What can a website make clearer?

Keep product names, prices, stock information and specifications clear and consistent. Review crawler rules and CDN protections against the access you actually intend. Important information should be readable without relying only on an image.

The agentic-commerce readiness guide develops website checks. Clear information is useful groundwork; it is not a promise that an external agent will select or buy a product.

Sources and scope

Source-language entry records a source check on 2026-10-14. Anthropic’s architectural distinction was reopened on 2026-10-26. Its article is dated 2024-12-19; this page does not adopt it as a current inventory of every agent product.

← Back to glossary